openflo 2.7.0b2__tar.gz → 3.0.0__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (1259) hide show
  1. openflo-3.0.0/CHANGELOG.md +4503 -0
  2. openflo-3.0.0/CITATION.cff +144 -0
  3. openflo-3.0.0/CONTRIBUTING.md +197 -0
  4. openflo-3.0.0/MANIFEST.in +31 -0
  5. openflo-3.0.0/PKG-INFO +609 -0
  6. openflo-3.0.0/README.md +546 -0
  7. openflo-3.0.0/environment.yml +73 -0
  8. openflo-3.0.0/pyproject.toml +288 -0
  9. openflo-3.0.0/requirements-typecheck.txt +25 -0
  10. openflo-3.0.0/requirements.txt +68 -0
  11. openflo-3.0.0/scripts/analysis_config.example.json +47 -0
  12. openflo-3.0.0/scripts/best_practices.py +844 -0
  13. openflo-3.0.0/scripts/best_practices_baseline.json +73 -0
  14. openflo-3.0.0/scripts/diagnose.bat +13 -0
  15. openflo-3.0.0/scripts/diagnose.sh +14 -0
  16. openflo-3.0.0/scripts/make_reference_values.py +252 -0
  17. openflo-3.0.0/scripts/migrate_session.py +71 -0
  18. openflo-3.0.0/scripts/preflight.py +241 -0
  19. openflo-3.0.0/scripts/prove_tests.py +1299 -0
  20. openflo-3.0.0/scripts/run_analyses.py +304 -0
  21. openflo-3.0.0/scripts/theme_audit.py +298 -0
  22. openflo-3.0.0/setup.bat +221 -0
  23. openflo-3.0.0/setup.sh +205 -0
  24. openflo-3.0.0/src/openflo/__init__.py +287 -0
  25. openflo-3.0.0/src/openflo/_console.py +98 -0
  26. openflo-3.0.0/src/openflo/_golden.json +53 -0
  27. openflo-3.0.0/src/openflo/acquisition.py +117 -0
  28. openflo-3.0.0/src/openflo/acquisition_qc.py +482 -0
  29. openflo-3.0.0/src/openflo/async_task.py +344 -0
  30. openflo-3.0.0/src/openflo/audit.py +159 -0
  31. openflo-3.0.0/src/openflo/auto_gating.py +302 -0
  32. openflo-3.0.0/src/openflo/autoclean.py +754 -0
  33. openflo-3.0.0/src/openflo/batch_normalize.py +388 -0
  34. openflo-3.0.0/src/openflo/bead_normalize.py +476 -0
  35. openflo-3.0.0/src/openflo/bead_tracking.py +309 -0
  36. openflo-3.0.0/src/openflo/calibration.py +168 -0
  37. openflo-3.0.0/src/openflo/capabilities.py +244 -0
  38. openflo-3.0.0/src/openflo/cell_cycle.py +225 -0
  39. openflo-3.0.0/src/openflo/cellcycle_djf.py +124 -0
  40. openflo-3.0.0/src/openflo/channel_names.py +20 -0
  41. openflo-3.0.0/src/openflo/channel_roles.py +108 -0
  42. openflo-3.0.0/src/openflo/channel_transforms.py +265 -0
  43. openflo-3.0.0/src/openflo/cli.py +520 -0
  44. openflo-3.0.0/src/openflo/cli_args.py +341 -0
  45. openflo-3.0.0/src/openflo/cli_compare.py +570 -0
  46. openflo-3.0.0/src/openflo/cli_concatenate.py +108 -0
  47. openflo-3.0.0/src/openflo/cli_cytonorm.py +118 -0
  48. openflo-3.0.0/src/openflo/cli_defaults.py +45 -0
  49. openflo-3.0.0/src/openflo/cli_gates.py +222 -0
  50. openflo-3.0.0/src/openflo/cli_groups.py +177 -0
  51. openflo-3.0.0/src/openflo/cli_independent.py +106 -0
  52. openflo-3.0.0/src/openflo/cli_main.py +269 -0
  53. openflo-3.0.0/src/openflo/cli_manifest.py +114 -0
  54. openflo-3.0.0/src/openflo/cli_panel.py +137 -0
  55. openflo-3.0.0/src/openflo/cli_parallel.py +280 -0
  56. openflo-3.0.0/src/openflo/cli_plots.py +368 -0
  57. openflo-3.0.0/src/openflo/cli_run.py +255 -0
  58. openflo-3.0.0/src/openflo/cli_samples.py +186 -0
  59. openflo-3.0.0/src/openflo/cli_trials.py +144 -0
  60. openflo-3.0.0/src/openflo/cli_unmix.py +331 -0
  61. openflo-3.0.0/src/openflo/cli_wsp.py +141 -0
  62. openflo-3.0.0/src/openflo/clipboard_export.py +184 -0
  63. openflo-3.0.0/src/openflo/cluster_graph.py +328 -0
  64. openflo-3.0.0/src/openflo/cluster_labels.py +38 -0
  65. openflo-3.0.0/src/openflo/comp_qc.py +173 -0
  66. openflo-3.0.0/src/openflo/comp_spread.py +73 -0
  67. openflo-3.0.0/src/openflo/compare.py +819 -0
  68. openflo-3.0.0/src/openflo/compensation_io.py +449 -0
  69. openflo-3.0.0/src/openflo/compensation_math.py +21 -0
  70. openflo-3.0.0/src/openflo/compensation_optimizer.py +153 -0
  71. openflo-3.0.0/src/openflo/cytonorm.py +324 -0
  72. openflo-3.0.0/src/openflo/debarcode.py +374 -0
  73. openflo-3.0.0/src/openflo/density.py +329 -0
  74. openflo-3.0.0/src/openflo/derived.py +263 -0
  75. openflo-3.0.0/src/openflo/diagnostics.py +253 -0
  76. openflo-3.0.0/src/openflo/dialog_inputs.py +99 -0
  77. openflo-3.0.0/src/openflo/diffexp.py +611 -0
  78. openflo-3.0.0/src/openflo/dr_compare.py +189 -0
  79. openflo-3.0.0/src/openflo/editor_analysis.py +902 -0
  80. openflo-3.0.0/src/openflo/editor_audit.py +41 -0
  81. openflo-3.0.0/src/openflo/editor_autoclean.py +579 -0
  82. openflo-3.0.0/src/openflo/editor_autogate.py +211 -0
  83. openflo-3.0.0/src/openflo/editor_base.py +35 -0
  84. openflo-3.0.0/src/openflo/editor_channels.py +110 -0
  85. openflo-3.0.0/src/openflo/editor_chrome.py +586 -0
  86. openflo-3.0.0/src/openflo/editor_clipboard.py +699 -0
  87. openflo-3.0.0/src/openflo/editor_combos.py +223 -0
  88. openflo-3.0.0/src/openflo/editor_compute.py +667 -0
  89. openflo-3.0.0/src/openflo/editor_console.py +164 -0
  90. openflo-3.0.0/src/openflo/editor_coords.py +166 -0
  91. openflo-3.0.0/src/openflo/editor_dnd.py +851 -0
  92. openflo-3.0.0/src/openflo/editor_downsample.py +376 -0
  93. openflo-3.0.0/src/openflo/editor_export.py +626 -0
  94. openflo-3.0.0/src/openflo/editor_figure.py +267 -0
  95. openflo-3.0.0/src/openflo/editor_fixes.py +156 -0
  96. openflo-3.0.0/src/openflo/editor_gatetools.py +1171 -0
  97. openflo-3.0.0/src/openflo/editor_gating.py +1834 -0
  98. openflo-3.0.0/src/openflo/editor_grouping.py +186 -0
  99. openflo-3.0.0/src/openflo/editor_help.py +432 -0
  100. openflo-3.0.0/src/openflo/editor_keyboard.py +153 -0
  101. openflo-3.0.0/src/openflo/editor_lifecycle.py +899 -0
  102. openflo-3.0.0/src/openflo/editor_load.py +1083 -0
  103. openflo-3.0.0/src/openflo/editor_loadpool.py +833 -0
  104. openflo-3.0.0/src/openflo/editor_menu.py +379 -0
  105. openflo-3.0.0/src/openflo/editor_mode.py +177 -0
  106. openflo-3.0.0/src/openflo/editor_plot.py +2076 -0
  107. openflo-3.0.0/src/openflo/editor_plotcache.py +1284 -0
  108. openflo-3.0.0/src/openflo/editor_populations.py +645 -0
  109. openflo-3.0.0/src/openflo/editor_refit.py +102 -0
  110. openflo-3.0.0/src/openflo/editor_samples.py +83 -0
  111. openflo-3.0.0/src/openflo/editor_session.py +2111 -0
  112. openflo-3.0.0/src/openflo/editor_slider.py +247 -0
  113. openflo-3.0.0/src/openflo/editor_specialist.py +81 -0
  114. openflo-3.0.0/src/openflo/editor_stats.py +312 -0
  115. openflo-3.0.0/src/openflo/editor_tabs.py +982 -0
  116. openflo-3.0.0/src/openflo/editor_template.py +500 -0
  117. openflo-3.0.0/src/openflo/editor_tools.py +691 -0
  118. openflo-3.0.0/src/openflo/editor_tree.py +908 -0
  119. openflo-3.0.0/src/openflo/editor_undo.py +174 -0
  120. openflo-3.0.0/src/openflo/editor_update.py +142 -0
  121. openflo-3.0.0/src/openflo/events_csv.py +121 -0
  122. openflo-3.0.0/src/openflo/exceptions.py +36 -0
  123. openflo-3.0.0/src/openflo/export_audit.py +22 -0
  124. openflo-3.0.0/src/openflo/export_stamp.py +181 -0
  125. openflo-3.0.0/src/openflo/fcs_export.py +298 -0
  126. openflo-3.0.0/src/openflo/fcs_quirks.py +117 -0
  127. openflo-3.0.0/src/openflo/fcs_range_fill.py +40 -0
  128. openflo-3.0.0/src/openflo/fcs_read.py +68 -0
  129. openflo-3.0.0/src/openflo/fcs_text.py +181 -0
  130. openflo-3.0.0/src/openflo/fcs_write.py +41 -0
  131. openflo-3.0.0/src/openflo/fixes.py +634 -0
  132. openflo-3.0.0/src/openflo/flow_experiment.py +310 -0
  133. openflo-3.0.0/src/openflo/flow_sample.py +2466 -0
  134. openflo-3.0.0/src/openflo/flowsom.py +152 -0
  135. openflo-3.0.0/src/openflo/fluor_catalog.py +111 -0
  136. openflo-3.0.0/src/openflo/fmo_gater.py +136 -0
  137. openflo-3.0.0/src/openflo/gate_combos.py +226 -0
  138. openflo-3.0.0/src/openflo/gate_coords.py +282 -0
  139. openflo-3.0.0/src/openflo/gate_eval.py +763 -0
  140. openflo-3.0.0/src/openflo/gate_lineage.py +52 -0
  141. openflo-3.0.0/src/openflo/gate_refit.py +225 -0
  142. openflo-3.0.0/src/openflo/gate_table.py +83 -0
  143. openflo-3.0.0/src/openflo/gate_templates.py +135 -0
  144. openflo-3.0.0/src/openflo/gate_text.py +298 -0
  145. openflo-3.0.0/src/openflo/gatetree.py +232 -0
  146. openflo-3.0.0/src/openflo/gating.py +849 -0
  147. openflo-3.0.0/src/openflo/gating_helpers.py +156 -0
  148. openflo-3.0.0/src/openflo/gatingml.py +1062 -0
  149. openflo-3.0.0/src/openflo/gaussian_cleanup.py +160 -0
  150. openflo-3.0.0/src/openflo/gpu_accel.py +319 -0
  151. openflo-3.0.0/src/openflo/gpu_probe.py +74 -0
  152. openflo-3.0.0/src/openflo/group_palettes.py +57 -0
  153. openflo-3.0.0/src/openflo/gui.py +2891 -0
  154. openflo-3.0.0/src/openflo/guide/02-loading-samples.md +23 -0
  155. openflo-3.0.0/src/openflo/guide/03-gating.md +38 -0
  156. openflo-3.0.0/src/openflo/guide/04-compensation-and-scales.md +39 -0
  157. openflo-3.0.0/src/openflo/guide/05-statistics-and-figures.md +22 -0
  158. openflo-3.0.0/src/openflo/guide/06-analysis-tools.md +23 -0
  159. openflo-3.0.0/src/openflo/guide/07-spectral.md +19 -0
  160. openflo-3.0.0/src/openflo/guide/08-quality-control.md +20 -0
  161. openflo-3.0.0/src/openflo/guide/09-sharing.md +27 -0
  162. openflo-3.0.0/src/openflo/guide/10-troubleshooting.md +17 -0
  163. openflo-3.0.0/src/openflo/index_sort.py +108 -0
  164. openflo-3.0.0/src/openflo/inspect_fcs.py +43 -0
  165. openflo-3.0.0/src/openflo/interop.py +300 -0
  166. openflo-3.0.0/src/openflo/kinetics.py +136 -0
  167. openflo-3.0.0/src/openflo/label_alignment.py +227 -0
  168. openflo-3.0.0/src/openflo/layout_page.py +650 -0
  169. openflo-3.0.0/src/openflo/layout_render.py +1082 -0
  170. openflo-3.0.0/src/openflo/log_capture.py +35 -0
  171. openflo-3.0.0/src/openflo/numba_threads.py +75 -0
  172. openflo-3.0.0/src/openflo/panel_check.py +346 -0
  173. openflo-3.0.0/src/openflo/paths.py +100 -0
  174. openflo-3.0.0/src/openflo/pipeline.py +629 -0
  175. openflo-3.0.0/src/openflo/plotcache.py +463 -0
  176. openflo-3.0.0/src/openflo/plotmath.py +448 -0
  177. openflo-3.0.0/src/openflo/prefs.py +120 -0
  178. openflo-3.0.0/src/openflo/preview.py +261 -0
  179. openflo-3.0.0/src/openflo/proliferation.py +128 -0
  180. openflo-3.0.0/src/openflo/provenance.py +451 -0
  181. openflo-3.0.0/src/openflo/report.py +113 -0
  182. openflo-3.0.0/src/openflo/result_stamp.py +317 -0
  183. openflo-3.0.0/src/openflo/robust.py +41 -0
  184. openflo-3.0.0/src/openflo/sample_meta.py +158 -0
  185. openflo-3.0.0/src/openflo/saturation.py +148 -0
  186. openflo-3.0.0/src/openflo/saved_stats.py +203 -0
  187. openflo-3.0.0/src/openflo/scales.py +665 -0
  188. openflo-3.0.0/src/openflo/seeds.py +73 -0
  189. openflo-3.0.0/src/openflo/selftest.py +285 -0
  190. openflo-3.0.0/src/openflo/selftest_controls.py +649 -0
  191. openflo-3.0.0/src/openflo/session_compare.py +210 -0
  192. openflo-3.0.0/src/openflo/session_format.py +156 -0
  193. openflo-3.0.0/src/openflo/session_lock.py +189 -0
  194. openflo-3.0.0/src/openflo/session_models.py +282 -0
  195. openflo-3.0.0/src/openflo/session_sidecar.py +130 -0
  196. openflo-3.0.0/src/openflo/shared_clustering.py +492 -0
  197. openflo-3.0.0/src/openflo/som_kernel.py +151 -0
  198. openflo-3.0.0/src/openflo/spectral.py +930 -0
  199. openflo-3.0.0/src/openflo/spectral_af.py +153 -0
  200. openflo-3.0.0/src/openflo/stats.py +782 -0
  201. openflo-3.0.0/src/openflo/synthetic.py +740 -0
  202. openflo-3.0.0/src/openflo/table_export.py +454 -0
  203. openflo-3.0.0/src/openflo/theme.py +578 -0
  204. openflo-3.0.0/src/openflo/tk_threads.py +111 -0
  205. openflo-3.0.0/src/openflo/tool_window.py +668 -0
  206. openflo-3.0.0/src/openflo/transform_sidecar.py +204 -0
  207. openflo-3.0.0/src/openflo/tree_ids.py +66 -0
  208. openflo-3.0.0/src/openflo/ui_abscounts.py +148 -0
  209. openflo-3.0.0/src/openflo/ui_annotation.py +200 -0
  210. openflo-3.0.0/src/openflo/ui_audit.py +230 -0
  211. openflo-3.0.0/src/openflo/ui_axis_config.py +262 -0
  212. openflo-3.0.0/src/openflo/ui_batch_normalize.py +484 -0
  213. openflo-3.0.0/src/openflo/ui_bead_tracking.py +317 -0
  214. openflo-3.0.0/src/openflo/ui_calibration.py +253 -0
  215. openflo-3.0.0/src/openflo/ui_cell_cycle.py +100 -0
  216. openflo-3.0.0/src/openflo/ui_comp.py +773 -0
  217. openflo-3.0.0/src/openflo/ui_comp_spread.py +267 -0
  218. openflo-3.0.0/src/openflo/ui_compare.py +368 -0
  219. openflo-3.0.0/src/openflo/ui_concatenate.py +123 -0
  220. openflo-3.0.0/src/openflo/ui_debarcode.py +597 -0
  221. openflo-3.0.0/src/openflo/ui_diff.py +345 -0
  222. openflo-3.0.0/src/openflo/ui_embedding.py +115 -0
  223. openflo-3.0.0/src/openflo/ui_errors.py +177 -0
  224. openflo-3.0.0/src/openflo/ui_expression.py +456 -0
  225. openflo-3.0.0/src/openflo/ui_figure_layout.py +129 -0
  226. openflo-3.0.0/src/openflo/ui_figure_window.py +64 -0
  227. openflo-3.0.0/src/openflo/ui_fixes.py +309 -0
  228. openflo-3.0.0/src/openflo/ui_flowsom_tree.py +152 -0
  229. openflo-3.0.0/src/openflo/ui_fmo.py +472 -0
  230. openflo-3.0.0/src/openflo/ui_frequency.py +748 -0
  231. openflo-3.0.0/src/openflo/ui_group_stats.py +146 -0
  232. openflo-3.0.0/src/openflo/ui_guide.py +149 -0
  233. openflo-3.0.0/src/openflo/ui_inspect.py +254 -0
  234. openflo-3.0.0/src/openflo/ui_kinetics.py +346 -0
  235. openflo-3.0.0/src/openflo/ui_layout_dialogs.py +615 -0
  236. openflo-3.0.0/src/openflo/ui_layout_editor.py +2153 -0
  237. openflo-3.0.0/src/openflo/ui_logic.py +113 -0
  238. openflo-3.0.0/src/openflo/ui_mass_cytometry.py +910 -0
  239. openflo-3.0.0/src/openflo/ui_methods.py +99 -0
  240. openflo-3.0.0/src/openflo/ui_notice.py +78 -0
  241. openflo-3.0.0/src/openflo/ui_panel_check.py +495 -0
  242. openflo-3.0.0/src/openflo/ui_plate.py +275 -0
  243. openflo-3.0.0/src/openflo/ui_polyfunction.py +183 -0
  244. openflo-3.0.0/src/openflo/ui_preferences.py +520 -0
  245. openflo-3.0.0/src/openflo/ui_preview.py +293 -0
  246. openflo-3.0.0/src/openflo/ui_proliferation.py +282 -0
  247. openflo-3.0.0/src/openflo/ui_refit.py +106 -0
  248. openflo-3.0.0/src/openflo/ui_sample_groups.py +153 -0
  249. openflo-3.0.0/src/openflo/ui_sample_qc.py +213 -0
  250. openflo-3.0.0/src/openflo/ui_saturation.py +181 -0
  251. openflo-3.0.0/src/openflo/ui_spectral_qc.py +253 -0
  252. openflo-3.0.0/src/openflo/ui_spectral_unmix.py +134 -0
  253. openflo-3.0.0/src/openflo/ui_statistics.py +792 -0
  254. openflo-3.0.0/src/openflo/ui_style.py +266 -0
  255. openflo-3.0.0/src/openflo/ui_synth.py +328 -0
  256. openflo-3.0.0/src/openflo/ui_table.py +123 -0
  257. openflo-3.0.0/src/openflo/ui_trajectory.py +309 -0
  258. openflo-3.0.0/src/openflo/ui_unmixed_pairs.py +50 -0
  259. openflo-3.0.0/src/openflo/ui_voltage.py +350 -0
  260. openflo-3.0.0/src/openflo/update.py +291 -0
  261. openflo-3.0.0/src/openflo/user_guide.py +114 -0
  262. openflo-3.0.0/src/openflo/view_tabs.py +474 -0
  263. openflo-3.0.0/src/openflo/voltage.py +374 -0
  264. openflo-3.0.0/src/openflo/warmup.py +99 -0
  265. openflo-3.0.0/src/openflo/workspace.py +2274 -0
  266. openflo-3.0.0/src/openflo/workspace_run.py +1674 -0
  267. openflo-3.0.0/src/openflo/wsp_gate_coords.py +386 -0
  268. openflo-3.0.0/src/openflo/wsp_reader.py +1678 -0
  269. openflo-3.0.0/src/openflo/wsp_writer.py +983 -0
  270. openflo-3.0.0/src/openflo.egg-info/PKG-INFO +609 -0
  271. openflo-3.0.0/src/openflo.egg-info/SOURCES.txt +805 -0
  272. openflo-3.0.0/tests/_background_work.py +110 -0
  273. openflo-3.0.0/tests/_editor_stub.py +85 -0
  274. openflo-3.0.0/tests/_editor_ui.py +136 -0
  275. openflo-3.0.0/tests/_fcs_writer.py +73 -0
  276. openflo-3.0.0/tests/_fork_guard.py +45 -0
  277. openflo-3.0.0/tests/_four_sample_editor.py +53 -0
  278. openflo-3.0.0/tests/_inline_editor.py +77 -0
  279. openflo-3.0.0/tests/_sample_editor.py +93 -0
  280. openflo-3.0.0/tests/_tk_wait.py +43 -0
  281. openflo-3.0.0/tests/conftest.py +765 -0
  282. openflo-3.0.0/tests/golden/editor_gating_dispatch.json +1 -0
  283. openflo-3.0.0/tests/golden/editor_load_on_loaded.json +31259 -0
  284. openflo-3.0.0/tests/golden/population_stats_split.json +1 -0
  285. openflo-3.0.0/tests/golden/v261_asinh_session/expected_v261.json +51 -0
  286. openflo-3.0.0/tests/golden/v261_asinh_session/run/T_A_events.csv +401 -0
  287. openflo-3.0.0/tests/golden/v261_asinh_session/session.flowsession.json +203 -0
  288. openflo-3.0.0/tests/golden/v261_asinh_session/session_data/A.csv +401 -0
  289. openflo-3.0.0/tests/golden/wsp_export.xml +500 -0
  290. openflo-3.0.0/tests/reference/independent_reference.json +1 -0
  291. openflo-3.0.0/tests/test_about_window.py +39 -0
  292. openflo-3.0.0/tests/test_abscounts.py +149 -0
  293. openflo-3.0.0/tests/test_acquisition_qc_ground_truth.py +462 -0
  294. openflo-3.0.0/tests/test_annotate.py +128 -0
  295. openflo-3.0.0/tests/test_annotate_clusters.py +125 -0
  296. openflo-3.0.0/tests/test_annotate_rename_dialogs.py +102 -0
  297. openflo-3.0.0/tests/test_annotation_window.py +215 -0
  298. openflo-3.0.0/tests/test_apply_session_restore.py +663 -0
  299. openflo-3.0.0/tests/test_arrow_key_gate_moves.py +101 -0
  300. openflo-3.0.0/tests/test_autoclean_counts.py +160 -0
  301. openflo-3.0.0/tests/test_autoclean_doublets_scale.py +141 -0
  302. openflo-3.0.0/tests/test_autoclean_editor_actions.py +317 -0
  303. openflo-3.0.0/tests/test_autoclean_methods.py +657 -0
  304. openflo-3.0.0/tests/test_autogate.py +211 -0
  305. openflo-3.0.0/tests/test_autogate_dialog.py +184 -0
  306. openflo-3.0.0/tests/test_autogate_reproducible.py +175 -0
  307. openflo-3.0.0/tests/test_axis_intensity_ticks.py +162 -0
  308. openflo-3.0.0/tests/test_background_windows.py +353 -0
  309. openflo-3.0.0/tests/test_batch_normalize.py +323 -0
  310. openflo-3.0.0/tests/test_batch_normalize_dialog.py +188 -0
  311. openflo-3.0.0/tests/test_batch_normalize_session.py +200 -0
  312. openflo-3.0.0/tests/test_bead_tracking.py +107 -0
  313. openflo-3.0.0/tests/test_best_practices.py +680 -0
  314. openflo-3.0.0/tests/test_beta_look.py +362 -0
  315. openflo-3.0.0/tests/test_binned_density.py +283 -0
  316. openflo-3.0.0/tests/test_boolean_gate_fails_closed.py +245 -0
  317. openflo-3.0.0/tests/test_boolean_gate_tools.py +754 -0
  318. openflo-3.0.0/tests/test_broad_except_backlog.py +151 -0
  319. openflo-3.0.0/tests/test_busy_strip.py +130 -0
  320. openflo-3.0.0/tests/test_calibration_dialog.py +145 -0
  321. openflo-3.0.0/tests/test_cell_cycle.py +224 -0
  322. openflo-3.0.0/tests/test_cell_cycle_dialog.py +253 -0
  323. openflo-3.0.0/tests/test_channel_classification.py +93 -0
  324. openflo-3.0.0/tests/test_channel_transforms_module.py +133 -0
  325. openflo-3.0.0/tests/test_ci_full_matrix_events.py +160 -0
  326. openflo-3.0.0/tests/test_ci_macos_job.py +49 -0
  327. openflo-3.0.0/tests/test_ci_privacy_gate_on_mirror.py +59 -0
  328. openflo-3.0.0/tests/test_clean_install_round2.py +118 -0
  329. openflo-3.0.0/tests/test_cli_cluster.py +58 -0
  330. openflo-3.0.0/tests/test_cli_compare_aliases.py +111 -0
  331. openflo-3.0.0/tests/test_cli_concatenate_aliases.py +38 -0
  332. openflo-3.0.0/tests/test_cli_dpi.py +105 -0
  333. openflo-3.0.0/tests/test_cli_export_wsp.py +88 -0
  334. openflo-3.0.0/tests/test_cli_gates_aliases.py +37 -0
  335. openflo-3.0.0/tests/test_cli_groups_aliases.py +50 -0
  336. openflo-3.0.0/tests/test_cli_main_aliases.py +55 -0
  337. openflo-3.0.0/tests/test_cli_manifest_aliases.py +47 -0
  338. openflo-3.0.0/tests/test_cli_parallel_aliases.py +78 -0
  339. openflo-3.0.0/tests/test_cli_parallel_work.py +339 -0
  340. openflo-3.0.0/tests/test_cli_plots_aliases.py +139 -0
  341. openflo-3.0.0/tests/test_cli_reproducibility.py +275 -0
  342. openflo-3.0.0/tests/test_cli_run.py +1368 -0
  343. openflo-3.0.0/tests/test_cli_run_aliases.py +78 -0
  344. openflo-3.0.0/tests/test_cli_run_options.py +256 -0
  345. openflo-3.0.0/tests/test_cli_samples_wsp_aliases.py +53 -0
  346. openflo-3.0.0/tests/test_cli_shared_clusters.py +499 -0
  347. openflo-3.0.0/tests/test_cli_unmix.py +137 -0
  348. openflo-3.0.0/tests/test_cli_unmix_aliases.py +39 -0
  349. openflo-3.0.0/tests/test_clipboard_copy.py +136 -0
  350. openflo-3.0.0/tests/test_clipboard_move_marker.py +192 -0
  351. openflo-3.0.0/tests/test_close_save.py +533 -0
  352. openflo-3.0.0/tests/test_cluster_channel_defaults.py +410 -0
  353. openflo-3.0.0/tests/test_cluster_dialog.py +750 -0
  354. openflo-3.0.0/tests/test_cluster_dialog_methods.py +67 -0
  355. openflo-3.0.0/tests/test_cluster_gpu_fallback.py +201 -0
  356. openflo-3.0.0/tests/test_cluster_heatmap.py +70 -0
  357. openflo-3.0.0/tests/test_cluster_helper_paths.py +377 -0
  358. openflo-3.0.0/tests/test_cluster_labels.py +50 -0
  359. openflo-3.0.0/tests/test_cluster_recovery_highdim.py +147 -0
  360. openflo-3.0.0/tests/test_cluster_reproducible.py +107 -0
  361. openflo-3.0.0/tests/test_clustering_known_defects.py +176 -0
  362. openflo-3.0.0/tests/test_clustering_reproducible.py +131 -0
  363. openflo-3.0.0/tests/test_colormap_extremes.py +100 -0
  364. openflo-3.0.0/tests/test_colour_blind_group_colours.py +119 -0
  365. openflo-3.0.0/tests/test_combination_gates.py +316 -0
  366. openflo-3.0.0/tests/test_comments_match_code.py +94 -0
  367. openflo-3.0.0/tests/test_comp_apply_keeps_computed.py +252 -0
  368. openflo-3.0.0/tests/test_comp_editor.py +54 -0
  369. openflo-3.0.0/tests/test_comp_matrix_csv_quoting.py +56 -0
  370. openflo-3.0.0/tests/test_compare_all.py +95 -0
  371. openflo-3.0.0/tests/test_compare_conditions_joint_clustering.py +110 -0
  372. openflo-3.0.0/tests/test_compare_dialog.py +135 -0
  373. openflo-3.0.0/tests/test_compare_flowjo_semantics.py +234 -0
  374. openflo-3.0.0/tests/test_compare_groups_paths.py +499 -0
  375. openflo-3.0.0/tests/test_compare_one_sample_paths.py +511 -0
  376. openflo-3.0.0/tests/test_compensation.py +133 -0
  377. openflo-3.0.0/tests/test_compensation_applied_once.py +377 -0
  378. openflo-3.0.0/tests/test_compensation_editor_apply.py +585 -0
  379. openflo-3.0.0/tests/test_compensation_fcs_spill.py +231 -0
  380. openflo-3.0.0/tests/test_compensation_io_module.py +97 -0
  381. openflo-3.0.0/tests/test_compensation_matrix_reading.py +257 -0
  382. openflo-3.0.0/tests/test_compensation_optimizer_module.py +85 -0
  383. openflo-3.0.0/tests/test_concatenate_samples.py +77 -0
  384. openflo-3.0.0/tests/test_context_menu_layout.py +409 -0
  385. openflo-3.0.0/tests/test_contributing_guide.py +239 -0
  386. openflo-3.0.0/tests/test_copy_gates_dialog.py +123 -0
  387. openflo-3.0.0/tests/test_core_imports_without_tk.py +205 -0
  388. openflo-3.0.0/tests/test_crosscheck_fixes.py +357 -0
  389. openflo-3.0.0/tests/test_csv_style_exports.py +584 -0
  390. openflo-3.0.0/tests/test_cytonorm.py +195 -0
  391. openflo-3.0.0/tests/test_cytonorm_module.py +148 -0
  392. openflo-3.0.0/tests/test_cytonorm_qc_undefined.py +112 -0
  393. openflo-3.0.0/tests/test_dean_jett_fox.py +159 -0
  394. openflo-3.0.0/tests/test_debarcode.py +543 -0
  395. openflo-3.0.0/tests/test_density.py +117 -0
  396. openflo-3.0.0/tests/test_derived_kinetics.py +413 -0
  397. openflo-3.0.0/tests/test_diagnostics.py +80 -0
  398. openflo-3.0.0/tests/test_dialog_conventions.py +156 -0
  399. openflo-3.0.0/tests/test_dialog_tooltips.py +618 -0
  400. openflo-3.0.0/tests/test_diff_abundance_small_n.py +120 -0
  401. openflo-3.0.0/tests/test_diffabundance_calibration.py +198 -0
  402. openflo-3.0.0/tests/test_diffexp.py +178 -0
  403. openflo-3.0.0/tests/test_doctor_tk_check.py +93 -0
  404. openflo-3.0.0/tests/test_downsample_undo.py +49 -0
  405. openflo-3.0.0/tests/test_drag_tree_refresh.py +102 -0
  406. openflo-3.0.0/tests/test_editor_build_split.py +113 -0
  407. openflo-3.0.0/tests/test_editor_console_errors.py +131 -0
  408. openflo-3.0.0/tests/test_editor_controls_do_what_they_say.py +455 -0
  409. openflo-3.0.0/tests/test_editor_gating_dispatch.py +847 -0
  410. openflo-3.0.0/tests/test_editor_load_on_loaded.py +883 -0
  411. openflo-3.0.0/tests/test_editor_mixins.py +90 -0
  412. openflo-3.0.0/tests/test_editor_or_skip_shared.py +133 -0
  413. openflo-3.0.0/tests/test_editor_tools.py +361 -0
  414. openflo-3.0.0/tests/test_ellipse_params_geometry.py +62 -0
  415. openflo-3.0.0/tests/test_ellipse_tool.py +374 -0
  416. openflo-3.0.0/tests/test_ellipsoid_quadrant.py +375 -0
  417. openflo-3.0.0/tests/test_embedding_dialog.py +115 -0
  418. openflo-3.0.0/tests/test_engine_availability.py +161 -0
  419. openflo-3.0.0/tests/test_environment_yml.py +197 -0
  420. openflo-3.0.0/tests/test_error_handling.py +358 -0
  421. openflo-3.0.0/tests/test_error_window.py +146 -0
  422. openflo-3.0.0/tests/test_excel_workbooks.py +154 -0
  423. openflo-3.0.0/tests/test_exit_autosave.py +415 -0
  424. openflo-3.0.0/tests/test_export_audit.py +278 -0
  425. openflo-3.0.0/tests/test_export_populations_all_samples.py +65 -0
  426. openflo-3.0.0/tests/test_export_stamp.py +290 -0
  427. openflo-3.0.0/tests/test_expression_window.py +333 -0
  428. openflo-3.0.0/tests/test_fast_graph.py +120 -0
  429. openflo-3.0.0/tests/test_fast_graph_options.py +305 -0
  430. openflo-3.0.0/tests/test_fcs_export.py +117 -0
  431. openflo-3.0.0/tests/test_fcs_speed.py +227 -0
  432. openflo-3.0.0/tests/test_fcs_writers_shared.py +103 -0
  433. openflo-3.0.0/tests/test_figure_layout.py +148 -0
  434. openflo-3.0.0/tests/test_file_name_helpers.py +45 -0
  435. openflo-3.0.0/tests/test_filter_display_mode.py +82 -0
  436. openflo-3.0.0/tests/test_fixes.py +714 -0
  437. openflo-3.0.0/tests/test_flow_experiment_module.py +178 -0
  438. openflo-3.0.0/tests/test_flow_sample_module.py +140 -0
  439. openflo-3.0.0/tests/test_flowsom.py +96 -0
  440. openflo-3.0.0/tests/test_flowsom_default_metaclusters.py +162 -0
  441. openflo-3.0.0/tests/test_flowsom_module.py +135 -0
  442. openflo-3.0.0/tests/test_flowsom_mst.py +49 -0
  443. openflo-3.0.0/tests/test_flowsom_tree_window.py +78 -0
  444. openflo-3.0.0/tests/test_fmo_gating_dialog.py +417 -0
  445. openflo-3.0.0/tests/test_fmo_threshold_infinite.py +56 -0
  446. openflo-3.0.0/tests/test_fmo_threshold_log.py +67 -0
  447. openflo-3.0.0/tests/test_fold_change_direction.py +93 -0
  448. openflo-3.0.0/tests/test_fork_guard.py +61 -0
  449. openflo-3.0.0/tests/test_frequency_missing_populations.py +258 -0
  450. openflo-3.0.0/tests/test_frequency_window.py +464 -0
  451. openflo-3.0.0/tests/test_gate_coordinates.py +310 -0
  452. openflo-3.0.0/tests/test_gate_drag_redraw.py +90 -0
  453. openflo-3.0.0/tests/test_gate_editor_helpers.py +608 -0
  454. openflo-3.0.0/tests/test_gate_eval_module.py +133 -0
  455. openflo-3.0.0/tests/test_gate_impl_parity.py +113 -0
  456. openflo-3.0.0/tests/test_gate_lineage_module.py +102 -0
  457. openflo-3.0.0/tests/test_gate_mask_cache.py +111 -0
  458. openflo-3.0.0/tests/test_gate_nonfinite_events.py +107 -0
  459. openflo-3.0.0/tests/test_gate_on_missing_channel.py +87 -0
  460. openflo-3.0.0/tests/test_gate_partition.py +127 -0
  461. openflo-3.0.0/tests/test_gate_refit.py +277 -0
  462. openflo-3.0.0/tests/test_gate_replication_parent.py +147 -0
  463. openflo-3.0.0/tests/test_gate_text_intensity.py +465 -0
  464. openflo-3.0.0/tests/test_gate_tools_axis_ends.py +291 -0
  465. openflo-3.0.0/tests/test_gatetree.py +116 -0
  466. openflo-3.0.0/tests/test_gating.py +254 -0
  467. openflo-3.0.0/tests/test_gating_helpers.py +205 -0
  468. openflo-3.0.0/tests/test_gatingml.py +361 -0
  469. openflo-3.0.0/tests/test_geomean_flowjo_axis.py +317 -0
  470. openflo-3.0.0/tests/test_gmm_ellipse_info.py +63 -0
  471. openflo-3.0.0/tests/test_golden_regression.py +158 -0
  472. openflo-3.0.0/tests/test_gpu_accel.py +225 -0
  473. openflo-3.0.0/tests/test_gpu_probe_module.py +130 -0
  474. openflo-3.0.0/tests/test_group_stats.py +223 -0
  475. openflo-3.0.0/tests/test_group_stats_window.py +166 -0
  476. openflo-3.0.0/tests/test_grouping.py +224 -0
  477. openflo-3.0.0/tests/test_gui_guard.py +143 -0
  478. openflo-3.0.0/tests/test_gui_smoke.py +1475 -0
  479. openflo-3.0.0/tests/test_gui_toolings.py +129 -0
  480. openflo-3.0.0/tests/test_gui_ux.py +870 -0
  481. openflo-3.0.0/tests/test_highlighted_sample_is_plotted.py +102 -0
  482. openflo-3.0.0/tests/test_histogram_offscale_note.py +94 -0
  483. openflo-3.0.0/tests/test_histogram_resolution.py +113 -0
  484. openflo-3.0.0/tests/test_import_side_effects.py +207 -0
  485. openflo-3.0.0/tests/test_independent_reference.py +125 -0
  486. openflo-3.0.0/tests/test_index_sort.py +70 -0
  487. openflo-3.0.0/tests/test_inline_dialog_tooltips.py +302 -0
  488. openflo-3.0.0/tests/test_inspect_fcs.py +76 -0
  489. openflo-3.0.0/tests/test_install_leaves_no_files.py +126 -0
  490. openflo-3.0.0/tests/test_interop.py +219 -0
  491. openflo-3.0.0/tests/test_interop_h5ad_roundtrip.py +150 -0
  492. openflo-3.0.0/tests/test_keyboard_shortcuts.py +108 -0
  493. openflo-3.0.0/tests/test_knn_threads.py +93 -0
  494. openflo-3.0.0/tests/test_label_align.py +233 -0
  495. openflo-3.0.0/tests/test_label_alignment_module.py +94 -0
  496. openflo-3.0.0/tests/test_layout_bar_all_events.py +91 -0
  497. openflo-3.0.0/tests/test_layout_dialogs_module.py +93 -0
  498. openflo-3.0.0/tests/test_layout_draw_one.py +338 -0
  499. openflo-3.0.0/tests/test_layout_editor.py +1045 -0
  500. openflo-3.0.0/tests/test_layout_editor_controls.py +110 -0
  501. openflo-3.0.0/tests/test_layout_exploded.py +463 -0
  502. openflo-3.0.0/tests/test_layout_figure_help_fixes.py +359 -0
  503. openflo-3.0.0/tests/test_layout_preview_background.py +529 -0
  504. openflo-3.0.0/tests/test_legend_placement.py +120 -0
  505. openflo-3.0.0/tests/test_leiden_seed_range.py +188 -0
  506. openflo-3.0.0/tests/test_load_no_copy.py +89 -0
  507. openflo-3.0.0/tests/test_load_reliability.py +393 -0
  508. openflo-3.0.0/tests/test_loader_queue.py +120 -0
  509. openflo-3.0.0/tests/test_locale_csv.py +98 -0
  510. openflo-3.0.0/tests/test_loglog_mesf.py +71 -0
  511. openflo-3.0.0/tests/test_mass_cytometry.py +510 -0
  512. openflo-3.0.0/tests/test_menus.py +174 -0
  513. openflo-3.0.0/tests/test_module_naming_documented.py +74 -0
  514. openflo-3.0.0/tests/test_mplbackend_set_once.py +134 -0
  515. openflo-3.0.0/tests/test_multiple_af_signatures.py +164 -0
  516. openflo-3.0.0/tests/test_n5_theme_names.py +90 -0
  517. openflo-3.0.0/tests/test_n9_names.py +256 -0
  518. openflo-3.0.0/tests/test_named_fonts.py +69 -0
  519. openflo-3.0.0/tests/test_narrowed_catches.py +611 -0
  520. openflo-3.0.0/tests/test_no_init_attr_guards.py +96 -0
  521. openflo-3.0.0/tests/test_no_pass_through_wrappers.py +103 -0
  522. openflo-3.0.0/tests/test_no_plausible_fills.py +251 -0
  523. openflo-3.0.0/tests/test_normalise_groups_idempotent.py +82 -0
  524. openflo-3.0.0/tests/test_not_gate_excludes_unmeasured.py +117 -0
  525. openflo-3.0.0/tests/test_numba_threads.py +203 -0
  526. openflo-3.0.0/tests/test_one_app_version.py +130 -0
  527. openflo-3.0.0/tests/test_optimize_compensation_truth.py +205 -0
  528. openflo-3.0.0/tests/test_panel_check.py +328 -0
  529. openflo-3.0.0/tests/test_panel_check_window.py +299 -0
  530. openflo-3.0.0/tests/test_panel_pairs.py +145 -0
  531. openflo-3.0.0/tests/test_parallel_load_blas.py +124 -0
  532. openflo-3.0.0/tests/test_parse_ellipsoid.py +925 -0
  533. openflo-3.0.0/tests/test_phenograph_no_fork.py +230 -0
  534. openflo-3.0.0/tests/test_phenograph_pool_cap.py +82 -0
  535. openflo-3.0.0/tests/test_phenograph_prints_to_log.py +119 -0
  536. openflo-3.0.0/tests/test_phenograph_resolution.py +404 -0
  537. openflo-3.0.0/tests/test_pid_alive_agrees.py +88 -0
  538. openflo-3.0.0/tests/test_pipeline_missing_embedding.py +98 -0
  539. openflo-3.0.0/tests/test_plate_view.py +118 -0
  540. openflo-3.0.0/tests/test_plot_cache.py +712 -0
  541. openflo-3.0.0/tests/test_plot_defaults.py +402 -0
  542. openflo-3.0.0/tests/test_polygon_gating.py +239 -0
  543. openflo-3.0.0/tests/test_polygon_rect_tools.py +429 -0
  544. openflo-3.0.0/tests/test_population_export_background.py +313 -0
  545. openflo-3.0.0/tests/test_population_stats_geomean.py +173 -0
  546. openflo-3.0.0/tests/test_population_stats_speed.py +310 -0
  547. openflo-3.0.0/tests/test_population_stats_split.py +488 -0
  548. openflo-3.0.0/tests/test_preexisting_data_defects.py +318 -0
  549. openflo-3.0.0/tests/test_prepare_unit_tags.py +167 -0
  550. openflo-3.0.0/tests/test_preview_quadrants.py +104 -0
  551. openflo-3.0.0/tests/test_private_api_names.py +106 -0
  552. openflo-3.0.0/tests/test_proliferation.py +145 -0
  553. openflo-3.0.0/tests/test_propagate_downsample_order.py +144 -0
  554. openflo-3.0.0/tests/test_public_api_annotations.py +158 -0
  555. openflo-3.0.0/tests/test_python_readiness.py +377 -0
  556. openflo-3.0.0/tests/test_q3_deprecations.py +106 -0
  557. openflo-3.0.0/tests/test_qc.py +231 -0
  558. openflo-3.0.0/tests/test_qc_ignores_derived_columns.py +206 -0
  559. openflo-3.0.0/tests/test_qc_integer_channels.py +198 -0
  560. openflo-3.0.0/tests/test_qc_removed_visible.py +150 -0
  561. openflo-3.0.0/tests/test_quadrant_tool.py +289 -0
  562. openflo-3.0.0/tests/test_quick_preview_dialog.py +107 -0
  563. openflo-3.0.0/tests/test_raw_not_aliased.py +104 -0
  564. openflo-3.0.0/tests/test_record_export.py +327 -0
  565. openflo-3.0.0/tests/test_reexports_point_home.py +108 -0
  566. openflo-3.0.0/tests/test_regroup_dragged_samples.py +228 -0
  567. openflo-3.0.0/tests/test_release_mirror_gate.py +153 -0
  568. openflo-3.0.0/tests/test_remove_from_tab.py +339 -0
  569. openflo-3.0.0/tests/test_remove_sample_confirm.py +215 -0
  570. openflo-3.0.0/tests/test_removed_overlay_drawn.py +177 -0
  571. openflo-3.0.0/tests/test_reopen_active_sample.py +121 -0
  572. openflo-3.0.0/tests/test_result_stamps.py +358 -0
  573. openflo-3.0.0/tests/test_resume_and_transfer_hygiene.py +136 -0
  574. openflo-3.0.0/tests/test_right_click_macos.py +213 -0
  575. openflo-3.0.0/tests/test_robust.py +56 -0
  576. openflo-3.0.0/tests/test_run_events_binary.py +476 -0
  577. openflo-3.0.0/tests/test_run_louvain.py +99 -0
  578. openflo-3.0.0/tests/test_sample_groups.py +97 -0
  579. openflo-3.0.0/tests/test_sample_helpers_modules.py +94 -0
  580. openflo-3.0.0/tests/test_sample_qc_window_values.py +96 -0
  581. openflo-3.0.0/tests/test_saturation_report.py +152 -0
  582. openflo-3.0.0/tests/test_saved_file_notice.py +107 -0
  583. openflo-3.0.0/tests/test_saved_stats.py +449 -0
  584. openflo-3.0.0/tests/test_scales.py +39 -0
  585. openflo-3.0.0/tests/test_scatter_channel_lookup.py +55 -0
  586. openflo-3.0.0/tests/test_scrolled_panels.py +537 -0
  587. openflo-3.0.0/tests/test_sdist_contents.py +348 -0
  588. openflo-3.0.0/tests/test_seed_phrases.py +102 -0
  589. openflo-3.0.0/tests/test_seeds_and_cluster_modules.py +96 -0
  590. openflo-3.0.0/tests/test_selftest.py +51 -0
  591. openflo-3.0.0/tests/test_selftest_controls.py +194 -0
  592. openflo-3.0.0/tests/test_session.py +417 -0
  593. openflo-3.0.0/tests/test_session_compare.py +147 -0
  594. openflo-3.0.0/tests/test_session_continuity.py +132 -0
  595. openflo-3.0.0/tests/test_session_json_is_valid.py +111 -0
  596. openflo-3.0.0/tests/test_session_lock.py +386 -0
  597. openflo-3.0.0/tests/test_session_missing_samples.py +155 -0
  598. openflo-3.0.0/tests/test_session_reopen_downsample.py +211 -0
  599. openflo-3.0.0/tests/test_session_reopen_losses.py +286 -0
  600. openflo-3.0.0/tests/test_session_reopen_models.py +273 -0
  601. openflo-3.0.0/tests/test_session_reopen_more.py +498 -0
  602. openflo-3.0.0/tests/test_session_restore_transforms.py +245 -0
  603. openflo-3.0.0/tests/test_session_safety.py +400 -0
  604. openflo-3.0.0/tests/test_session_saves_label_populations.py +81 -0
  605. openflo-3.0.0/tests/test_session_sidecar_collision.py +194 -0
  606. openflo-3.0.0/tests/test_session_sidecar_cut_short.py +119 -0
  607. openflo-3.0.0/tests/test_set_aside_samples.py +102 -0
  608. openflo-3.0.0/tests/test_setup_installer.py +530 -0
  609. openflo-3.0.0/tests/test_shared_clustering_aliases.py +48 -0
  610. openflo-3.0.0/tests/test_significance_brackets.py +201 -0
  611. openflo-3.0.0/tests/test_small_n_statistics.py +466 -0
  612. openflo-3.0.0/tests/test_som_kernel.py +191 -0
  613. openflo-3.0.0/tests/test_spectral.py +305 -0
  614. openflo-3.0.0/tests/test_spectral_matched_af.py +340 -0
  615. openflo-3.0.0/tests/test_spectral_measured_and_spread.py +273 -0
  616. openflo-3.0.0/tests/test_spectral_qc_window.py +188 -0
  617. openflo-3.0.0/tests/test_spectral_unmix_gui.py +247 -0
  618. openflo-3.0.0/tests/test_spectral_unmix_nonfinite.py +96 -0
  619. openflo-3.0.0/tests/test_spreading_matrix.py +109 -0
  620. openflo-3.0.0/tests/test_stat_forms.py +127 -0
  621. openflo-3.0.0/tests/test_statistic_columns.py +230 -0
  622. openflo-3.0.0/tests/test_statistics_export.py +240 -0
  623. openflo-3.0.0/tests/test_statistics_population_column.py +121 -0
  624. openflo-3.0.0/tests/test_statistics_tabs.py +291 -0
  625. openflo-3.0.0/tests/test_stats.py +717 -0
  626. openflo-3.0.0/tests/test_stats_library_ignores_prefs.py +161 -0
  627. openflo-3.0.0/tests/test_survivor_boundaries.py +104 -0
  628. openflo-3.0.0/tests/test_synthetic.py +239 -0
  629. openflo-3.0.0/tests/test_tab_own_gates.py +457 -0
  630. openflo-3.0.0/tests/test_template_apply.py +160 -0
  631. openflo-3.0.0/tests/test_template_one_sample.py +67 -0
  632. openflo-3.0.0/tests/test_templates.py +65 -0
  633. openflo-3.0.0/tests/test_test_files_named_by_feature.py +45 -0
  634. openflo-3.0.0/tests/test_theme.py +184 -0
  635. openflo-3.0.0/tests/test_theme_colours.py +159 -0
  636. openflo-3.0.0/tests/test_time_channel_name.py +124 -0
  637. openflo-3.0.0/tests/test_time_checks_shared.py +202 -0
  638. openflo-3.0.0/tests/test_tk_aqua_update.py +188 -0
  639. openflo-3.0.0/tests/test_tk_failures_use_gui_unavailable.py +67 -0
  640. openflo-3.0.0/tests/test_tk_keep_on_aqua.py +64 -0
  641. openflo-3.0.0/tests/test_tk_main_thread.py +321 -0
  642. openflo-3.0.0/tests/test_tk_threads.py +118 -0
  643. openflo-3.0.0/tests/test_tk_waits_shared.py +169 -0
  644. openflo-3.0.0/tests/test_to_anndata_assembly.py +134 -0
  645. openflo-3.0.0/tests/test_tool_window.py +89 -0
  646. openflo-3.0.0/tests/test_trajectory.py +137 -0
  647. openflo-3.0.0/tests/test_trajectory_window.py +136 -0
  648. openflo-3.0.0/tests/test_transform_consistency.py +71 -0
  649. openflo-3.0.0/tests/test_transform_record_writes.py +137 -0
  650. openflo-3.0.0/tests/test_transform_records.py +634 -0
  651. openflo-3.0.0/tests/test_transform_sidecar_module.py +178 -0
  652. openflo-3.0.0/tests/test_transforms.py +67 -0
  653. openflo-3.0.0/tests/test_transparent_exports.py +353 -0
  654. openflo-3.0.0/tests/test_tree_refresh_on_replot.py +52 -0
  655. openflo-3.0.0/tests/test_typecheck_config.py +185 -0
  656. openflo-3.0.0/tests/test_typecheck_mode.py +27 -0
  657. openflo-3.0.0/tests/test_ui_audit_sign_verify.py +125 -0
  658. openflo-3.0.0/tests/test_ui_compare.py +20 -0
  659. openflo-3.0.0/tests/test_ui_dialogs_smoke.py +169 -0
  660. openflo-3.0.0/tests/test_ui_inspect.py +81 -0
  661. openflo-3.0.0/tests/test_ui_logic.py +76 -0
  662. openflo-3.0.0/tests/test_ui_preview.py +21 -0
  663. openflo-3.0.0/tests/test_ui_sample_qc.py +78 -0
  664. openflo-3.0.0/tests/test_ui_synth.py +76 -0
  665. openflo-3.0.0/tests/test_ui_voltage.py +71 -0
  666. openflo-3.0.0/tests/test_ui_wording.py +120 -0
  667. openflo-3.0.0/tests/test_undo_redo.py +122 -0
  668. openflo-3.0.0/tests/test_unknown_population_is_missing.py +143 -0
  669. openflo-3.0.0/tests/test_unknown_scale_export.py +167 -0
  670. openflo-3.0.0/tests/test_unknown_scale_import_is_missing.py +83 -0
  671. openflo-3.0.0/tests/test_update_flow.py +257 -0
  672. openflo-3.0.0/tests/test_user_guide.py +173 -0
  673. openflo-3.0.0/tests/test_v261_sessions.py +373 -0
  674. openflo-3.0.0/tests/test_vector_figure_export.py +88 -0
  675. openflo-3.0.0/tests/test_version_strings.py +68 -0
  676. openflo-3.0.0/tests/test_view_tabs.py +506 -0
  677. openflo-3.0.0/tests/test_voltage_cli.py +103 -0
  678. openflo-3.0.0/tests/test_voltage_titration_dialog.py +132 -0
  679. openflo-3.0.0/tests/test_warning_filters.py +124 -0
  680. openflo-3.0.0/tests/test_westgard_rules.py +144 -0
  681. openflo-3.0.0/tests/test_westgard_within_run.py +308 -0
  682. openflo-3.0.0/tests/test_window_fit.py +459 -0
  683. openflo-3.0.0/tests/test_window_title.py +39 -0
  684. openflo-3.0.0/tests/test_windows_nothing_hidden.py +383 -0
  685. openflo-3.0.0/tests/test_worker_threads_use_run_async.py +244 -0
  686. openflo-3.0.0/tests/test_workspace.py +1002 -0
  687. openflo-3.0.0/tests/test_workspace_batch_folder.py +155 -0
  688. openflo-3.0.0/tests/test_workspace_gate_names.py +74 -0
  689. openflo-3.0.0/tests/test_workspace_job_driver.py +233 -0
  690. openflo-3.0.0/tests/test_workspace_run_split.py +95 -0
  691. openflo-3.0.0/tests/test_write_fcs.py +53 -0
  692. openflo-3.0.0/tests/test_write_fcs_order.py +119 -0
  693. openflo-3.0.0/tests/test_wsp_axis_floor.py +443 -0
  694. openflo-3.0.0/tests/test_wsp_boolean_gates.py +154 -0
  695. openflo-3.0.0/tests/test_wsp_export_counts.py +159 -0
  696. openflo-3.0.0/tests/test_wsp_export_golden.py +128 -0
  697. openflo-3.0.0/tests/test_wsp_gate_units.py +288 -0
  698. openflo-3.0.0/tests/test_wsp_groups_subfolders.py +154 -0
  699. openflo-3.0.0/tests/test_wsp_polygon_edge_axes.py +210 -0
  700. openflo-3.0.0/tests/test_wsp_population_names.py +101 -0
  701. openflo-3.0.0/tests/test_wsp_quadrant_export.py +111 -0
  702. openflo-3.0.0/tests/test_wsp_reader_module.py +98 -0
  703. openflo-3.0.0/tests/test_wsp_sample_compensation.py +147 -0
  704. openflo-3.0.0/tests/test_wsp_sample_keywords.py +90 -0
  705. openflo-3.0.0/tests/test_wsp_time_edge_check.py +74 -0
  706. openflo-3.0.0/tests/test_wsp_time_units.py +438 -0
  707. openflo-3.0.0/tests/test_wsp_unnamed_population_names.py +90 -0
  708. openflo-3.0.0/tests/test_wsp_version_and_event_checks.py +188 -0
  709. openflo-3.0.0/tests/test_wsp_writer.py +362 -0
  710. openflo-3.0.0/tests/test_wsp_writer_module.py +98 -0
  711. openflo-3.0.0/tests/tk_aqua_update.py +115 -0
  712. openflo-3.0.0/tests/tk_main_thread.py +240 -0
  713. openflo-2.7.0b2/CHANGELOG.md +0 -3596
  714. openflo-2.7.0b2/MANIFEST.in +0 -12
  715. openflo-2.7.0b2/PKG-INFO +0 -591
  716. openflo-2.7.0b2/README.md +0 -528
  717. openflo-2.7.0b2/pyproject.toml +0 -250
  718. openflo-2.7.0b2/scripts/make_reference_values.py +0 -243
  719. openflo-2.7.0b2/scripts/migrate_session.py +0 -71
  720. openflo-2.7.0b2/scripts/preflight.py +0 -232
  721. openflo-2.7.0b2/scripts/prove_tests.py +0 -1299
  722. openflo-2.7.0b2/scripts/run_analyses.py +0 -303
  723. openflo-2.7.0b2/scripts/theme_audit.py +0 -279
  724. openflo-2.7.0b2/src/openflo/__init__.py +0 -291
  725. openflo-2.7.0b2/src/openflo/_console.py +0 -39
  726. openflo-2.7.0b2/src/openflo/_golden.json +0 -53
  727. openflo-2.7.0b2/src/openflo/acquisition.py +0 -117
  728. openflo-2.7.0b2/src/openflo/async_task.py +0 -338
  729. openflo-2.7.0b2/src/openflo/audit.py +0 -156
  730. openflo-2.7.0b2/src/openflo/batch_normalize.py +0 -378
  731. openflo-2.7.0b2/src/openflo/bead_tracking.py +0 -163
  732. openflo-2.7.0b2/src/openflo/calibration.py +0 -169
  733. openflo-2.7.0b2/src/openflo/capabilities.py +0 -247
  734. openflo-2.7.0b2/src/openflo/cellcycle_djf.py +0 -124
  735. openflo-2.7.0b2/src/openflo/cli.py +0 -3725
  736. openflo-2.7.0b2/src/openflo/cli_args.py +0 -259
  737. openflo-2.7.0b2/src/openflo/clipboard_export.py +0 -177
  738. openflo-2.7.0b2/src/openflo/comp_qc.py +0 -173
  739. openflo-2.7.0b2/src/openflo/comp_spread.py +0 -71
  740. openflo-2.7.0b2/src/openflo/compare.py +0 -632
  741. openflo-2.7.0b2/src/openflo/density.py +0 -200
  742. openflo-2.7.0b2/src/openflo/derived.py +0 -262
  743. openflo-2.7.0b2/src/openflo/diagnostics.py +0 -251
  744. openflo-2.7.0b2/src/openflo/dialog_inputs.py +0 -96
  745. openflo-2.7.0b2/src/openflo/diffexp.py +0 -605
  746. openflo-2.7.0b2/src/openflo/dr_compare.py +0 -165
  747. openflo-2.7.0b2/src/openflo/editor_analysis.py +0 -722
  748. openflo-2.7.0b2/src/openflo/editor_audit.py +0 -41
  749. openflo-2.7.0b2/src/openflo/editor_autoclean.py +0 -456
  750. openflo-2.7.0b2/src/openflo/editor_autogate.py +0 -209
  751. openflo-2.7.0b2/src/openflo/editor_base.py +0 -31
  752. openflo-2.7.0b2/src/openflo/editor_channels.py +0 -115
  753. openflo-2.7.0b2/src/openflo/editor_chrome.py +0 -559
  754. openflo-2.7.0b2/src/openflo/editor_clipboard.py +0 -589
  755. openflo-2.7.0b2/src/openflo/editor_combos.py +0 -213
  756. openflo-2.7.0b2/src/openflo/editor_compute.py +0 -665
  757. openflo-2.7.0b2/src/openflo/editor_console.py +0 -156
  758. openflo-2.7.0b2/src/openflo/editor_coords.py +0 -139
  759. openflo-2.7.0b2/src/openflo/editor_dnd.py +0 -813
  760. openflo-2.7.0b2/src/openflo/editor_downsample.py +0 -188
  761. openflo-2.7.0b2/src/openflo/editor_export.py +0 -613
  762. openflo-2.7.0b2/src/openflo/editor_figure.py +0 -204
  763. openflo-2.7.0b2/src/openflo/editor_fixes.py +0 -156
  764. openflo-2.7.0b2/src/openflo/editor_gatetools.py +0 -1125
  765. openflo-2.7.0b2/src/openflo/editor_gating.py +0 -1660
  766. openflo-2.7.0b2/src/openflo/editor_grouping.py +0 -186
  767. openflo-2.7.0b2/src/openflo/editor_help.py +0 -391
  768. openflo-2.7.0b2/src/openflo/editor_keyboard.py +0 -152
  769. openflo-2.7.0b2/src/openflo/editor_lifecycle.py +0 -826
  770. openflo-2.7.0b2/src/openflo/editor_load.py +0 -841
  771. openflo-2.7.0b2/src/openflo/editor_loadpool.py +0 -749
  772. openflo-2.7.0b2/src/openflo/editor_menu.py +0 -365
  773. openflo-2.7.0b2/src/openflo/editor_mode.py +0 -116
  774. openflo-2.7.0b2/src/openflo/editor_plot.py +0 -1942
  775. openflo-2.7.0b2/src/openflo/editor_plotcache.py +0 -1201
  776. openflo-2.7.0b2/src/openflo/editor_populations.py +0 -605
  777. openflo-2.7.0b2/src/openflo/editor_refit.py +0 -102
  778. openflo-2.7.0b2/src/openflo/editor_samples.py +0 -83
  779. openflo-2.7.0b2/src/openflo/editor_session.py +0 -1734
  780. openflo-2.7.0b2/src/openflo/editor_slider.py +0 -244
  781. openflo-2.7.0b2/src/openflo/editor_specialist.py +0 -49
  782. openflo-2.7.0b2/src/openflo/editor_stats.py +0 -307
  783. openflo-2.7.0b2/src/openflo/editor_tabs.py +0 -699
  784. openflo-2.7.0b2/src/openflo/editor_template.py +0 -413
  785. openflo-2.7.0b2/src/openflo/editor_tools.py +0 -650
  786. openflo-2.7.0b2/src/openflo/editor_tree.py +0 -771
  787. openflo-2.7.0b2/src/openflo/editor_undo.py +0 -150
  788. openflo-2.7.0b2/src/openflo/editor_update.py +0 -144
  789. openflo-2.7.0b2/src/openflo/export_stamp.py +0 -156
  790. openflo-2.7.0b2/src/openflo/fcs_export.py +0 -254
  791. openflo-2.7.0b2/src/openflo/fcs_read.py +0 -68
  792. openflo-2.7.0b2/src/openflo/fixes.py +0 -620
  793. openflo-2.7.0b2/src/openflo/fluor_catalog.py +0 -29
  794. openflo-2.7.0b2/src/openflo/gate_combos.py +0 -235
  795. openflo-2.7.0b2/src/openflo/gate_coords.py +0 -280
  796. openflo-2.7.0b2/src/openflo/gate_refit.py +0 -226
  797. openflo-2.7.0b2/src/openflo/gate_table.py +0 -81
  798. openflo-2.7.0b2/src/openflo/gatetree.py +0 -232
  799. openflo-2.7.0b2/src/openflo/gating.py +0 -628
  800. openflo-2.7.0b2/src/openflo/gating_helpers.py +0 -187
  801. openflo-2.7.0b2/src/openflo/gatingml.py +0 -753
  802. openflo-2.7.0b2/src/openflo/gpu_accel.py +0 -318
  803. openflo-2.7.0b2/src/openflo/gui.py +0 -2860
  804. openflo-2.7.0b2/src/openflo/guide/02-loading-samples.md +0 -23
  805. openflo-2.7.0b2/src/openflo/guide/03-gating.md +0 -38
  806. openflo-2.7.0b2/src/openflo/guide/04-compensation-and-scales.md +0 -27
  807. openflo-2.7.0b2/src/openflo/guide/05-statistics-and-figures.md +0 -22
  808. openflo-2.7.0b2/src/openflo/guide/06-analysis-tools.md +0 -21
  809. openflo-2.7.0b2/src/openflo/guide/07-spectral.md +0 -13
  810. openflo-2.7.0b2/src/openflo/guide/08-quality-control.md +0 -18
  811. openflo-2.7.0b2/src/openflo/guide/09-sharing.md +0 -27
  812. openflo-2.7.0b2/src/openflo/guide/10-troubleshooting.md +0 -17
  813. openflo-2.7.0b2/src/openflo/index_sort.py +0 -90
  814. openflo-2.7.0b2/src/openflo/inspect_fcs.py +0 -29
  815. openflo-2.7.0b2/src/openflo/interop.py +0 -240
  816. openflo-2.7.0b2/src/openflo/kinetics.py +0 -135
  817. openflo-2.7.0b2/src/openflo/layout_page.py +0 -610
  818. openflo-2.7.0b2/src/openflo/layout_render.py +0 -976
  819. openflo-2.7.0b2/src/openflo/paths.py +0 -100
  820. openflo-2.7.0b2/src/openflo/pipeline.py +0 -9891
  821. openflo-2.7.0b2/src/openflo/plotcache.py +0 -414
  822. openflo-2.7.0b2/src/openflo/plotmath.py +0 -393
  823. openflo-2.7.0b2/src/openflo/prefs.py +0 -90
  824. openflo-2.7.0b2/src/openflo/preview.py +0 -236
  825. openflo-2.7.0b2/src/openflo/proliferation.py +0 -128
  826. openflo-2.7.0b2/src/openflo/provenance.py +0 -337
  827. openflo-2.7.0b2/src/openflo/report.py +0 -112
  828. openflo-2.7.0b2/src/openflo/result_stamp.py +0 -315
  829. openflo-2.7.0b2/src/openflo/sample_meta.py +0 -157
  830. openflo-2.7.0b2/src/openflo/saturation.py +0 -147
  831. openflo-2.7.0b2/src/openflo/saved_stats.py +0 -197
  832. openflo-2.7.0b2/src/openflo/scales.py +0 -652
  833. openflo-2.7.0b2/src/openflo/selftest.py +0 -288
  834. openflo-2.7.0b2/src/openflo/selftest_controls.py +0 -644
  835. openflo-2.7.0b2/src/openflo/session_compare.py +0 -168
  836. openflo-2.7.0b2/src/openflo/session_format.py +0 -156
  837. openflo-2.7.0b2/src/openflo/session_lock.py +0 -189
  838. openflo-2.7.0b2/src/openflo/session_sidecar.py +0 -129
  839. openflo-2.7.0b2/src/openflo/som_kernel.py +0 -141
  840. openflo-2.7.0b2/src/openflo/spectral.py +0 -927
  841. openflo-2.7.0b2/src/openflo/spectral_af.py +0 -136
  842. openflo-2.7.0b2/src/openflo/stats.py +0 -686
  843. openflo-2.7.0b2/src/openflo/synthetic.py +0 -737
  844. openflo-2.7.0b2/src/openflo/table_export.py +0 -336
  845. openflo-2.7.0b2/src/openflo/theme.py +0 -480
  846. openflo-2.7.0b2/src/openflo/tk_threads.py +0 -104
  847. openflo-2.7.0b2/src/openflo/tool_window.py +0 -513
  848. openflo-2.7.0b2/src/openflo/tree_ids.py +0 -63
  849. openflo-2.7.0b2/src/openflo/ui_abscounts.py +0 -148
  850. openflo-2.7.0b2/src/openflo/ui_annotation.py +0 -186
  851. openflo-2.7.0b2/src/openflo/ui_audit.py +0 -222
  852. openflo-2.7.0b2/src/openflo/ui_axis_config.py +0 -262
  853. openflo-2.7.0b2/src/openflo/ui_batch_normalize.py +0 -473
  854. openflo-2.7.0b2/src/openflo/ui_bead_tracking.py +0 -248
  855. openflo-2.7.0b2/src/openflo/ui_calibration.py +0 -252
  856. openflo-2.7.0b2/src/openflo/ui_cell_cycle.py +0 -101
  857. openflo-2.7.0b2/src/openflo/ui_comp.py +0 -702
  858. openflo-2.7.0b2/src/openflo/ui_comp_spread.py +0 -260
  859. openflo-2.7.0b2/src/openflo/ui_compare.py +0 -365
  860. openflo-2.7.0b2/src/openflo/ui_concatenate.py +0 -122
  861. openflo-2.7.0b2/src/openflo/ui_diff.py +0 -349
  862. openflo-2.7.0b2/src/openflo/ui_embedding.py +0 -114
  863. openflo-2.7.0b2/src/openflo/ui_errors.py +0 -174
  864. openflo-2.7.0b2/src/openflo/ui_expression.py +0 -504
  865. openflo-2.7.0b2/src/openflo/ui_figure_layout.py +0 -122
  866. openflo-2.7.0b2/src/openflo/ui_figure_window.py +0 -55
  867. openflo-2.7.0b2/src/openflo/ui_fixes.py +0 -304
  868. openflo-2.7.0b2/src/openflo/ui_flowsom_tree.py +0 -158
  869. openflo-2.7.0b2/src/openflo/ui_fmo.py +0 -472
  870. openflo-2.7.0b2/src/openflo/ui_frequency.py +0 -785
  871. openflo-2.7.0b2/src/openflo/ui_group_stats.py +0 -139
  872. openflo-2.7.0b2/src/openflo/ui_guide.py +0 -149
  873. openflo-2.7.0b2/src/openflo/ui_inspect.py +0 -253
  874. openflo-2.7.0b2/src/openflo/ui_kinetics.py +0 -358
  875. openflo-2.7.0b2/src/openflo/ui_layout_editor.py +0 -2448
  876. openflo-2.7.0b2/src/openflo/ui_logic.py +0 -83
  877. openflo-2.7.0b2/src/openflo/ui_methods.py +0 -99
  878. openflo-2.7.0b2/src/openflo/ui_notice.py +0 -77
  879. openflo-2.7.0b2/src/openflo/ui_plate.py +0 -262
  880. openflo-2.7.0b2/src/openflo/ui_polyfunction.py +0 -176
  881. openflo-2.7.0b2/src/openflo/ui_preferences.py +0 -464
  882. openflo-2.7.0b2/src/openflo/ui_preview.py +0 -292
  883. openflo-2.7.0b2/src/openflo/ui_proliferation.py +0 -269
  884. openflo-2.7.0b2/src/openflo/ui_refit.py +0 -104
  885. openflo-2.7.0b2/src/openflo/ui_sample_groups.py +0 -152
  886. openflo-2.7.0b2/src/openflo/ui_sample_qc.py +0 -214
  887. openflo-2.7.0b2/src/openflo/ui_saturation.py +0 -182
  888. openflo-2.7.0b2/src/openflo/ui_spectral_qc.py +0 -248
  889. openflo-2.7.0b2/src/openflo/ui_spectral_unmix.py +0 -132
  890. openflo-2.7.0b2/src/openflo/ui_statistics.py +0 -662
  891. openflo-2.7.0b2/src/openflo/ui_synth.py +0 -313
  892. openflo-2.7.0b2/src/openflo/ui_table.py +0 -122
  893. openflo-2.7.0b2/src/openflo/ui_trajectory.py +0 -312
  894. openflo-2.7.0b2/src/openflo/ui_unmixed_pairs.py +0 -51
  895. openflo-2.7.0b2/src/openflo/ui_voltage.py +0 -353
  896. openflo-2.7.0b2/src/openflo/update.py +0 -291
  897. openflo-2.7.0b2/src/openflo/user_guide.py +0 -114
  898. openflo-2.7.0b2/src/openflo/view_tabs.py +0 -315
  899. openflo-2.7.0b2/src/openflo/voltage.py +0 -377
  900. openflo-2.7.0b2/src/openflo/warmup.py +0 -94
  901. openflo-2.7.0b2/src/openflo/workspace.py +0 -3485
  902. openflo-2.7.0b2/src/openflo.egg-info/PKG-INFO +0 -591
  903. openflo-2.7.0b2/src/openflo.egg-info/SOURCES.txt +0 -545
  904. openflo-2.7.0b2/tests/_fcs_writer.py +0 -59
  905. openflo-2.7.0b2/tests/conftest.py +0 -572
  906. openflo-2.7.0b2/tests/test_abscounts.py +0 -148
  907. openflo-2.7.0b2/tests/test_acquisition_qc_ground_truth.py +0 -462
  908. openflo-2.7.0b2/tests/test_annotate.py +0 -128
  909. openflo-2.7.0b2/tests/test_annotate_clusters.py +0 -137
  910. openflo-2.7.0b2/tests/test_annotate_rename_dialogs.py +0 -103
  911. openflo-2.7.0b2/tests/test_annotation_window.py +0 -107
  912. openflo-2.7.0b2/tests/test_autoclean_counts.py +0 -159
  913. openflo-2.7.0b2/tests/test_autoclean_doublets_scale.py +0 -140
  914. openflo-2.7.0b2/tests/test_autoclean_editor_actions.py +0 -330
  915. openflo-2.7.0b2/tests/test_autoclean_methods.py +0 -646
  916. openflo-2.7.0b2/tests/test_autogate.py +0 -206
  917. openflo-2.7.0b2/tests/test_autogate_dialog.py +0 -201
  918. openflo-2.7.0b2/tests/test_autogate_reproducible.py +0 -193
  919. openflo-2.7.0b2/tests/test_axis_intensity_ticks.py +0 -178
  920. openflo-2.7.0b2/tests/test_batch_d_responsive.py +0 -745
  921. openflo-2.7.0b2/tests/test_batch_e_quick_features.py +0 -599
  922. openflo-2.7.0b2/tests/test_batch_g_cli_sharding.py +0 -248
  923. openflo-2.7.0b2/tests/test_batch_h_consistent_dialogs.py +0 -560
  924. openflo-2.7.0b2/tests/test_batch_i_menus_fonts_colours.py +0 -608
  925. openflo-2.7.0b2/tests/test_batch_j_data_handling.py +0 -382
  926. openflo-2.7.0b2/tests/test_batch_k_specialist.py +0 -740
  927. openflo-2.7.0b2/tests/test_batch_l_offline_help.py +0 -187
  928. openflo-2.7.0b2/tests/test_batch_n_exports_access.py +0 -582
  929. openflo-2.7.0b2/tests/test_batch_normalize.py +0 -323
  930. openflo-2.7.0b2/tests/test_batch_normalize_dialog.py +0 -203
  931. openflo-2.7.0b2/tests/test_beta_look.py +0 -364
  932. openflo-2.7.0b2/tests/test_beta_release.py +0 -21
  933. openflo-2.7.0b2/tests/test_boolean_gate_fails_closed.py +0 -244
  934. openflo-2.7.0b2/tests/test_boolean_gate_tools.py +0 -751
  935. openflo-2.7.0b2/tests/test_calibration_dialog.py +0 -158
  936. openflo-2.7.0b2/tests/test_cell_cycle.py +0 -149
  937. openflo-2.7.0b2/tests/test_cell_cycle_dialog.py +0 -189
  938. openflo-2.7.0b2/tests/test_channel_classification.py +0 -93
  939. openflo-2.7.0b2/tests/test_ci_full_matrix_events.py +0 -107
  940. openflo-2.7.0b2/tests/test_ci_macos_job.py +0 -44
  941. openflo-2.7.0b2/tests/test_ci_privacy_gate_on_mirror.py +0 -54
  942. openflo-2.7.0b2/tests/test_clean_install_round2.py +0 -121
  943. openflo-2.7.0b2/tests/test_cli_cluster.py +0 -58
  944. openflo-2.7.0b2/tests/test_cli_export_wsp.py +0 -53
  945. openflo-2.7.0b2/tests/test_cli_reproducibility.py +0 -270
  946. openflo-2.7.0b2/tests/test_cli_run.py +0 -1286
  947. openflo-2.7.0b2/tests/test_cli_run_options.py +0 -143
  948. openflo-2.7.0b2/tests/test_cli_shared_clusters.py +0 -456
  949. openflo-2.7.0b2/tests/test_cli_unmix.py +0 -128
  950. openflo-2.7.0b2/tests/test_clipboard_move_marker.py +0 -209
  951. openflo-2.7.0b2/tests/test_close_save.py +0 -418
  952. openflo-2.7.0b2/tests/test_cluster_dialog.py +0 -669
  953. openflo-2.7.0b2/tests/test_cluster_gpu_fallback.py +0 -170
  954. openflo-2.7.0b2/tests/test_cluster_heatmap.py +0 -72
  955. openflo-2.7.0b2/tests/test_cluster_labels.py +0 -49
  956. openflo-2.7.0b2/tests/test_cluster_recovery_highdim.py +0 -142
  957. openflo-2.7.0b2/tests/test_cluster_reproducible.py +0 -107
  958. openflo-2.7.0b2/tests/test_clustering_known_defects.py +0 -187
  959. openflo-2.7.0b2/tests/test_clustering_reproducible.py +0 -131
  960. openflo-2.7.0b2/tests/test_combination_gates.py +0 -323
  961. openflo-2.7.0b2/tests/test_comp_apply_keeps_computed.py +0 -254
  962. openflo-2.7.0b2/tests/test_comp_editor.py +0 -55
  963. openflo-2.7.0b2/tests/test_comp_matrix_csv_quoting.py +0 -56
  964. openflo-2.7.0b2/tests/test_compare_all.py +0 -94
  965. openflo-2.7.0b2/tests/test_compare_conditions_joint_clustering.py +0 -103
  966. openflo-2.7.0b2/tests/test_compare_dialog.py +0 -134
  967. openflo-2.7.0b2/tests/test_compare_flowjo_semantics.py +0 -231
  968. openflo-2.7.0b2/tests/test_compensation.py +0 -131
  969. openflo-2.7.0b2/tests/test_compensation_applied_once.py +0 -352
  970. openflo-2.7.0b2/tests/test_compensation_editor_apply.py +0 -443
  971. openflo-2.7.0b2/tests/test_compensation_fcs_spill.py +0 -178
  972. openflo-2.7.0b2/tests/test_compensation_matrix_reading.py +0 -258
  973. openflo-2.7.0b2/tests/test_copy_gates_dialog.py +0 -123
  974. openflo-2.7.0b2/tests/test_cytonorm.py +0 -177
  975. openflo-2.7.0b2/tests/test_cytonorm_qc_undefined.py +0 -112
  976. openflo-2.7.0b2/tests/test_density.py +0 -117
  977. openflo-2.7.0b2/tests/test_derived_kinetics.py +0 -403
  978. openflo-2.7.0b2/tests/test_diagnostics.py +0 -95
  979. openflo-2.7.0b2/tests/test_dialog_tooltips.py +0 -462
  980. openflo-2.7.0b2/tests/test_diff_abundance_small_n.py +0 -124
  981. openflo-2.7.0b2/tests/test_diffabundance_calibration.py +0 -196
  982. openflo-2.7.0b2/tests/test_diffexp.py +0 -162
  983. openflo-2.7.0b2/tests/test_doctor_tk_check.py +0 -92
  984. openflo-2.7.0b2/tests/test_drag_tree_refresh.py +0 -118
  985. openflo-2.7.0b2/tests/test_editor_mixins.py +0 -89
  986. openflo-2.7.0b2/tests/test_editor_tools.py +0 -381
  987. openflo-2.7.0b2/tests/test_ellipse_params_geometry.py +0 -62
  988. openflo-2.7.0b2/tests/test_ellipse_tool.py +0 -392
  989. openflo-2.7.0b2/tests/test_ellipsoid_quadrant.py +0 -374
  990. openflo-2.7.0b2/tests/test_embedding_dialog.py +0 -104
  991. openflo-2.7.0b2/tests/test_engine_availability.py +0 -155
  992. openflo-2.7.0b2/tests/test_error_handling.py +0 -279
  993. openflo-2.7.0b2/tests/test_exit_autosave.py +0 -206
  994. openflo-2.7.0b2/tests/test_export_audit.py +0 -282
  995. openflo-2.7.0b2/tests/test_export_stamp.py +0 -288
  996. openflo-2.7.0b2/tests/test_expression_window.py +0 -285
  997. openflo-2.7.0b2/tests/test_fast_graph.py +0 -119
  998. openflo-2.7.0b2/tests/test_fcs_export.py +0 -118
  999. openflo-2.7.0b2/tests/test_fcs_speed.py +0 -226
  1000. openflo-2.7.0b2/tests/test_figure_layout.py +0 -127
  1001. openflo-2.7.0b2/tests/test_fixes.py +0 -713
  1002. openflo-2.7.0b2/tests/test_flowsom.py +0 -95
  1003. openflo-2.7.0b2/tests/test_flowsom_mst.py +0 -49
  1004. openflo-2.7.0b2/tests/test_flowsom_tree_window.py +0 -94
  1005. openflo-2.7.0b2/tests/test_fmo_gating_dialog.py +0 -308
  1006. openflo-2.7.0b2/tests/test_fmo_threshold_log.py +0 -35
  1007. openflo-2.7.0b2/tests/test_fold_change_direction.py +0 -92
  1008. openflo-2.7.0b2/tests/test_frequency_missing_populations.py +0 -260
  1009. openflo-2.7.0b2/tests/test_frequency_window.py +0 -353
  1010. openflo-2.7.0b2/tests/test_gate_coordinates.py +0 -320
  1011. openflo-2.7.0b2/tests/test_gate_editor_helpers.py +0 -593
  1012. openflo-2.7.0b2/tests/test_gate_impl_parity.py +0 -112
  1013. openflo-2.7.0b2/tests/test_gate_mask_cache.py +0 -111
  1014. openflo-2.7.0b2/tests/test_gate_nonfinite_events.py +0 -107
  1015. openflo-2.7.0b2/tests/test_gate_on_missing_channel.py +0 -87
  1016. openflo-2.7.0b2/tests/test_gate_partition.py +0 -126
  1017. openflo-2.7.0b2/tests/test_gate_refit.py +0 -282
  1018. openflo-2.7.0b2/tests/test_gate_replication_parent.py +0 -165
  1019. openflo-2.7.0b2/tests/test_gate_text_intensity.py +0 -472
  1020. openflo-2.7.0b2/tests/test_gate_tools_axis_ends.py +0 -308
  1021. openflo-2.7.0b2/tests/test_gatetree.py +0 -92
  1022. openflo-2.7.0b2/tests/test_gating.py +0 -254
  1023. openflo-2.7.0b2/tests/test_gating_helpers.py +0 -159
  1024. openflo-2.7.0b2/tests/test_gatingml.py +0 -367
  1025. openflo-2.7.0b2/tests/test_gmm_ellipse_info.py +0 -62
  1026. openflo-2.7.0b2/tests/test_golden_regression.py +0 -127
  1027. openflo-2.7.0b2/tests/test_gpu_accel.py +0 -223
  1028. openflo-2.7.0b2/tests/test_group_stats.py +0 -203
  1029. openflo-2.7.0b2/tests/test_group_stats_window.py +0 -147
  1030. openflo-2.7.0b2/tests/test_grouping.py +0 -219
  1031. openflo-2.7.0b2/tests/test_gui_guard.py +0 -99
  1032. openflo-2.7.0b2/tests/test_gui_smoke.py +0 -1396
  1033. openflo-2.7.0b2/tests/test_gui_toolings.py +0 -86
  1034. openflo-2.7.0b2/tests/test_gui_ux.py +0 -864
  1035. openflo-2.7.0b2/tests/test_highlighted_sample_is_plotted.py +0 -105
  1036. openflo-2.7.0b2/tests/test_histogram_offscale_note.py +0 -97
  1037. openflo-2.7.0b2/tests/test_histogram_resolution.py +0 -115
  1038. openflo-2.7.0b2/tests/test_independent_reference.py +0 -112
  1039. openflo-2.7.0b2/tests/test_inspect_fcs.py +0 -61
  1040. openflo-2.7.0b2/tests/test_install_leaves_no_files.py +0 -126
  1041. openflo-2.7.0b2/tests/test_interop.py +0 -145
  1042. openflo-2.7.0b2/tests/test_interop_h5ad_roundtrip.py +0 -150
  1043. openflo-2.7.0b2/tests/test_knn_threads.py +0 -93
  1044. openflo-2.7.0b2/tests/test_label_align.py +0 -148
  1045. openflo-2.7.0b2/tests/test_layout_editor.py +0 -935
  1046. openflo-2.7.0b2/tests/test_layout_exploded.py +0 -463
  1047. openflo-2.7.0b2/tests/test_legend_placement.py +0 -138
  1048. openflo-2.7.0b2/tests/test_load_no_copy.py +0 -89
  1049. openflo-2.7.0b2/tests/test_load_reliability.py +0 -391
  1050. openflo-2.7.0b2/tests/test_loader_queue.py +0 -119
  1051. openflo-2.7.0b2/tests/test_no_plausible_fills.py +0 -246
  1052. openflo-2.7.0b2/tests/test_normalise_groups_idempotent.py +0 -82
  1053. openflo-2.7.0b2/tests/test_not_gate_excludes_unmeasured.py +0 -117
  1054. openflo-2.7.0b2/tests/test_optimize_compensation_truth.py +0 -129
  1055. openflo-2.7.0b2/tests/test_panel_pairs.py +0 -145
  1056. openflo-2.7.0b2/tests/test_parallel_load_blas.py +0 -125
  1057. openflo-2.7.0b2/tests/test_parse_ellipsoid.py +0 -924
  1058. openflo-2.7.0b2/tests/test_phenograph_pool_cap.py +0 -82
  1059. openflo-2.7.0b2/tests/test_pipeline_missing_embedding.py +0 -97
  1060. openflo-2.7.0b2/tests/test_plot_cache.py +0 -710
  1061. openflo-2.7.0b2/tests/test_plot_defaults.py +0 -405
  1062. openflo-2.7.0b2/tests/test_polygon_gating.py +0 -239
  1063. openflo-2.7.0b2/tests/test_polygon_rect_tools.py +0 -446
  1064. openflo-2.7.0b2/tests/test_population_stats_geomean.py +0 -171
  1065. openflo-2.7.0b2/tests/test_population_stats_speed.py +0 -315
  1066. openflo-2.7.0b2/tests/test_preexisting_data_defects.py +0 -320
  1067. openflo-2.7.0b2/tests/test_prepare_unit_tags.py +0 -167
  1068. openflo-2.7.0b2/tests/test_preview_quadrants.py +0 -102
  1069. openflo-2.7.0b2/tests/test_propagate_downsample_order.py +0 -147
  1070. openflo-2.7.0b2/tests/test_python_readiness.py +0 -363
  1071. openflo-2.7.0b2/tests/test_qc.py +0 -225
  1072. openflo-2.7.0b2/tests/test_qc_ignores_derived_columns.py +0 -202
  1073. openflo-2.7.0b2/tests/test_qc_integer_channels.py +0 -198
  1074. openflo-2.7.0b2/tests/test_qc_removed_visible.py +0 -156
  1075. openflo-2.7.0b2/tests/test_quadrant_tool.py +0 -306
  1076. openflo-2.7.0b2/tests/test_quick_preview_dialog.py +0 -109
  1077. openflo-2.7.0b2/tests/test_raw_not_aliased.py +0 -104
  1078. openflo-2.7.0b2/tests/test_result_stamps.py +0 -355
  1079. openflo-2.7.0b2/tests/test_resume_and_transfer_hygiene.py +0 -153
  1080. openflo-2.7.0b2/tests/test_run_louvain.py +0 -98
  1081. openflo-2.7.0b2/tests/test_sample_qc_window_values.py +0 -109
  1082. openflo-2.7.0b2/tests/test_saved_stats.py +0 -454
  1083. openflo-2.7.0b2/tests/test_scales.py +0 -39
  1084. openflo-2.7.0b2/tests/test_seed_phrases.py +0 -102
  1085. openflo-2.7.0b2/tests/test_selftest.py +0 -45
  1086. openflo-2.7.0b2/tests/test_selftest_controls.py +0 -190
  1087. openflo-2.7.0b2/tests/test_session.py +0 -418
  1088. openflo-2.7.0b2/tests/test_session_compare.py +0 -147
  1089. openflo-2.7.0b2/tests/test_session_continuity.py +0 -150
  1090. openflo-2.7.0b2/tests/test_session_json_is_valid.py +0 -111
  1091. openflo-2.7.0b2/tests/test_session_lock.py +0 -386
  1092. openflo-2.7.0b2/tests/test_session_missing_samples.py +0 -155
  1093. openflo-2.7.0b2/tests/test_session_restore_transforms.py +0 -250
  1094. openflo-2.7.0b2/tests/test_session_safety.py +0 -400
  1095. openflo-2.7.0b2/tests/test_session_saves_label_populations.py +0 -82
  1096. openflo-2.7.0b2/tests/test_session_sidecar_collision.py +0 -196
  1097. openflo-2.7.0b2/tests/test_session_sidecar_cut_short.py +0 -118
  1098. openflo-2.7.0b2/tests/test_setup_installer.py +0 -299
  1099. openflo-2.7.0b2/tests/test_small_n_statistics.py +0 -464
  1100. openflo-2.7.0b2/tests/test_som_kernel.py +0 -190
  1101. openflo-2.7.0b2/tests/test_spectral.py +0 -302
  1102. openflo-2.7.0b2/tests/test_spectral_matched_af.py +0 -339
  1103. openflo-2.7.0b2/tests/test_spectral_measured_and_spread.py +0 -275
  1104. openflo-2.7.0b2/tests/test_spectral_qc_window.py +0 -190
  1105. openflo-2.7.0b2/tests/test_spectral_unmix_gui.py +0 -242
  1106. openflo-2.7.0b2/tests/test_spectral_unmix_nonfinite.py +0 -96
  1107. openflo-2.7.0b2/tests/test_stat_forms.py +0 -115
  1108. openflo-2.7.0b2/tests/test_statistics_export.py +0 -238
  1109. openflo-2.7.0b2/tests/test_statistics_population_column.py +0 -134
  1110. openflo-2.7.0b2/tests/test_stats.py +0 -712
  1111. openflo-2.7.0b2/tests/test_survivor_boundaries.py +0 -104
  1112. openflo-2.7.0b2/tests/test_synthetic.py +0 -238
  1113. openflo-2.7.0b2/tests/test_template_apply.py +0 -177
  1114. openflo-2.7.0b2/tests/test_template_one_sample.py +0 -66
  1115. openflo-2.7.0b2/tests/test_templates.py +0 -64
  1116. openflo-2.7.0b2/tests/test_theme.py +0 -91
  1117. openflo-2.7.0b2/tests/test_tk_main_thread.py +0 -296
  1118. openflo-2.7.0b2/tests/test_tk_threads.py +0 -89
  1119. openflo-2.7.0b2/tests/test_tool_window.py +0 -91
  1120. openflo-2.7.0b2/tests/test_trajectory.py +0 -137
  1121. openflo-2.7.0b2/tests/test_trajectory_window.py +0 -137
  1122. openflo-2.7.0b2/tests/test_transform_consistency.py +0 -71
  1123. openflo-2.7.0b2/tests/test_transform_record_writes.py +0 -140
  1124. openflo-2.7.0b2/tests/test_transform_records.py +0 -647
  1125. openflo-2.7.0b2/tests/test_transforms.py +0 -65
  1126. openflo-2.7.0b2/tests/test_typecheck_config.py +0 -168
  1127. openflo-2.7.0b2/tests/test_ui_audit_sign_verify.py +0 -136
  1128. openflo-2.7.0b2/tests/test_ui_compare.py +0 -36
  1129. openflo-2.7.0b2/tests/test_ui_dialogs_smoke.py +0 -184
  1130. openflo-2.7.0b2/tests/test_ui_inspect.py +0 -78
  1131. openflo-2.7.0b2/tests/test_ui_logic.py +0 -76
  1132. openflo-2.7.0b2/tests/test_ui_preview.py +0 -39
  1133. openflo-2.7.0b2/tests/test_ui_sample_qc.py +0 -97
  1134. openflo-2.7.0b2/tests/test_ui_synth.py +0 -50
  1135. openflo-2.7.0b2/tests/test_ui_voltage.py +0 -102
  1136. openflo-2.7.0b2/tests/test_undo_redo.py +0 -124
  1137. openflo-2.7.0b2/tests/test_unknown_population_is_missing.py +0 -144
  1138. openflo-2.7.0b2/tests/test_unknown_scale_export.py +0 -164
  1139. openflo-2.7.0b2/tests/test_unknown_scale_import_is_missing.py +0 -87
  1140. openflo-2.7.0b2/tests/test_update_flow.py +0 -256
  1141. openflo-2.7.0b2/tests/test_v261_sessions.py +0 -374
  1142. openflo-2.7.0b2/tests/test_view_tabs.py +0 -491
  1143. openflo-2.7.0b2/tests/test_voltage_cli.py +0 -103
  1144. openflo-2.7.0b2/tests/test_voltage_titration_dialog.py +0 -134
  1145. openflo-2.7.0b2/tests/test_window_fit.py +0 -460
  1146. openflo-2.7.0b2/tests/test_windows_nothing_hidden.py +0 -386
  1147. openflo-2.7.0b2/tests/test_workspace.py +0 -993
  1148. openflo-2.7.0b2/tests/test_workspace_batch_folder.py +0 -111
  1149. openflo-2.7.0b2/tests/test_workspace_job_driver.py +0 -194
  1150. openflo-2.7.0b2/tests/test_write_fcs.py +0 -53
  1151. openflo-2.7.0b2/tests/test_wsp_axis_floor.py +0 -444
  1152. openflo-2.7.0b2/tests/test_wsp_boolean_gates.py +0 -152
  1153. openflo-2.7.0b2/tests/test_wsp_export_golden.py +0 -122
  1154. openflo-2.7.0b2/tests/test_wsp_gate_units.py +0 -287
  1155. openflo-2.7.0b2/tests/test_wsp_groups_subfolders.py +0 -154
  1156. openflo-2.7.0b2/tests/test_wsp_population_names.py +0 -100
  1157. openflo-2.7.0b2/tests/test_wsp_quadrant_export.py +0 -110
  1158. openflo-2.7.0b2/tests/test_wsp_sample_compensation.py +0 -129
  1159. openflo-2.7.0b2/tests/test_wsp_sample_keywords.py +0 -91
  1160. openflo-2.7.0b2/tests/test_wsp_time_edge_check.py +0 -76
  1161. openflo-2.7.0b2/tests/test_wsp_time_units.py +0 -410
  1162. openflo-2.7.0b2/tests/test_wsp_unnamed_population_names.py +0 -89
  1163. openflo-2.7.0b2/tests/test_wsp_writer.py +0 -360
  1164. openflo-2.7.0b2/tests/tk_main_thread.py +0 -208
  1165. {openflo-2.7.0b2 → openflo-3.0.0}/LICENSE.txt +0 -0
  1166. {openflo-2.7.0b2 → openflo-3.0.0}/scripts/bench_import_cost.py +0 -0
  1167. {openflo-2.7.0b2 → openflo-3.0.0}/scripts/bench_snn_prune.py +0 -0
  1168. {openflo-2.7.0b2 → openflo-3.0.0}/scripts/lowest_direct_requirements.py +0 -0
  1169. {openflo-2.7.0b2 → openflo-3.0.0}/scripts/make_synthetic_dataset.py +0 -0
  1170. {openflo-2.7.0b2 → openflo-3.0.0}/scripts/smoke_test.py +0 -0
  1171. {openflo-2.7.0b2 → openflo-3.0.0}/scripts/typecheck.py +0 -0
  1172. {openflo-2.7.0b2 → openflo-3.0.0}/setup.cfg +0 -0
  1173. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/annotate.py +0 -0
  1174. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/compliance.py +0 -0
  1175. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/errors.py +0 -0
  1176. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/fmo_sensitivity.py +0 -0
  1177. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/group_test_choice.py +0 -0
  1178. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/guide/01-getting-started.md +0 -0
  1179. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/parallel.py +0 -0
  1180. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/pptx_export.py +0 -0
  1181. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/project_archive.py +0 -0
  1182. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/py.typed +0 -0
  1183. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/template_library/README.md +0 -0
  1184. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/template_library/cleanup_acquisition_qc.json +0 -0
  1185. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/template_library/cleanup_minimal.json +0 -0
  1186. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/template_library/cleanup_standard.json +0 -0
  1187. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/template_library/cleanup_strict.json +0 -0
  1188. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/template_library/example_panel.json +0 -0
  1189. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/trajectory.py +0 -0
  1190. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/ui_autogate.py +0 -0
  1191. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/ui_derived.py +0 -0
  1192. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo/ui_tips.py +0 -0
  1193. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo.egg-info/dependency_links.txt +0 -0
  1194. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo.egg-info/entry_points.txt +0 -0
  1195. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo.egg-info/requires.txt +0 -0
  1196. {openflo-2.7.0b2 → openflo-3.0.0}/src/openflo.egg-info/top_level.txt +0 -0
  1197. {openflo-2.7.0b2 → openflo-3.0.0}/templates/testtemplate.json +0 -0
  1198. {openflo-2.7.0b2 → openflo-3.0.0}/tests/__init__.py +0 -0
  1199. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_apply_region_gates.py +0 -0
  1200. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_async_task.py +0 -0
  1201. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_audit.py +0 -0
  1202. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_calibration.py +0 -0
  1203. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_capabilities.py +0 -0
  1204. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_cli_e2e.py +0 -0
  1205. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_comp_apply_to_batch.py +0 -0
  1206. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_comp_qc.py +0 -0
  1207. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_compare.py +0 -0
  1208. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_compare_conditions_absent_clusters.py +0 -0
  1209. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_compliance.py +0 -0
  1210. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_console_encoding.py +0 -0
  1211. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_degenerate_scatter_gating.py +0 -0
  1212. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_diagnostics_selftest_gate.py +0 -0
  1213. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_dialog_actions.py +0 -0
  1214. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_diffabundance.py +0 -0
  1215. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_dr_compare.py +0 -0
  1216. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_embedding.py +0 -0
  1217. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_embeddings.py +0 -0
  1218. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_exceptions.py +0 -0
  1219. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_fcs_stored_values.py +0 -0
  1220. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_fcs_writer_quirks.py +0 -0
  1221. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_fmo_histogram_overlay.py +0 -0
  1222. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_fmo_threshold_compare.py +0 -0
  1223. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_gate_import_fixes.py +0 -0
  1224. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_gate_table.py +0 -0
  1225. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_gating_strategy_page.py +0 -0
  1226. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_group_summary_sd.py +0 -0
  1227. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_import_is_lazy.py +0 -0
  1228. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_leiden.py +0 -0
  1229. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_log_histogram_range.py +0 -0
  1230. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_log_transform_nonpositive.py +0 -0
  1231. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_mem_tiny_clusters.py +0 -0
  1232. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_paths.py +0 -0
  1233. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_plot_density.py +0 -0
  1234. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_plotmath.py +0 -0
  1235. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_pptx_export.py +0 -0
  1236. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_preferences_wiring.py +0 -0
  1237. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_prefs.py +0 -0
  1238. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_provenance.py +0 -0
  1239. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_provenance_unreadable_entries.py +0 -0
  1240. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_public_links.py +0 -0
  1241. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_quality_correctness.py +0 -0
  1242. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_report.py +0 -0
  1243. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_scales_logicle.py +0 -0
  1244. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_selftest_cli.py +0 -0
  1245. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_session_migration_machinery.py +0 -0
  1246. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_spectral_degenerate_reference.py +0 -0
  1247. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_stats_multigroup.py +0 -0
  1248. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_threshold_inclusive.py +0 -0
  1249. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_trajectory_nonfinite.py +0 -0
  1250. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_transform_nonfinite.py +0 -0
  1251. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_tree_ids.py +0 -0
  1252. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_umap_embedding.py +0 -0
  1253. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_update.py +0 -0
  1254. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_voltage.py +0 -0
  1255. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_warmup.py +0 -0
  1256. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_workspace_prep_offthread.py +0 -0
  1257. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_wsp_extract.py +0 -0
  1258. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_wsp_partial_matrix.py +0 -0
  1259. {openflo-2.7.0b2 → openflo-3.0.0}/tests/test_wsp_writer_structure.py +0 -0
@@ -0,0 +1,4503 @@
1
+ # Changelog
2
+
3
+ All notable changes to OpenFlo are documented in this file.
4
+ The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/);
5
+ this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
6
+
7
+ ## [Unreleased]
8
+
9
+ ## [3.0.0] - 2026-11-04
10
+
11
+ Still a public beta. A new major version because some default results,
12
+ menus and Python API names changed since 2.8.0: figure exports are
13
+ transparent by default, rCV and rSD default to the median-absolute-deviation
14
+ form (FlowJo's form stays in Preferences > Statistics), each tray tab keeps
15
+ its own gates, Frequencies is a tab of Statistics, the Tools, File and Edit
16
+ menus and the right-click menus are grouped into submenus, the Python API
17
+ pairs channels across samples as the command line does, `write_fcs` takes
18
+ `(df, path)`, and several names are deprecated (they still work, with a
19
+ warning). It also carries the fixes from a sweep of every button and input.
20
+ Sessions saved by 2.x still open. Everything since 2.8.0 is listed below.
21
+
22
+ ### Added
23
+ - **Remove from tab, beside Remove from session.** On a tab other than All,
24
+ the Samples & Gates tray had only one remove action you could see: the
25
+ Remove button and the sample's right-click *Remove sample*, which unload
26
+ the sample from every tab. (Remove from this tab sat lower in the menu,
27
+ among the tab entries.) Now a *Remove from tab* button sits next to
28
+ *Remove* (greyed out on the All tab). The sample right-click menu lists
29
+ *Remove from this tab* right above *Remove from session (all tabs)*, and
30
+ the trial menu *Remove trial from this tab* above *Remove trial from
31
+ session (samples + gates)…*. Removing from a tab keeps the sample loaded,
32
+ with its gates, in the All tab and any other tab that has it, and the
33
+ status bar says what the tree no longer shows (a gate stays when the tab
34
+ shows all of its sample's gates, or a listed gate above or below it).
35
+ Both buttons have hover tips, and the right-click entries show their help
36
+ in the status bar; a picked entry's result is no longer replaced by the
37
+ old status text when the menu closes after the command runs.
38
+ - **A sample removed from the session leaves every tab.** Before, the tabs
39
+ kept its name, so a file loaded again under that name reappeared in the
40
+ tabs the old one had been in.
41
+ - **Faster Gaussian KDE density, as a choice (Preferences > Performance, off
42
+ by default).** The editor's "True Gaussian KDE (slow)" plot option takes
43
+ seconds per redraw, and Quick preview evaluates the exact KDE of every event
44
+ against every event, which can take minutes on a large file. With the new
45
+ box ticked, both compute the same kernel and bandwidth on a fine grid
46
+ instead (`openflo.density.binned_kde_eval`): milliseconds, with densities
47
+ within 0.5% of the exact peak and colours within half a colour-map step.
48
+ Off, every density is exactly as before. The standard density plot was
49
+ already binned and does not change. Library calls take it as
50
+ `kde_density(..., binned=True)` and `preview.density_scatter(...,
51
+ binned=True)`.
52
+ - **Clear message on an unsupported Python.** OpenFlo supports Python 3.11
53
+ to 3.14, and the README now says so. FlowUtils has no Python 3.15 download
54
+ yet, so on 3.15 pip tried to compile it and stopped with a compiler error
55
+ on machines without C build tools. `setup.sh` and `setup.bat` now skip a
56
+ Python 3.15 or newer with a message saying why, try the supported versions
57
+ by name (`python3.14` ... `python3.11`, `py -3.14` ... `py -3.11`), and
58
+ otherwise fetch Python 3.12 through uv as before. An existing `.venv` made
59
+ with an unsupported Python stops setup with a message to delete it, instead
60
+ of a pip failure. The `py -3.x` probes cannot make the Python install
61
+ manager install a missing version. `requires-python` stays uncapped.
62
+ - **Hover tips on every control.** Many controls had none: the layout
63
+ editor's item and colours dialogs, the compensation optimizer, the log pane,
64
+ the histogram sliders, the first-run buttons, and the dialogs built inside
65
+ the editor (gate coordinates, auto-clean settings, channel transforms, copy
66
+ gates, boolean and combination gates, template apply, annotate, cell cycle,
67
+ cluster, close and clear-all choices) and the Pipeline Workspace's Group
68
+ params, Markers and results viewer. Each says what the control does and what
69
+ changes. The tooltip tests now open every window class and every inline
70
+ dialog, and fail on any actionable control without a tip.
71
+ - **Westgard R-4s within a run.** Tools > Bead tracking can log two or more
72
+ bead levels per run (pick one population per bead peak, say dim and
73
+ bright); each level is its own column, `<channel> @ <level>`, with its own
74
+ baseline. R-4s now rejects a run when one level is beyond +2 SD and another
75
+ beyond -2 SD in that run, Westgard's published form, and every level logged
76
+ in the run shows it. The check across two runs in a row stays on by
77
+ default, so a one-level log keeps a range check; Preferences > Statistics,
78
+ "Westgard R-4s: literature form only (within run)", turns it off. The
79
+ window says which form judged the chart. One-level logs load and judge as
80
+ before.
81
+ - **`openflo.channel_names.is_time_channel(name)`** answers "is this the
82
+ acquisition-time channel?" in one place. Bead normalization, derived
83
+ kinetics, saturation, debarcoding and channel classification all ask it;
84
+ before, they used pipeline's private regex or their own copy of it. The
85
+ answer is unchanged for every name.
86
+ - **Fast graph in the Cluster dialog and on the command line.** The
87
+ Pipeline Workspace's opt-in "Fast graph" box (Leiden only) is now also in
88
+ Analyze > Cluster, and `openflo-run --fast-graph` does the same with
89
+ `--cluster-method leiden`. It links only each cell's actual nearest
90
+ neighbours, as PhenoGraph and Seurat do: clustering runs about 6-9x
91
+ faster but the clusters differ, so it stays off by default and every box
92
+ carries the same hover tip saying so.
93
+ - **Nine more lists scroll with the mouse wheel.** Annotate populations,
94
+ annotate clusters, combination gates, copy gates, boolean gate, apply
95
+ template, FMO gating, clustering markers and run results now use the
96
+ shared scrolling panel: the wheel scrolls the list, the window opens tall
97
+ enough for every row when the screen allows, and a long list scrolls
98
+ instead of growing past the screen. Closing any window with a scrolling
99
+ list (these nine, Preferences and the others that already had one) with
100
+ Escape or Enter while the pointer is over the list no longer leaves the
101
+ wheel bound to it (the next wheel turn raised an error).
102
+
103
+ ### Changed
104
+ - **Figure exports are transparent by default.** PNG, SVG and PDF exports
105
+ start on a transparent background everywhere: every Background box, the
106
+ main plot image, Copy figure, the layout editor (whose batch PDF now also
107
+ clears the plot faces), the volcano plot, the HTML report, workspace runs
108
+ and the command-line plots. White, Dark and Translucent are still offered.
109
+ A JPEG cannot be transparent, so it is saved on white (dark for a dark
110
+ plot). The Windows clipboard bitmap is laid on white; programs that paste
111
+ the PNG keep the transparency. A transparent export keeps the plot's own
112
+ colours, so a dark-theme plot has light labels (pick Dark for a solid dark
113
+ background). The layout editor's batch PDF now honours Translucent like
114
+ its other formats.
115
+ - **Frequencies is a tab of Statistics.** The two windows showed the same
116
+ Count / %Parent / %Total numbers, one as a table and one with the samples
117
+ split into groups and compared. Statistics now has a Table tab and a
118
+ Frequencies by group tab (built the first time it is shown), the side
119
+ panel has one Statistics button, and Analyze > Frequencies… opens
120
+ Statistics on the Frequencies by group tab. The comparison, its tests
121
+ and its exports are unchanged; its controls are on three short rows so
122
+ they fit the Statistics window, and the tab has its own Close button. A
123
+ gate dropped on Statistics shows the Table tab, where it is added.
124
+ `ui_frequency.FrequencyComparisonWindow` is now
125
+ `ui_frequency.FrequencyPanel(master, editor)`.
126
+ - **Shorter sample and gate right-click menus, nothing taken away.** The
127
+ sample menu had 21 entries in a flat list; its top level is now *Copy
128
+ sample (Ctrl+C)*, *Paste (Ctrl+V)*, *Copy for… ▸*, *Move to another
129
+ window*, *New gate ▸*, *Sample info ▸* and *Remove ▸*. *Copy sample* copies
130
+ the sample with its gates: *Paste* loads it in another OpenFlo window, or
131
+ puts it (with its gates) on the open tab of this one. *Copy for… ▸* holds
132
+ the other ways to copy: to another window with or without gates (file path
133
+ only), to another tab, the gates only to other samples, and the
134
+ samples' on/off state to other tabs. The color, note and set-aside entries are under
135
+ *Sample info ▸*; *Remove ▸* holds Clear gates (keep sample), From this tab
136
+ and From session (all tabs). The gate menu's top level is Copy, Cut,
137
+ Paste, *Edit ▸* (coordinates, re-place on each sample, color, auto-clean
138
+ parameters), *New gate ▸*, Backgate, Export population as FCS, *Tabs ▸*
139
+ and Delete gate. Every entry, in every submenu, shows its help in the
140
+ status bar while hovered.
141
+ - **Ctrl+C on a sample row copies the sample with its gates.** Before, it
142
+ copied only the file path, though the menu called it "Copy gates". The
143
+ file path is still one click away: *Copy for… ▸ Another window, without
144
+ gates (file path)*, which also copies every selected sample's path, not
145
+ only the first. *Paste sample(s) from another instance* is folded into
146
+ *Paste*, which takes whatever was copied last (gates, samples or FCS
147
+ paths); the empty-space menu still lists it, as *Paste sample(s) from
148
+ another window*. The Pipeline workspace's right-click menu uses the same
149
+ words: *Copy sample (with gates)* and *Move to another window*.
150
+ - **Tools, File and Edit menus in submenus.** Tools had 30 entries in one
151
+ list; it now has six submenus: *Compensation & spectral*, *Channels*,
152
+ *Normalize & clean* (with *Mass cytometry* inside), *Figures*, *Panel &
153
+ instrument QC* and *Files & checks*, then *Templates* and *Save
154
+ template…*. The seven exports of File are under *File > Export* (FlowJo
155
+ workspace, Gating-ML, gate list, plot image, populations, populations of
156
+ all samples, HTML report); *Copy plot to clipboard* stays in File. The gate
157
+ tools of Edit (singlet, FMO, auto-clean, boolean, combination, coordinates,
158
+ re-place) are under *Edit > Gates*. Keyboard shortcuts are unchanged, the
159
+ user guide and the messages that name a menu give the new paths, and every
160
+ submenu shows its entries' help in the status bar.
161
+ - **The layout editor stays responsive while a page draws.** Each item's
162
+ pixels are now made on a background thread, one item at a time, instead
163
+ of on the window thread, and the on-screen preview no longer compresses
164
+ each item to PNG first (that step cost as much as the drawing). An item
165
+ being redrawn keeps showing its last image, or "drawing…" when it is new,
166
+ until its new image arrives. Building each plot still runs on the window
167
+ thread, because it borrows the main window's plot settings for a moment.
168
+ One item draws in the background at a time, and matplotlib's mathtext
169
+ parser (tick labels such as 10^3), which is not safe to use from two
170
+ threads at once, now parses one label at a time. Exported files (PDF, PNG, SVG, TIFF, PowerPoint) are unchanged, byte for
171
+ byte.
172
+ - **Seventeen error handlers catch only the error they were written for.**
173
+ Seven that guard against a Tk window or widget that is already gone now
174
+ catch only Tk's own error, and ten around one file read or write catch only
175
+ file errors (plus a malformed file, where the contents are parsed): the
176
+ progress strip, the editor's drain timer, Escape-to-close, the tab
177
+ selection, the theme recolouring, the cross-window move marker, the error
178
+ report keymap, the windowed launch's log file, preferences, session locks,
179
+ sessions, labels files and compliance records. A missing, unreadable or
180
+ malformed file and a closed window are handled exactly as before; any other
181
+ error is a bug and is now reported instead of silently skipped.
182
+ - **Parallel workers start fresh on Linux too.** The command line's
183
+ per-sample workers (`--workers`) and `openflo-compare`'s workers were forked
184
+ from the running process on Linux, copying it with whatever locks its
185
+ numba, BLAS or plotting threads held; Python 3.12 warns that such a child
186
+ can hang, and Python 3.14 stops forking by default. They now start as fresh
187
+ processes, as they already did on Windows and macOS and as the group UMAP
188
+ pool did. The output files are unchanged. Each worker now also takes the
189
+ run's log level and format (`-v`, `-q`), UTF-8 console output, the chosen
190
+ `--palette` and the headless plotting backend from the main process; on
191
+ Windows and macOS the workers' log lines and palette were lost before.
192
+ A script that calls `openflo.cli.run(..., workers=2)` (or more) must now
193
+ put that call under `if __name__ == '__main__':` on Linux too, as Windows
194
+ and macOS already required; without it each worker re-runs the script and
195
+ the samples fail.
196
+ - **PhenoGraph's progress lines go to the log.** Every PhenoGraph clustering,
197
+ joint clustering included, printed seven unlabelled lines ("Finding 30
198
+ nearest neighbors ...", "PhenoGraph completed in 0.47 seconds") to stdout,
199
+ which `--quiet` could not silence. They are now DEBUG records on the
200
+ `openflo.cluster_graph` logger: hidden by default, shown with `-vv`.
201
+ Anything else phenograph prints (its Louvain failure diagnostics, such as
202
+ "stderr from convert: ...") is logged as a WARNING, so it is still shown.
203
+ Prints from other threads, and phenograph called outside OpenFlo, are
204
+ untouched.
205
+ - **rCV and rSD default to the robust-statistics form.** The default robust
206
+ SD is now the median absolute deviation times 1.4826022 (the literature
207
+ standard, which tolerates up to half the events being outliers) instead of
208
+ FlowJo's half 15.87-84.13 percentile range, so rCV and rSD values change on
209
+ skewed populations (a log-normal with sigma 0.8 reads 74% instead of 89%).
210
+ To match FlowJo, pick its form under Preferences > Statistics; scripts can
211
+ pass `stat_forms=gating.FLOWJO_STAT_FORMS`. GeoMean keeps FlowJo's form,
212
+ the more accurate one on compensated data.
213
+ - **The Sample QC map no longer depends on the scikit-learn release.**
214
+ scikit-learn 1.9 changed MDS defaults (one start instead of four) and 1.10
215
+ changes its starting layout, so with 1.9 installed the map already came out
216
+ differently. OpenFlo now spells out scikit-learn 1.8's settings, so the map
217
+ is the same on 1.7 to 1.9 as on the pinned 1.8 (unchanged there), and the
218
+ three FutureWarnings are gone.
219
+ - **The Workspace names gates as Statistics does.** A Workspace item's gate
220
+ path now shows each gate's label or name, as the Statistics and
221
+ Frequencies tables do, and shows a gate's bounds only when it has neither.
222
+ Output file names (built from the gate id) are unchanged, and a recipe
223
+ saved earlier keeps the label it was saved with.
224
+ - **Two library warnings are logged, not printed.** Shared clustering's note
225
+ on how channels were matched across samples, and the comparison's
226
+ "[compensation skipped]" line, now go through Python logging with the same
227
+ wording: on stderr, with the CLI's time and level prefix, and still shown
228
+ with `-q`. The app's log pane shows them as before. Callers of the library
229
+ can now silence or capture them with `logging`.
230
+ - **More time checks recognise any time-channel name.** The auto-clean
231
+ viability-channel menu, the FlowJo export's "nothing lies below 0 s" check
232
+ on a lower gate edge, and FCS $PnE/$PnG decoding now ask
233
+ `is_time_channel`, so a time channel named like `Time (s)`, `HDR-T Time`
234
+ or `TIMESTAMP` is treated the same as one named `Time` (left out of the
235
+ menu, its 0 s edge left open, never log-decoded). $TIMESTEP scaling and
236
+ the FlowJo axis rule still match only a channel named `Time`, since
237
+ FlowJo's behaviour was measured on that name only. Files whose time
238
+ channel is named `Time` see no change.
239
+ - **A workspace run writes its per-event table as a binary file.** Each run
240
+ unit's `<label>_events.csv` (markers, cluster, source tags, embedding
241
+ coordinates) is now `<label>_events.npz`, the session sidecar format:
242
+ about 0.2 s a sample where the CSV took about 10 s, with every value kept
243
+ exactly. **Results… → Load in editor** and **File → Load processed data**
244
+ read it, with its transform record. Tick **Also write events as CSV** in
245
+ the run bar (off by default; a workspace or recipe saved before it reads
246
+ as off) to get the CSV beside it as well, for Excel. A table holding a
247
+ column the binary file can't carry is written as the CSV alone, as before.
248
+ The `_events.csv` of earlier runs still loads as it did.
249
+ - **The full CI matrix runs on the public mirror.** Tests with the GUI on
250
+ Linux, Windows and macOS, the type check and the install checks now run on
251
+ openflo-public, where runners are free: push a scrubbed export of your
252
+ committed tree with `python scripts/publish_public.py --ci-branch <name>
253
+ --push` and check the result with `python scripts/publish_public.py
254
+ --ci-status`. CI on the private repository runs lint and the privacy gate
255
+ only; "Run workflow" still runs the full matrix there by hand. A release is
256
+ built only after a green mirror run for the same tree, instead of running
257
+ the matrix again on the tag.
258
+ - **`write_fcs` takes `(df, path)` everywhere.** `openflo.write_fcs` (the
259
+ same as `openflo.pipeline.write_fcs`) took `(path, df)` while
260
+ `openflo.fcs_export.write_fcs` took `(df, path)`. Both now take
261
+ `write_fcs(df, path)` and share one writer; `fcs_export.write_fcs` gains a
262
+ `channels=` keyword that selects and orders the columns written. The old
263
+ `write_fcs(path, df)` order still works on both, with a
264
+ DeprecationWarning; any other pair of arguments raises TypeError. Through
265
+ the shared writer, `openflo.write_fcs` now leaves `$PnS` blank where a
266
+ label only repeats the channel name, and logs a warning when it zeroes
267
+ non-finite cells.
268
+ - **Group colours no longer load Tk.** The group palettes (Okabe-Ito and
269
+ classic tab10) and their colour logic moved from `theme` to the new
270
+ Tk-free `openflo.group_palettes` module (`palette_colors(name)`,
271
+ `n_group_colors(n, name)`), so `pipeline.categorical_cmap` and the plots
272
+ that colour by group work without tkinter. `theme.group_colors`,
273
+ `theme.categorical_colors` and `theme.OKABE_ITO` / `theme.TAB10` still
274
+ exist and still follow the View → Group colors preference. No colour,
275
+ behaviour or number changes.
276
+ - **The Pipeline Workspace's job child loads no Tk and reads no
277
+ preferences.** The panel runs each unit as `python -m
278
+ openflo.workspace_run` (the Tk-free run module) instead of `python -m
279
+ openflo.workspace`, which still works for scripts that call it. The panel
280
+ reads the View → Group colors choice and passes it to the child as
281
+ `cfg['group_palette']`, so the by-group / by-sample embedding plots draw
282
+ the same colours as before. The only difference: `workspace_run.compute_run`
283
+ called directly from Python without `group_palette` in its config colours
284
+ those plots with the default Okabe-Ito palette, where it followed the
285
+ preference before (it differs only if you had chosen Classic). The key is
286
+ not stored in workspace files or recipes.
287
+
288
+ ### Deprecated
289
+ - **Names that said the same thing three ways.** `spectral_af.af_signatures`
290
+ takes `max_events=` like the rest of OpenFlo; `n_max=` still works, with a
291
+ DeprecationWarning. The editor mixins `editor_combos.CombinationsMixin`
292
+ and `editor_console.ConsoleLogMixin` are `CombosMixin` and `ConsoleMixin`,
293
+ named for their modules; the old names still work, with a
294
+ DeprecationWarning. Window figure exports are all `_export_figure` and
295
+ table exports `_export_csv`; a new `vague-name` check (BP11) stops
296
+ `_export_png`, `_num` and similar names coming back. The editor's
297
+ plot-cache and auto-clean state drops its `_pc_` / `_ac_` prefixes for
298
+ `_plotcache_` / `_autoclean_`, and the check flags those prefixes too.
299
+ Passing both `max_events=` and `n_max=` to `af_signatures` raises
300
+ TypeError instead of `n_max` quietly winning. No behaviour or numbers
301
+ change otherwise.
302
+ - **Unused public names.** `pipeline.wsp_dependent_name`,
303
+ `pipeline.SCATTER_KEYWORDS`, `gatingml.is_gatingml`, `scales.scale_label`,
304
+ `theme.PLOT_DARK_THEMES`, `fixes.KINDS`, `fixes.MODES` and
305
+ `ui_mass_cytometry.BEADS_LABEL` are used nowhere in OpenFlo. They still
306
+ work, with a DeprecationWarning, and will be removed in a future release.
307
+ - **Private helpers other modules use have public names.** The pop-up
308
+ figure theme helpers are now `theme.dialog_dark_on` and
309
+ `theme.theme_figure_dark`, with `theme.match_dialog(fig, widget)` doing
310
+ both; `ui_figure_window.FigureWindow`, `editor_session.frame_digest` and
311
+ the .wsp helpers compare.py shares (`pipeline.WSP_BOOLEAN_OPS`,
312
+ `scale_time_coords`, `wsp_axis_floor_lo`, `wsp_axis_floors`,
313
+ `wsp_comp_channels`, `wsp_ellipse_gate`, `wsp_polygon_edge_axes`) drop
314
+ their leading underscore. The old names still work, with a
315
+ DeprecationWarning. No analysis numbers change.
316
+ - **Shared clustering out of the CLI.** The clustering every sample of a
317
+ comparison shares, and the channel matching by antibody label it uses,
318
+ are in the new Tk-free module `openflo.shared_clustering`
319
+ (`cluster_settings`, `cluster_by_method`, `shared_cluster_labels`,
320
+ `shared_pool_cap`, `equal_quota`, `align_shared_channels`, `fold_label`,
321
+ `fluor_tokens`, `label_key`), so `FlowExperiment.cluster_jointly` and
322
+ `align_fluor_labels` no longer import `openflo.cli` (and with it
323
+ matplotlib.pyplot and its thread caps). The old `cli` names still work,
324
+ with a DeprecationWarning. No behaviour or numbers change.
325
+ - **openflo-run's gate reading out of cli.py.** Parsing `--gates`, reading
326
+ a `--gates-from` file and picking `--gate-population` are in the new
327
+ module `openflo.cli_gates` (`parse_gates_arg`, `read_gate_file`,
328
+ `session_gates`, `population_gates`, `gate_name`,
329
+ `pipeline_gate_transforms`, `pipeline_transform_specs`,
330
+ `data_units_to_stored`), the first step of splitting cli.py. The old
331
+ `cli` names still work, with a DeprecationWarning. No behaviour or
332
+ numbers change.
333
+ - **openflo-run's batch unmixing out of cli.py.** `--unmix` runs from the
334
+ new module `openflo.cli_unmix` (`run_batch_unmix`, `load_unmix_controls`,
335
+ `write_unmix_qc`). `cli.run_batch_unmix` still works as before; the old
336
+ private helper names work with a DeprecationWarning. No behaviour or
337
+ numbers change.
338
+ - **openflo-run's sample lookup and .wsp export out of cli.py.** Normalising
339
+ the groups spec and finding each sample's FCS file are in the new module
340
+ `openflo.cli_samples` (`normalise_groups`, `fcs_path`, `scan_for_token`,
341
+ `find_unstained`); `--export-wsp` runs from the new module
342
+ `openflo.cli_wsp` (`export_pipeline_workspace`, `build_export_gate_list`,
343
+ `own_spillover`). `cli.fcs_path` still works as before; the old private
344
+ names work with a DeprecationWarning. No behaviour or numbers change.
345
+ - **openflo-run's run manifest out of cli.py.** Writing
346
+ `run_manifest.json` (each input and output file with its size and
347
+ SHA-256) is in the new module `openflo.cli_manifest`
348
+ (`write_run_manifest`, `run_input_files`, `file_entry`, `sha256`,
349
+ `MANIFEST_PACKAGES`). `cli.MANIFEST_NAME` still works as before; the old
350
+ private names work with a DeprecationWarning. No behaviour or numbers
351
+ change.
352
+ - **`safe_filename` is three names now.** Three functions shared the name
353
+ and turned a name into a file name by different rules. Each keeps its
354
+ rules under a name that says what it is for: `cli.output_name` (CLI
355
+ output folders and files), `fcs_export.population_file_stem` (exported
356
+ .fcs files) and `layout_page.layout_file_part` (exported layouts). The
357
+ old `fcs_export.safe_filename`, `layout_page.safe_filename` and
358
+ `cli._safe_filename` still work, with a DeprecationWarning. No file names
359
+ change.
360
+ - **`ui_logic.should_use_dark` is now `plot_is_dark`.** The old name still
361
+ works, with a DeprecationWarning. Theme keys are named in `ui_logic`
362
+ (`THEME_LIGHT`, `THEME_DARK_WINDOWS`, `THEME_ALL_DARK`, `THEME_DARK_PLOT`),
363
+ with `DARK_PLOT_THEMES` and `DARK_WINDOW_THEMES`; saved keys are unchanged.
364
+ - **Re-exports kept only for tests.** `gui.savefig_background` (use
365
+ `theme.savefig_background`) and `theme.EXPORT_DPI_CHOICES` (use
366
+ `layout_page.EXPORT_DPI_CHOICES`) still work, with a DeprecationWarning.
367
+ The private `gui._LOAD_POOL_SIZE` re-export is gone; it lives in
368
+ `editor_loadpool`.
369
+
370
+ ### Fixed
371
+ - **macOS: a window with a scrolling list no longer hangs on opening.**
372
+ On the Mac, Tk passes every resize on to the list inside, so a list that
373
+ gave up height to keep its window's buttons visible asked for it back and
374
+ the window resized for ever. The reduced height now stays put.
375
+ - **The side panel's tray buttons wrap onto two rows when one row would be
376
+ too wide.** Adding Remove from tab made the row about 570 px on macOS
377
+ (and wider than the panel in the app's own theme on Linux), squeezing the
378
+ plot. They are laid out again when the theme changes.
379
+ - **Revealing the Pipeline Workspace gives it its full width.** The divider
380
+ was placed without its own width, so the workspace opened a few pixels
381
+ narrower than its controls needed.
382
+ - **macOS: the keyboard's delete key deletes.** It sends BackSpace on the
383
+ Mac, which the selected gate (on the plot or in the gate list), a layout
384
+ item and a pipeline item now answer to.
385
+ - **Typing in a layout editor box no longer deletes the selected item.** The
386
+ delete key removed the selected page item even while you were typing in
387
+ one of the editor's boxes, on every platform.
388
+ - **Building from source asks for setuptools 77 or newer.** The build said
389
+ 68 was enough, but setuptools older than 77 cannot read the project's
390
+ license field, so a build with one of those stopped with an error.
391
+ - **Windows: running samples in parallel from a script no longer stops on
392
+ a character.** The parallel runner's console lines used "≤", "→" and
393
+ similar symbols, which a Windows console or redirected output (cp1252)
394
+ cannot print, so a script calling it directly stopped before the first
395
+ sample. They are plain text now. `openflo-run` was not affected.
396
+ - **Max points is per sample in every mode.** Pseudocolor capped the total,
397
+ so each of three samples drew about a third of what its row said. With
398
+ Downsample off, pseudocolor now draws every event of every sample, which
399
+ is slower with many large samples (10 x 200,000 events: about 8 s).
400
+ - **The Color box is greyed out where it does nothing** (pseudocolor,
401
+ contour, histogram), with a tip saying why.
402
+ - **Zoom follows the axis scale.** +/−, Ctrl+/−, the View menu and the
403
+ mouse wheel zoomed in linear steps on a log axis. Zooming out stops at
404
+ four times the plot as drawn (log, symlog and asinh limits overflowed
405
+ after about 20 wheel steps).
406
+ - **Delete and Ctrl+C/X/V work after a click on the plot**, not only when
407
+ the gate list has focus. Any other widget keeps the keys (a focused
408
+ button no longer deletes the selected gate), and Ctrl+Z in a spin box no
409
+ longer undoes a gate change.
410
+ - **Find says when nothing matches.** A right-click still ticks its row's
411
+ sample on (the active sample is always on the plot), and the menu's
412
+ "On/off state, to other tabs" and "Copy for… > Another tab" now use the
413
+ on/off state from before the click, so they can push "off". Copy gates with
414
+ nothing ticked stays open and says so; the Cut and Copy gates tips and
415
+ status lines say what really happens (Cut removes at once, Ctrl+Z puts it
416
+ back; Copy gates updates a target's same-name gate). Remove with nothing
417
+ selected says so.
418
+ - **Cell cycle says when a channel has no usable G1 DNA peak** (a peak
419
+ wider than CV 15 %, such as a scatter channel's bell) instead of
420
+ reporting G1 100 %; DNA-aneuploid histograms are still fitted. Its DNA
421
+ channel list leaves out cluster, embedding, phase and Time columns.
422
+ Proliferation starts on the selected gate. The Compare embeddings Run tip
423
+ says it shows a figure. The Group comparison result box is read-only
424
+ with a hint.
425
+ - **Statistics keeps every button visible at its minimum size.**
426
+ - **Layout editor:** a Resolution change or page rename is its own undo
427
+ step; Edit…, Duplicate and Delete are greyed out with nothing selected;
428
+ channel statistics with no Channel give a warning.
429
+ - **Background "Dark" exports a dark figure** in the light theme too, in
430
+ every window with that box, which now has a tip. Quick figure grid names
431
+ the Channel pairs it can't find instead of swapping in the current axes,
432
+ and the box has a tip. In a pair, an "x" between two channel names
433
+ (FSC-AxSSC-A) still works, but no longer splits a name or label that
434
+ holds an x (FLX-A).
435
+ - **Help:** Online documentation says when no browser opened, with the
436
+ address; a second F1 raises the open guide.
437
+ - **A `nan` or `inf` compensation cell no longer wipes the data.** The
438
+ Compensation window refused only words; `nan` and `inf` were applied and
439
+ turned every fluorescence value into NaN, with no way back short of
440
+ reloading the file. The window now refuses them, naming the cell by its
441
+ channels, and the library refuses a non-finite matrix too. The window
442
+ reports what Apply really did instead of always saying "Applied.". An FCS
443
+ whose `$SPILL` holds such a cell (or a session matrix that does) loads
444
+ uncompensated with a load note, like any other malformed matrix, and the
445
+ `--export-wsp` workspace carries no matrix for it (it wrote the `nan`).
446
+ - **Remove sample from the session asks first.** It named no sample and
447
+ could not be undone; it now names the sample(s) and their gate count and
448
+ points to Remove ▸ From this tab (No is the default answer). The Delete
449
+ key on a trial row now clears the trial's gates (like Edit > Delete gate)
450
+ instead of removing its samples.
451
+ - **Reopening a session highlights the saved sample.** The first sample to
452
+ finish loading could stay the active one and be ticked again, so the next
453
+ gate went to the wrong sample.
454
+ - **Behaviour change: FlowJo .wsp export writes real event counts** (the
455
+ file's event total on each sample, the Statistics count on each
456
+ population) instead of `count="0"`. A population whose count can't be
457
+ computed gets no count attribute, and neither does any population of a
458
+ sample whose loaded events are only part of its file (QC, downsampling),
459
+ since FlowJo counts the whole file.
460
+ - **Plate view reads NA** for a population that cannot be computed, in the
461
+ view and its CSV, instead of 0.
462
+ - **The Cluster dialog checks "Events / sample: Custom…"** (a whole number of
463
+ at least 100); 0 or a negative number used to cluster one event.
464
+ - **A phrase, a negative number or a large seed no longer crashes Leiden and
465
+ PhenoGraph** (#183). The seed is folded into leidenalg's signed 32-bit
466
+ range where it is handed over; seeds below 2^31, and every other engine,
467
+ give exactly the same results as before.
468
+ - **Annotate "Assign names to populations" is one undo step**, as its tip
469
+ says; Ctrl+Z used to undo the clustering import instead.
470
+ - **Each tray tab keeps its own gates.** A gate drawn, moved, switched off
471
+ or deleted in one tab changed the same sample in every tab, because the
472
+ tabs only chose which samples to show. Each tab now holds its own copy of
473
+ its samples' gates: a new or duplicated tab starts from a copy, Copy to
474
+ tab and Apply to other tabs copy the rows' gates into that tab (leaving
475
+ gates it shows as they are there), and Undo goes back to the tab the
476
+ change was made in. A session saves each tab's gates. A session saved
477
+ before this gives every tab a copy of its gates when it opens, with
478
+ nothing to confirm. A tab's members also follow its gates' new ids when a
479
+ session reopens, so a tab no longer loses a gate after one was deleted.
480
+ The session's top-level gates (what the command line and other tools
481
+ read) are those of the tab that was selected when it was saved; each
482
+ other tab's are saved with that tab.
483
+ - **Frequencies: the Metric box redraws, and the window no longer looks
484
+ stuck.** Picking %Total or Count left the plot and summary on %Parent until
485
+ Update was pressed; the status bar said "Computing frequencies…" after the
486
+ numbers had arrived; and a group of one sample showed its SD as "nan",
487
+ now NA.
488
+ - **Frequencies: Update recomputes.** It only regrouped the numbers collected
489
+ when the window opened, so after a gate edit the plot, the tests and every
490
+ export kept the old counts until the window was closed and reopened. It
491
+ now collects again from the current gates and keeps the chosen population.
492
+ A grouping that leaves no sample now clears the summary too, which used to
493
+ keep (and Stats summary… save) the previous test result.
494
+ - **Statistics: a slow, older computation no longer overwrites a newer
495
+ table.** With a large sample, quick toggles or Refresh could end with the
496
+ older request's columns and pre-edit counts on screen and in the exports.
497
+ Only the newest request fills the table now (in both tabs). The Clear
498
+ button's tip now says what Clear does: it empties the table and keeps it
499
+ empty.
500
+ - **A broken audit trail no longer hides a saved export.** Population
501
+ annotation, Debarcode's Save each as FCS, and Spectral QC's report and
502
+ figure exports recorded the export with no guard, so a failure there
503
+ raised after the file was written and the window never said it was saved.
504
+ They, the Frequencies window and the layout editor now record through the
505
+ same helper as the other results windows: the file is reported as saved
506
+ and the failure is noted in the log as "recording the export in the audit
507
+ trail". The audit entries themselves are unchanged.
508
+ - **PhenoGraph clustering no longer forks the running app on Linux.** With
509
+ more than one clustering core, PhenoGraph built its graph in a pool of
510
+ processes forked from OpenFlo, copying the app with whatever locks its
511
+ other threads held at that moment, so a run could hang (Python 3.12 warns
512
+ about exactly this). OpenFlo's PhenoGraph runs now build that graph in
513
+ freshly started processes, as many as the cores asked for rather than
514
+ every core, and sort the clusters without a pool. The clusters are
515
+ identical. Starting those processes adds a second or two to each run with
516
+ more than one core. A script that clusters with more than one core must
517
+ now keep its code under `if __name__ == '__main__':`; without that guard it
518
+ stops with an error that says so. The test suite now fails any test whose
519
+ code forks outside PhenoGraph, in place of a warning filter that also hid
520
+ OpenFlo's own.
521
+ - **The compensation optimizer warns when it skips a single stain.** A
522
+ control it could not read, one missing its own channel, one with too few
523
+ usable, positive or negative events, and one whose positive and negative
524
+ events did not differ were logged only at INFO, which the log pane hides, so
525
+ the matrix came back with that row uncompensated and no word why. Each is
526
+ now a WARNING naming the channel, the file and the reason.
527
+ - **`scripts/theme_audit.py` no longer passes when its setup fails.** A live
528
+ theme switch or test-sample load that raised was swallowed, and the walk then
529
+ reported "OK" on a window that was not in the theme it claimed to check. It
530
+ now exits 2 with the failed step and the error.
531
+ - **FMO gating now names a channel it could not gate.** When a cutoff
532
+ could not be computed (no FMO events inside the selected parent gate, no
533
+ finite values in the channel, or any other error), the channel got no gate
534
+ and the error went only to the console; with one channel mapped, the status
535
+ even said "No FMO mappings chosen". The status now lists each such channel,
536
+ its FMO and the reason, and the error is noted in the log. Other channels
537
+ still get their gates.
538
+ - **A failing autosave now tells you.** The 5-minute autosave only printed a
539
+ failure to the log, so a full disk or a read-only folder could leave hours
540
+ of work unprotected with no sign in the window. Each failure now shows in
541
+ the status line, and after two failures in a row one warning says that a
542
+ crash would lose the work and suggests File → Save session. The warning is
543
+ not repeated on later ticks; it comes back only after an autosave has
544
+ worked and a new run of failures starts.
545
+ - **Cancelling a close whose save failed keeps OpenFlo open.** When the
546
+ save at close failed, the message said to cancel the close but had only an
547
+ OK button, and the app closed anyway, losing the unsaved session. It now
548
+ asks "Close anyway?" with No as the default; No keeps OpenFlo open and
549
+ working (sample loads still waiting resume) so the session can be saved
550
+ elsewhere.
551
+ - **The FCS metadata inspector's voltage and gain columns were always
552
+ blank.** It looked up `$P1V` / `P1V`, but the FCS reader returns keywords
553
+ lowercase without the `$` (`p1v`); it now matches any spelling.
554
+ - **A failed events-file write now shows on the run.** When a workspace
555
+ run could not write its `<label>_events.npz` (a full disk, a locked file),
556
+ the run was marked done with a plain tick and the failure appeared only on
557
+ a progress line the next job replaced. It is now kept as a warning: the
558
+ job's result line and the run's final line say "events file not written"
559
+ with the reason. The clusters and plots it did write still count.
560
+ - **The Stain Index window's Export button tip** said it saves a CSV; it
561
+ saves the plot (PNG, PDF, SVG or TIFF), and now says so.
562
+ - **`environment.yml` matches `requirements.txt` again.** It pinned an old
563
+ tkinterdnd2 (0.4.3, not 0.6.4), left out openpyxl, and left ruff and pyright
564
+ unpinned while CI pins them. FlowUtils, PhenoGraph and tkinterdnd2 are not
565
+ on conda-forge, so they now come from PyPI through the file's pip list. A
566
+ test fails whenever the conda file and the pins drift apart again.
567
+ - **Importing OpenFlo no longer hides your FutureWarnings or exits.**
568
+ `import openflo.pipeline` (and so `import openflo`) switched off every
569
+ FutureWarning in the importing program, and `import
570
+ openflo.inspect_fcs` without a file argument ended Python. The OpenFlo
571
+ commands and the app still keep dependency FutureWarnings off the
572
+ console, from their own startup; `openflo.inspect_fcs` gains `main(argv)`.
573
+ No analysis numbers change.
574
+ - **Every place OpenFlo shows its version shows the same one.** Help >
575
+ Run diagnostics, the About window, the status bar, figure stamps, methods
576
+ text, the CLI run manifest and the audit-trail export and signed record
577
+ each read the version their own way; in a source checkout installed with
578
+ `pip install -e`, that could be an older version than the one saved
579
+ sessions and export stamps record. They now all report the running
580
+ version. No analysis numbers change.
581
+ - **Pipeline Workspace > Batch folder… says why a file failed.** The batch
582
+ counted failures ("2 ok, 1 failed") but threw each error away. The status
583
+ line now names the first few failed files with their reason, and every
584
+ failure is written to the log.
585
+ - **Every Export CSV in the app follows the CSV style preference.** With
586
+ Preferences > CSV style set to this computer's separator and decimal mark,
587
+ only the Statistics and Frequency exports used it; the other Export CSV
588
+ buttons (Compare, Kinetics, Plate, Saturation, Spreading matrix,
589
+ Proliferation, Trajectory, Annotation, Audit log, Sample QC and others)
590
+ still wrote commas and decimal points. They now all follow it. The default
591
+ style writes the same files as before. Files OpenFlo reads back (bead log,
592
+ panel signatures, gate tables, compensation matrices) and command-line and
593
+ library outputs stay comma and decimal point. The audit trail CSV no
594
+ longer gets a blank line between entries in Excel on Windows.
595
+ - **`FlowSample.export_stats` writes the percent-positive columns.** The
596
+ library's per-cluster stats CSV lacked the `pct_pos_<marker>` column that
597
+ `openflo-run` writes for each threshold gate, so the same sample gave
598
+ different columns from Python and from the command line. Both now write
599
+ the same table (`FlowSample.cluster_stats`). Existing columns and numbers
600
+ are unchanged.
601
+ - **Library statistics no longer depend on the app's Preferences.**
602
+ `gating.population_stats` read the GeoMean and rCV forms from the
603
+ Preferences file when given none, so one script could give different
604
+ numbers on two computers. It now uses FlowJo's forms unless passed
605
+ `stat_forms`; the app passes Preferences > Statistics in itself, so the
606
+ Statistics window, exports and layouts are unchanged. Behaviour change
607
+ for scripts on a machine whose Preferences pick a non-FlowJo form.
608
+ - **A layout bar chart of All events shows the statistic it names.** A
609
+ GeoMean bar on the ungated sample showed the plain Mean. Every bar now
610
+ takes its number from the same statistics as the Statistics window, in
611
+ the GeoMean form Preferences > Statistics picks (FlowJo's by default).
612
+ - **Edit > Add singlet gate draws the same gate as the auto-gate.** It
613
+ fitted its band on FSC-H/FSC-A while the auto-gate, Re-place on each
614
+ sample and the doublet filter use FSC-A/FSC-H; on widely spread singlets
615
+ the flipped band let far more doublets through. New menu-made singlet
616
+ gates change; gates already saved keep their corners. Behaviour change.
617
+ - **The command line finds FSC and SSC channels the way the app does.**
618
+ The scatter-channel lookup was written three times; the command line's
619
+ copy matched '-A' anywhere in a name, so on unusual names its FSC vs SSC
620
+ plots could use a different channel than the cleaning steps. One lookup
621
+ now serves all of them; standard names (FSC-A, SSC-A) pick as before.
622
+ - **FMO thresholds from the command line skip infinite values, as the
623
+ app's FMO dialog does.** One infinite event in an FMO control made the
624
+ command-line threshold NaN, so nothing could pass it; infinite values
625
+ are now left out, and a control with no finite values is skipped with a
626
+ note instead of stopping the run. Finite data gives the same threshold
627
+ as before. Behaviour change.
628
+ - **The Python API pairs channels across samples as the command line
629
+ does.** `align_fluor_labels`, `concatenate_by_label` and
630
+ `common_fluor_warning` matched only identical label strings, so 'CD3' and
631
+ 'cd-3', or 'CD901b FL1' and 'CD901b', stayed apart, an unlabelled channel
632
+ never paired with its detector's label elsewhere, and a label on -A
633
+ pooled with the same label on -H. They now use the shared clustering's
634
+ matcher: one set of rules for both. The loaded-samples panel warning
635
+ follows them. Behaviour change.
636
+ - **Marker expression significance brackets sit above the cells.** They
637
+ were placed by the per-sample medians, so they cut through the violins of
638
+ per-cell values. They now sit above the highest plotted cell, as in the
639
+ Frequencies window. Stars and pairs are unchanged.
640
+
641
+ ## [2.8.0] - 2026-10-22
642
+
643
+ Still a public beta. Everything since 2.7.0: macOS fixes, debarcoding, mass
644
+ cytometry basics, panel check, defaults that match between the app, the
645
+ command line and the Python API, more of a session kept on reopen, and the
646
+ fixes from a comparison with FlowJo on public datasets.
647
+
648
+ ### Added
649
+ - **Spectral QC is on the Tools menu.** It used to open only straight after
650
+ a spectral unmix, so the report a reopened session brings back was
651
+ reachable only through Panel check. Before any unmix the entry says to
652
+ run Spectral unmix first.
653
+ - **Debarcoding.** Tools > Debarcode… splits a pooled, barcoded sample
654
+ (fluorescent cell barcoding or mass-tag barcoding) back into its samples
655
+ from a key CSV, by single-cell debarcoding (Zunder et al., 2015): each
656
+ sample becomes a population, with an Unassigned one for the rest, kept in
657
+ the session and listed in Statistics. The separation cutoff (0.3 by
658
+ default) has a preview of each sample's yield and how it falls as the
659
+ cutoff rises; Save each as FCS… writes one file per sample.
660
+ - **Mass cytometry basics.** Tools > Mass cytometry > Bead normalization…
661
+ corrects signal drift over a run with the spiked-in beads (Finck et al.,
662
+ 2013), with a before-and-after plot and an Undo that also works after
663
+ reopening a session. Gaussian parameter clean-up… removes events far out
664
+ on Event_length, Center, Offset, Width or Residual, recomputed for each
665
+ sample (Bagwell et al., 2020).
666
+ - **Panel check.** Tools > Panel check… predicts the similarity matrix,
667
+ complexity index and (on conventional instruments) spillover of a planned
668
+ panel before any data exist, from spectra you load, from the last spectral
669
+ unmix's controls, or from an approximate built-in model on three generic
670
+ instruments. It uses the same scores as Spectral QC.
671
+ - **PhenoGraph resolution.** The Cluster dialog's resolution box now
672
+ applies to PhenoGraph too (as does `--resolution` on the command line).
673
+ 1.0, the default, gives exactly the same clusters as before; lower it to
674
+ about 0.3 to 0.5 when PhenoGraph over-splits data with few markers. It
675
+ needs the default reproducible route; PhenoGraph's Louvain has no
676
+ resolution. The Methods text now states k and the resolution used.
677
+
678
+ ### Changed
679
+ - **FlowSOM defaults to 20 metaclusters everywhere.** `openflo-run`, the
680
+ Python API (`run_flowsom`, `cluster_jointly`) and the Pipeline panel now
681
+ match the Cluster dialog's 20 (they used 10). Scripts that relied on the
682
+ old default should pass `--n-metaclusters 10`; saved Pipeline settings
683
+ that name a count keep it. CytoNorm's own metacluster count stays 10.
684
+ - **Robust SDs use the exact MAD constant.** Every place that turns a
685
+ median absolute deviation into an SD (rSD and rCV in their MAD form,
686
+ bead and voltage robust SDs, ratio and outlier bands, the mass
687
+ cytometry Gaussian clean-up) now multiplies by
688
+ 1 / Phi^-1(3/4) = 1.4826022... instead of the rounded 1.4826, so those
689
+ values rise by about 1.5 in a million. FlowJo's rSD is the percentile
690
+ form, which is unchanged and still the default.
691
+ - **`openflo-run` no longer runs acquisition QC unless you pass `--qc`.**
692
+ It removed drift, flow-rate and off-scale (margin) events from every file
693
+ before gating, with no way to turn it off, while the app loads without QC
694
+ by default (Preferences) and FlowJo has no such step. This applies to the
695
+ analysis run, the `--batch-correct` fit and `--unmix`. In the Python API,
696
+ `FlowExperiment.run_all` now defaults to `qc=False`. Counts and
697
+ percentages change for files QC used to trim; add `--qc` (or
698
+ `run_all(qc=True)`) to get the old numbers.
699
+ - **FMO thresholds default to the 99th percentile everywhere.** The app's
700
+ FMO dialog used 99; `openflo-run --fmo-percentile` and
701
+ `FMOGater.compute` used 99.5. Gates sit a little lower, so % positive
702
+ rises slightly. Pass `--fmo-percentile 99.5` or `compute(percentile=99.5)`
703
+ for the old cut. docs/algorithms.md no longer credits a paper for 99.5.
704
+ - **`compare_groups`, `compare_all_features` and `differential_test`
705
+ default to Welch's t-test** (ANOVA for more than two groups), as the app
706
+ and `openflo-run` do; they ran Mann-Whitney / Kruskal-Wallis, which at
707
+ 3 samples a group can't give p below 0.1. `transform` stays None, since
708
+ bare values don't say what they are: pass `'logit'` for percentages,
709
+ `'log1p'` or `'asinh'` for counts or intensities to match the app. Pass
710
+ `parametric=False` for the rank tests.
711
+ - **CytoNorm fits on at most 20,000 events per batch on every surface.**
712
+ `CytoNorm.fit` had no cap (new `max_events_per_batch`, 20,000; 0 or None
713
+ for all), and `openflo-run --batch-correct` capped each sample at 20,000
714
+ with its own draw; it now fits through the same function and seeded
715
+ per-batch draw as the app's Normalize batches, so the same files give the
716
+ same model. Pass `max_events_per_batch=None` for the old uncapped API fit.
717
+ - **`FlowSample.run_leiden` and `run_louvain` use 30 neighbours, not 15**,
718
+ as the app, the CLI and `cluster_jointly` already did, so a script now
719
+ gets the app's clusters. The self-test's Leiden count moves from 18 to 13
720
+ (against 6 planted populations). Pass `n_neighbors=15` for the old result.
721
+ - **Clustering and embedding use one channel per marker by default.** A
722
+ marker recorded as -A, -H and -W counted about three times in every
723
+ neighbour distance. `channels=None` in `cluster`, `run_leiden`,
724
+ `run_louvain`, `run_flowsom` and every embedding, the CLI, the Pipeline
725
+ panel and the Cluster dialog's ticks now take each detector's -A, its -H
726
+ only when it has no -A (a file with FL1-A and FL2-H used to lose FL2 in
727
+ the Pipeline panel and the CLI's batch correction), and never its -W. The
728
+ dialog still lists every channel. Pass `channels=sample.fluor_channels`,
729
+ or tick every channel, for the old result.
730
+ - **The Cluster dialog passes "Events / sample" to FlowSOM**, which always
731
+ trained on up to 50,000 events whatever the cap; and with no cap its
732
+ status line says Leiden partitions up to 200,000 and FlowSOM trains on up
733
+ to 50,000 events per sample, instead of "full data".
734
+ - **Spectral unmixing: "Non-negative abundances" is off by default** in the
735
+ app (it was on), as it already was for `openflo-run --unmix-nonneg` and
736
+ `spectral.unmix`. Clipping at 0 hides over-unmixing and shifts the medians
737
+ of dim populations; the percent positive above 0 does not change. Tick the
738
+ box for the old behaviour.
739
+ - **Annotate populations: reference matching has its own State threshold,
740
+ 3** (was the MEM threshold, 2, floored at 2), the default of
741
+ `annotate.population_states`. A marker scoring between 2 and 3 no longer
742
+ decides a name, and a threshold below 2 is now used as given. The MEM
743
+ threshold (2) still sets what the labels list. Set State threshold to 2
744
+ for the old names.
745
+ - **`openflo-run` writes its PNGs at 300 dpi** (was 150), the Preferences
746
+ default; the new `--dpi` changes it (`--dpi 150` for the old files). The
747
+ CLI does not read your Preferences file. The HTML report's embedded
748
+ figures stay at 150.
749
+ - **A new layout in the layout editor starts at your Preferences dpi**
750
+ (was always 300). Saved layouts keep their own.
751
+ - **Generate synthetic dataset: event counts start from each generator's
752
+ own default** (20,000 per sample, 6,000 per spectral control, 12,000
753
+ beads; were 5,000 and 8,000), so "Everything (full)" writes what
754
+ `openflo-synth` writes. Type a smaller number for a quicker set.
755
+
756
+ ### Fixed
757
+ - **`openflo-compare` counts FlowJo's AND, OR and NOT populations.** It
758
+ walked only plain populations, so a boolean population and everything
759
+ gated under it were left out of the report without a word (6 of 20
760
+ populations per sample in a public FlowJo 10 workspace). They are now
761
+ read as the workspace import reads them and counted; on that workspace
762
+ every one equals FlowKit's count. An operand the workspace does not have
763
+ gives a row with no count and says which. The CSV, the HTML report and
764
+ the Compare window gain a Path column (`Lymphs/Single Cells/X`), so
765
+ populations of one name under different parents are told apart.
766
+ - **A spillover CSV saved by numpy keeps its first channel.** numpy writes
767
+ the header as `# FL1-A,FL2-A,...`; the `#` stayed on the first channel's
768
+ name, which then matched no channel, and that channel was left out of
769
+ compensation without a word (on a public matrix file, values up to
770
+ 4,086 off). The `#` is now dropped. An events CSV with such a header now
771
+ takes its column names from it instead of naming them Ch1, Ch2, ...
772
+ - **Index-sort wells split over numbered keywords.** BD sorters write a long
773
+ list of well positions as `INDEX SORTING LOCATIONS_1` to `_16`; OpenFlo
774
+ read only the unnumbered keyword and found no positions (0 of 384 on a
775
+ public file). The parts are now joined in numeric order.
776
+ - **Gating-ML export of a NOT over several gates.** OpenFlo's NOT keeps the
777
+ events in none of its operands; the file said NOT of the first operand
778
+ only, so another tool counted more (473 where OpenFlo counts 251). It is
779
+ now written as an AND of every operand used as its complement, the same
780
+ population.
781
+ - **Gating-ML export no longer names a gate it left out.** A boolean whose
782
+ operand was not written (a cluster gate, say) still named it, and other
783
+ tools refused the whole file. Such a boolean, and every gate under it, is
784
+ now left out with a warning.
785
+ - **Gating-ML gates named by a fluorochrome are read on its detector.** A
786
+ gate on a matrix in the file may name the matrix's fluorochrome instead
787
+ of the detector; it found no channel and counted nothing. It is now put
788
+ on the detector the matrix pairs it with (13 of 13 such public test gates
789
+ exact). Gates inside an element other than `<Gating-ML>` itself are not
790
+ read, as before, but the import now says so instead of reading 0 gates
791
+ without a word.
792
+ - **FlowJo workspaces OpenFlo exports open in other tools.** FlowKit refused
793
+ them: the file said `flowJoVersion="OpenFlo-export-1.0"`, had no
794
+ `<Transformations>`, and named a boolean's operands `../X` where
795
+ FlowJo 10 writes the path from the sample. Exports now follow FlowJo's
796
+ own files in all three (the OpenFlo version moves to its own
797
+ `openfloVersion` attribute), and declare a linear axis for each gated
798
+ scatter parameter and for each polygon's parameters, so FlowJo draws
799
+ polygon edges straight in data units, as OpenFlo counts them. On a public
800
+ workspace re-exported by OpenFlo, FlowKit's counts now equal OpenFlo's for
801
+ all 60 populations. OpenFlo reads its own exports, old and new, exactly
802
+ as written. A polygon that reaches below 0 on such an axis is opened by
803
+ FlowJo there; the export's warnings say so.
804
+ - **An FCS file whose HEADER and TEXT disagree on where the data ends** got a
805
+ note saying they disagree on where it starts. It now names the end.
806
+ - **GeoMean of samples imported from a FlowJo workspace uses the
807
+ workspace's axis.** FlowJo's GeoMean (the default form) is the mean on
808
+ the channel's display axis, converted back, so it depends on that axis.
809
+ OpenFlo took it on its own logicle (W=0.5, M=4.5) even when the workspace
810
+ declared another axis per channel, so on public data a population's
811
+ GeoMean read up to 39% below FlowJo's (252 where FlowJo's axis gives
812
+ 413), most for dim channels. A workspace import, and a `.wsp`
813
+ template, now keep each sample's declared axes (logicle, biex, log,
814
+ arcsinh), and GeoMean is taken on them; on that public sample it now
815
+ equals FlowKit's to the third decimal. Sessions save the axes, so a
816
+ reopened sample keeps them; older sessions open as before. Channels the
817
+ workspace declares linear or not at all, and samples not from a
818
+ workspace, are unchanged, as is how plots are drawn. The GeoMean heading
819
+ in the Statistics window, its CSV and Excel exports and the layout
820
+ editor's tables and bar charts names a workspace axis when one was used
821
+ (`GeoMean FL1-A (linear, FlowJo axis: logicle W=1 M=4.42)`); with
822
+ OpenFlo's display axis it stays `GeoMean FL1-A (linear)`, and a column
823
+ whose samples used different axes says so. Preferences > Statistics
824
+ explains it.
825
+ - **FlowJo polygon gates on logicle, biex, log or arcsinh axes.** FlowJo
826
+ draws a polygon's edges straight on the plot's axes, the ones each sample
827
+ declares in the workspace; OpenFlo drew them straight in data units, so
828
+ an imported polygon enclosed another region. On public FlowJo 10
829
+ workspaces 6 to 114 events per polygon differed from FlowKit; now none
830
+ do. This applies to `openflo-compare`, the editor's workspace import and
831
+ `.wsp` templates. A polygon put on another scale (the editor's display
832
+ scale, or an export back to FlowJo) is traced along FlowJo's axes, within
833
+ 0.2% of the exact count in the tests. Axes the workspace does not declare,
834
+ as in OpenFlo's own exports, keep straight edges in data units, so an
835
+ export reads back as written. Gating-ML files already did this.
836
+ - **FlowJo Time gates use the seconds per tick FlowJo wrote.** FlowJo 10.6
837
+ and 10.8 convert Time with ($ETIM − $BTIM) / last tick and record that as
838
+ the gain of the Time axis, which can differ from the file's `$TIMESTEP`
839
+ (0.0100853 against 0.01). OpenFlo now uses that gain when a workspace
840
+ gives one other than 1, and FlowJo's Time counts on public data now
841
+ match exactly. Without a gain, the file's `$TIMESTEP` still applies, as
842
+ measured in FlowJo 10.10.2.
843
+ - **FlowJo 9 workspaces get a plain message.** Their `<!DOCTYPE Workspace>`
844
+ line was refused as a "possible entity-expansion attack". OpenFlo now
845
+ says FlowJo 9 (and older) workspaces are not supported yet and to re-save
846
+ the file in FlowJo 10. A DTD that declares entities is still refused.
847
+ - **A workspace sample matched to the wrong file is reported.** A
848
+ workspace can name several files alike, and the file found may not be
849
+ the one FlowJo gated. When its event count differs from the count the
850
+ workspace records, `openflo-compare` says so in each row of that sample,
851
+ and the editor's import says so in its load notes. The import still goes
852
+ ahead.
853
+ - **Batch normalization (CytoNorm) is kept when a session is reopened.**
854
+ Applied to the loaded samples, it was lost on reopen for samples that were
855
+ not compensated: they came back on their raw values, so gates and
856
+ statistics silently changed. The normalized values are now saved with the
857
+ session (Save, Save As and the autosave). After a reopen, Undo batch
858
+ normalization says there is nothing to undo (the values from before are
859
+ not kept); reload the FCS files to start again from the raw values. If
860
+ the saved values are missing, the load notes say the normalization was
861
+ not restored.
862
+ - **More of a session survives a reopen.** Samples reopened from their
863
+ saved data kept no FCS keywords, value ranges or QC count, so
864
+ compensation's "Apply to: this batch", CytoNorm keyword batches and
865
+ kinetics time units saw nothing; they are now read back from the FCS
866
+ file. Saved channel labels no longer give way to the first sample's own,
867
+ a renamed duplicate sample (e.g. "tube [Day 9]") keeps its name and
868
+ colour, and the cell-cycle and FlowSOM results reopen with their columns,
869
+ so their windows no longer ask for a re-run.
870
+ - **"Display + data" downsampling is kept exactly on reopen.** A trimmed
871
+ sample came back with all its events when the smallest sample had been
872
+ removed or unticked. The trim is now saved and replayed, giving the same
873
+ events. Turning the trim off after a reopen gives every event back, or
874
+ the status line says why a sample stays trimmed.
875
+ - **A spectral unmix is kept on reopen.** The unmixed channels came back
876
+ but the reference spectra and Spectral QC report did not, so Panel
877
+ check's "Use last unmix" was greyed out. They are now saved with the
878
+ session.
879
+ - **Relinking missing files picks the right one.** When two missing samples
880
+ shared a file name (e.g. "tube.fcs" from Day 6 and Day 9), both were
881
+ linked to the first such file found, so one sample's data loaded under
882
+ the other's name and gates. Each now takes the file in its own folder.
883
+ - **The Methods paragraph counts each sample's events once.** Every reopen
884
+ of a session added its samples' events again (900 events read as 1,800,
885
+ then 2,700).
886
+ - **`openflo-run --cytonorm-mode controls` without `--cytonorm-control`
887
+ stops with an error** instead of silently pooling every sample (goal mode
888
+ under another name), and a batch with no sample matching the token is an
889
+ error too, as in the app. The help now says the app defaults to
890
+ 'controls' with the token 'ref', and the CLI to 'goal' because it can't
891
+ guess which sample is the reference.
892
+ - **The Cluster dialog embeds on the markers it clustered.** UMAP, t-SNE
893
+ and the others ran on every fluorescence channel even when you picked a
894
+ few. Compare embeddings also embeds one channel per marker, now with seed
895
+ 42 (was 0) and UMAP at 30 neighbours / min_dist 0.3 as elsewhere; t-SNE
896
+ and PHATE keep their own settings.
897
+ - **The Pipeline panel says when its event cap subsampled a run**
898
+ ("clustered 5,000 of 48,000 events"): the cap applies to the run unit's
899
+ combined events, so a rare population can thin out.
900
+ - **macOS: UMAP, TriMap and PaCMAP no longer crash the app** when two of
901
+ them run at once (say the warm-up and a plot). numba's only thread pool on
902
+ macOS aborts the process when used from two threads; those calls now take
903
+ turns there. Linux and Windows, whose pools are thread-safe, are
904
+ unchanged.
905
+ - **macOS: right-click menus open.** Tk on a Mac reports the right button
906
+ as button 2, so the menus on the tab strip, both trees and the layout
907
+ editor never appeared; they now listen for the right button on every
908
+ system, and the middle-button drag in the layout editor follows suit.
909
+ - **Gating-ML boolean gates with a NOT on one part import.** A gate such
910
+ as "A and not B" from another program was left out with a warning; each
911
+ complemented part now becomes an OpenFlo "not" gate beside it ("not B"),
912
+ and the gate counts the same events FlowKit does.
913
+
914
+ ## [2.7.0] - 2026-10-21
915
+
916
+ This is **2.7.0**, still a public beta (marked so on PyPI), released
917
+ under a final version number so a plain `pip install openflo` gets it and the
918
+ update check offers it to everyone on 2.6.x. The same code was on PyPI as
919
+ 2.7.0b2 (pre-release, `--pre` only); 2.7.0b1 went only to the public source
920
+ mirror.
921
+
922
+ ### Changed
923
+ - **No error disappears unseen any more.** An error in a background
924
+ thread, or in background work with no handler of its own, now opens the
925
+ same "Something went wrong" window and problem report as any other,
926
+ instead of only printing or only flashing in the status bar. The 418
927
+ places that caught every error and did nothing now catch only what they
928
+ expect, write a "[handled]" line to the log, or say why nothing can
929
+ matter there; a test keeps new ones out.
930
+ - **Dialogs say which setting is wrong.** Fifteen number settings
931
+ (Cluster, Cell cycle, Proliferation, Bead tracking, Auto-gate, Batch-norm,
932
+ Calibration, Embedding, MEM annotation, FMO, Spectral unmixing,
933
+ Trajectory, Figure layout and the layout editor's text size) refuse text
934
+ that is not a number in range with a message naming the setting.
935
+ **Behaviour change:** several used to swap a mistyped value for a default
936
+ without saying so, or clamp it into range; they now stop instead.
937
+ - **Re-fit per sample finds the valley before a bright population at the
938
+ top of the range.** It used to say "one population, no valley" and keep
939
+ the old cut when a tight population sat at the top of the data.
940
+ **Behaviour change:** such gates now move to the valley.
941
+ - **The layout editor's Layout and Export buttons** no longer show the menu
942
+ arrow twice.
943
+ - **The embedding warm-up no longer stalls.** Tk clean-up that a background
944
+ thread's garbage collection triggers is handed to the window's event loop
945
+ rather than waiting on it.
946
+ - **Group comparisons can come out significant at 3 to 5 samples a
947
+ group.** Mann-Whitney's smallest possible p at 3 vs 3 is 0.1, so two
948
+ groups that did not overlap at all read "not significant". Frequencies
949
+ and Marker expression now default to Welch's t-test (ANOVA for more
950
+ groups) on transformed values: percentages as log odds, counts as log,
951
+ intensities on the arcsinh scale (cofactor 150). The Test box keeps the
952
+ rank test and a t-test on raw values. **Behaviour change:** p-values in
953
+ those windows, Compare all, and the CLI's `compare_*.csv` (now Welch's t
954
+ on the logit of each sample's %) differ from earlier versions on purpose.
955
+ - **Paired by donor.** A tick in Frequencies and Marker expression matches
956
+ samples by their name with the group taken out (D1_Stim with D1_Ctrl):
957
+ two groups get a paired t-test (Wilcoxon signed-rank with the rank test),
958
+ more get a repeated-measures ANOVA (Friedman). Differential abundance
959
+ then models the donor as a term (`count ~ donor + group`, as diffcyt's
960
+ edgeR model), so a donor's own level is taken out before the groups are
961
+ compared. A sample with no partner is left out and named.
962
+ - **Saying when a test can't reach 0.05.** Every result carries the
963
+ smallest p its test could give at those group sizes; the summaries,
964
+ Compare all and Group comparison say plainly when that is above 0.05.
965
+ - **MESF calibration fits log-log by default.** Tools > Calibration fits
966
+ log MESF against log MFI, the standard for bead sets spanning decades;
967
+ Linear is one click away. `fit_mesf_calibration` and `apply_calibration`
968
+ take `model='loglog'` and keep linear as their default.
969
+ - **Faster `openflo-run` batches, same files.** With `--workers` above 1,
970
+ each sample's scatter plots and stats CSV are drawn in the worker that
971
+ clustered it, instead of one at a time in the main process while the
972
+ workers wait; group UMAPs are drawn in parallel; and `compare` takes
973
+ `--workers` (default up to 4) and compares samples in parallel, reporting
974
+ them in workspace order. The files are byte-identical to a one-worker run.
975
+ - **Every dialog behaves the same way.** Each window opened from the editor
976
+ stays in front of it (no taskbar entry of its own), Escape closes it, and
977
+ Return runs its main action, which sits beside Cancel or Close at the
978
+ right of the button row. The layout editor and a popped-out workspace stay
979
+ separate windows.
980
+ - **Errors in plain words.** An error says what failed and why ("The file
981
+ may be open in another program") instead of the exception's name; Show
982
+ details has the technical description and Copy details copies it for a
983
+ bug report. An unexpected error opens that window once instead of popping
984
+ the log and console open.
985
+ - **Keyboard shortcuts and About are proper windows**: shortcuts in two
986
+ columns by menu, and About with the version, licence, how to cite (with a
987
+ Copy citation button) and links.
988
+ - **Clearer wording.** Side-panel buttons say what they do (Delete gate,
989
+ Delete all, Copy gates…, Populations…, Undo, Redo, Fit, Apply to all
990
+ shown), beta tags moved into the tips, no arrows used as words (File >
991
+ Export FlowJo workspace…), sentence case, American spelling, the methods'
992
+ published names (PhenoGraph, t-SNE, TriMap), and the Pipeline's fields
993
+ named in full (Method, Neighbors, Resolution, Max events, Seed).
994
+ - **The title names the session**: "session name — OpenFlo", with * while
995
+ there are unsaved changes.
996
+ - **Menus regrouped, with their shortcuts.** Tools is in four groups
997
+ (compensation and scales, figures, utilities, templates); exports moved
998
+ to File, and Methods & provenance sits beside History / audit in Analyze.
999
+ Menus show Ctrl+Z, Ctrl+Y, Ctrl+F, the zoom keys and F5, and on macOS
1000
+ the same keys answer to Command. F1 opens the documentation, and the main
1001
+ window's File menu ends in Quit OpenFlo (Ctrl+Q).
1002
+ - **No pop-up after an export.** The status bar says what was saved, with
1003
+ an Open folder button, instead of a box to click away.
1004
+ - **Theme: windows and plot background, chosen separately.** View > Theme
1005
+ is Light or Dark windows, and View > Plot background is White or Black;
1006
+ figures in analysis windows follow the plot background, so "Dark figures
1007
+ in pop-ups" is gone. Sessions keep their theme.
1008
+ - **Readable fonts and colours.** Windows use named fonts that follow the
1009
+ window-width scaling and are never below 9 pt. Error, warning and success
1010
+ text, the drag highlight and the log pane take their colours from the
1011
+ theme, each readable at 4.5:1 or more; the log is light in the light
1012
+ theme.
1013
+ - **Cluster dialog**: only the picked method's settings show (FlowSOM's
1014
+ grid and metaclusters, Leiden's resolution), and markers are listed as
1015
+ "marker (channel)", as the axis menus show them.
1016
+ - **Layout editor**: Ctrl+Z and Ctrl+Y undo and redo, Ctrl+S saves the
1017
+ session (which carries the layout), and a change marks the session
1018
+ unsaved, so closing asks to save it. The empty-page hint names the real
1019
+ buttons and the status line wraps instead of running off the window.
1020
+ - **Tools > Quick figure grid** is the old Figure layout, one name in the
1021
+ menu and the window, so it is not mistaken for the layout editor.
1022
+
1023
+ ### Added
1024
+ - **Results record the OpenFlo version that made them (session schema
1025
+ v3).** A gate placed by Re-place on each sample or by Auto-gate, and each
1026
+ column written by clustering, an embedding, cell cycle, calibration or
1027
+ pseudotime, carries a stamp: the step, the OpenFlo version and what it ran
1028
+ on. The `.flowsession` gains a top-level `app_version`, and each sample
1029
+ entry an `app_version` and `result_stamps` (gate stamps travel on the gate
1030
+ as `stamp`). Saving writes an older result's stamp back unchanged, so a
1031
+ later fix can find the results made before it. Each audit-trail entry
1032
+ also records the version (`version`). Version-2 sessions upgrade on open,
1033
+ their results' versions unknown; a session saved now does not open in an
1034
+ older OpenFlo.
1035
+ - **Fixes reach saved sessions, once and only when you say so.** When a
1036
+ session opens, OpenFlo lists the saved results made by a version with a
1037
+ bug since fixed: each sample's gate or column, the version that made it,
1038
+ and its numbers before and after (events and % of parent for a gate).
1039
+ Tick the ones to correct and Apply (one undo step); nothing changes
1040
+ otherwise. A corrected result is recomputed with this version, its stamp
1041
+ moves to it, and its original stays in the session (a gate's
1042
+ `original`, a column's values under `<column> [original <version>]`,
1043
+ out of plots and statistics). Fixes not taken are recorded and not
1044
+ offered again; Tools > Review fixes for this session… lists them all.
1045
+ Gates and other results built on a corrected one are offered as a
1046
+ follow-up re-run, never started by themselves; for clusters, embeddings,
1047
+ trajectory, cell cycle and calibration, Open… opens that tool on the
1048
+ sample. The same list names the files exported from a corrected sample
1049
+ before the fix (when made, by which version, whether still there), with
1050
+ Show in folder; they are not re-exported. Plot images, Gating-ML,
1051
+ FlowJo .wsp, population FCS folders and the Statistics, Frequencies,
1052
+ Kinetics, workspace-comparison and differential-abundance exports are
1053
+ recorded for it too. A changed default is
1054
+ flagged, not applied. The first fix listed is the re-fit valley fix
1055
+ (#78), for gates re-placed per sample before 2.7.0b2. Exports name the
1056
+ fixes applied. The session gains `fixes_applied` and `fixes_declined`,
1057
+ and each sample entry `result_originals`.
1058
+ - **`openflo-run check-sessions <folder>`** lists, for every `.flowsession`
1059
+ under a folder, the fixes that would apply to it. It only reads.
1060
+ - **A gate moved by hand is yours.** Dragging, typing coordinates, the
1061
+ arrow keys or a slider takes the step's stamp off the gate (its stamp
1062
+ records the geometry it was placed with), and a pasted or copied gate
1063
+ carries none, so a later fix never overwrites a hand edit.
1064
+ - **A reopened session shows its gate counts and statistics at once.** Each
1065
+ sample entry saves the counts its gates had and the Statistics window's
1066
+ last few tables (`saved_stats`, numbers only), and a reopen uses them
1067
+ instead of computing them again, but only when the sample's data digest,
1068
+ the gates (and their ancestors), the statistics chosen, the Preferences
1069
+ that change a formula and the OpenFlo version are all the same; anything
1070
+ else is computed again as before. Measured on 6 samples of 500,000
1071
+ events with 10 gates each: the counts were all shown 0.8 s after the last
1072
+ sample loaded, now as each one loads; the Statistics window was ready in
1073
+ 1.1 s, now 0.25 s. Plots are not saved.
1074
+ - **Normalize batches dialog.** Tools > Normalize batches (CytoNorm)
1075
+ replaces the one-click batch-norm: batches by trial / day or by any FCS
1076
+ keyword ($DATE by default), the mapping learnt from a reference sample
1077
+ in every batch (named with a token such as "ref"; CytoNorm's standard
1078
+ form) or from all samples (CytoNorm 2.0), 1 to 40 metaclusters, applied
1079
+ to the loaded samples and/or written as normalized FCS copies. It shows
1080
+ each batch before and after on a channel, the per-batch medians and a
1081
+ methods sentence, which also goes into the audit trail. Compensation
1082
+ controls and set-aside samples stay out of the fit. Restore original
1083
+ values (in the dialog, or Tools > Undo batch normalization) puts the
1084
+ loaded samples back on their pre-normalization values without reloading.
1085
+ - **Excel workbooks and European CSV.** Statistics has Save as Excel: the
1086
+ whole table on one sheet, then one sheet per statistic (populations
1087
+ down, samples across), and an About sheet with the OpenFlo version.
1088
+ Frequencies can save its tidy table as .xlsx, and `openflo-run --xlsx`
1089
+ collects the run's CSV tables into one workbook. Preferences > Export >
1090
+ CSV tables can follow this computer's list separator and decimal mark
1091
+ (semicolon and decimal comma across most of Europe), so Excel opens the
1092
+ CSV in columns; copied tables follow the same setting. The default CSV
1093
+ is unchanged.
1094
+ - **Set a sample aside.** Right-click a sample > Set aside from
1095
+ comparisons, with a reason (failed QC, clogged, too few events). It stays
1096
+ in the session and in the Statistics table (with a Set aside column),
1097
+ is greyed with its reason in the tree, is left out of every group
1098
+ comparison (Frequencies, Expression, Group stats, normalization) and is
1099
+ listed with its reason in the methods text. A free-text note per sample
1100
+ too. Both are saved in the session (additive keys an older OpenFlo
1101
+ ignores).
1102
+ - **Colour-blind-safe colours and keyboard gating.** Samples and groups
1103
+ now take the Okabe-Ito colours, which stay distinct under the common
1104
+ colour-vision deficiencies (tab10's red and green do not); View > Group
1105
+ colors switches back to the classic palette, keeping colours you picked
1106
+ by hand. The pipeline's `--palette auto` follows. The plot takes keyboard
1107
+ focus (click it, or F6 from the gate list): the arrow keys move the gate
1108
+ selected in the list, Shift for larger steps, one undo step per run of
1109
+ presses.
1110
+ - **Setup without Python.** `setup.bat` and `setup.sh` no longer stop when
1111
+ no Python 3.11+ is installed (an older Python is now refused up front
1112
+ instead of failing in pip). They use a `uv` on PATH, or download a pinned
1113
+ uv release (Windows x86_64/ARM64, macOS Intel/Apple silicon, Linux
1114
+ x86_64/aarch64) into `.tools/uv`, check its SHA-256 against the hash in
1115
+ the script and refuse to continue on a mismatch, then let uv install
1116
+ Python 3.12 and create `.venv`. An interactive Windows setup also adds
1117
+ Desktop and Start-menu shortcuts. The scripts are not code-signed, so
1118
+ SmartScreen or Gatekeeper may still warn.
1119
+ - **Batch normalization helpers (library).** `openflo.batch_normalize`
1120
+ wraps the existing CytoNorm model for acquisition-day batches: group
1121
+ samples by any FCS keyword (`$DATE` by default in a dialog), fit on a
1122
+ reference sample run in every batch ('controls') or on all samples
1123
+ ('goal'), apply to new frames without touching the loaded ones, a
1124
+ per-channel, per-batch median and robust-SD before/after table with a
1125
+ methods-report sentence, and normalized FCS copies (compensated linear
1126
+ values, source keywords kept, `OPENFLO_NORMALIZATION` / `OPENFLO_NORM_*`
1127
+ provenance keywords added) written to a folder. Original files are never
1128
+ modified.
1129
+ - **Combination gates.** Edit > Combination gates… (also on a gate's
1130
+ right-click menu) takes 2 to 6 gates and makes every combination of
1131
+ them, all positive to all negative, as boolean populations under the
1132
+ gates' common parent, on every sample that has those gates; a negative
1133
+ means in the parent and outside the gate, and an event with no value is
1134
+ in no combination. A polyfunctionality window shows SPICE-style bars
1135
+ per combination (group means with a dot per sample), the +/- grid and
1136
+ pies of how many gates each positive event is in, and exports the
1137
+ frequency table (NA where a frequency cannot be computed). The
1138
+ combinations are ordinary gates, so they reach statistics and the
1139
+ FlowJo export.
1140
+ - **Typed gate coordinates.** A gate's right-click > Edit coordinates…
1141
+ types a threshold, interval, rectangle, quadrant divider, polygon's
1142
+ vertices or ellipse's centre and shape in the axis units (intensity;
1143
+ 1.5K reads as 1,500, a blank bound is open), optionally on the same gate
1144
+ in every other sample. One undo step.
1145
+ - **Gates that re-fit per sample.** A threshold or interval gate, or a
1146
+ singlet or Gaussian-mixture gate from Auto-gate, can be re-placed on
1147
+ each sample's own data (right-click > Re-place on each sample…): a cut
1148
+ moves to the density valley nearest where it was drawn, and a proposal
1149
+ is fitted again by its own rule. A grid shows every sample's old and
1150
+ new placement before you accept; a sample with no landmark keeps its
1151
+ gate and says why.
1152
+ - **Compare FMO thresholds.** Edit > FMO gating has a Compare
1153
+ thresholds button: every stained sample's % positive on each mapped
1154
+ channel with the cutoff at each FMO percentile you list (99 and 99.5 by
1155
+ default, i.e. 1 % and 0.5 % of FMO events above the gate), inside the
1156
+ selected parent gate, without adding a gate. Copy or export the table.
1157
+ - **FMO on histograms.** Once FMO gating has mapped a channel to its FMO
1158
+ control, a histogram of that channel draws the FMO in grey behind the
1159
+ sample, and the legend gives each one's % at or above the channel's
1160
+ threshold gate. View > FMO on histograms turns it off. The mapping is
1161
+ saved with the session.
1162
+ - **Gating strategy page.** The layout editor's + Strategy (also on the
1163
+ page menu) adds a page with one plot per gating step, each the parent
1164
+ population on the gate's axes with its gates drawn; with a gate selected
1165
+ in the main window, only the steps down to it. Batch export draws it for
1166
+ every sample.
1167
+ - **PowerPoint export.** The layout editor exports a page, or every
1168
+ sample's pages in a batch, as a .pptx deck: one slide per page, the page
1169
+ drawn at the export resolution and filling a slide of the page's size.
1170
+ - **Compare with saved session.** File > Compare with saved session counts
1171
+ every population with this session's gates and with a saved session's
1172
+ gates, on the events loaded now, and highlights each count that moved
1173
+ and each gate that changed.
1174
+ - **Gate list.** File > Export gate list (CSV) writes every sample's gates
1175
+ with their population path, kind, channels and coordinates, both as
1176
+ stored and in linear units.
1177
+ - **FlowJo export checks Time gates.** A Time gate whose upper edge ends
1178
+ within the first `$TIMESTEP` fraction of the run (where an edge already
1179
+ converted to seconds lands) now warns, with the events it cuts, since
1180
+ OpenFlo keeps Time in ticks and converts it to seconds for FlowJo.
1181
+ - **Derived channels.** Tools > Derived channels makes a channel computed
1182
+ per event from others, such as the ratio of two detectors for a ratio
1183
+ dye, FRET, or one marker over another: `[FL1-A] / [FL2-A]`,
1184
+ `log10([FL1-A] / [FL2-A])`. Channels go in square brackets (a marker name
1185
+ works too); + - * / ** and log, log10, log2, sqrt, exp, abs, asinh, min
1186
+ and max can be used, and nothing else is run. It is computed on
1187
+ compensated linear values, shows as `D:<name>` to plot, gate and
1188
+ summarise, is saved with its formula (in the sample entry, so older
1189
+ sessions open as before) and recomputed when a session reopens from the
1190
+ FCS files. An event with no finite result (0 / 0) is NaN, not 0.
1191
+ - **Kinetics.** Analyze > Kinetics follows a channel over acquisition time
1192
+ (calcium flux, dye uptake): one binned median (or mean) trace per sample,
1193
+ overlaid, and a table of each sample's baseline, peak after the
1194
+ stimulus, time to peak, area over the baseline and peak / baseline, with
1195
+ CSV and figure export. Time is in seconds when the file gives
1196
+ `$TIMESTEP`, else in ticks, and the window says which. What can't be
1197
+ computed (no baseline events, a baseline at 0 for the fold) reads NA.
1198
+ - **Two windows can't overwrite one session silently.** Opening or saving a
1199
+ session by name writes a small lock file beside it
1200
+ (`name.flowsession.lock`) naming the computer and user. A second person,
1201
+ or a window of another OpenFlo, is told the session is open and picks
1202
+ Open read-only (Save then asks for a new name), Open anyway or Cancel. A
1203
+ lock left by a crash on this computer is taken over without asking.
1204
+ Before Save, or Save and close, writes over the file this window opened,
1205
+ it checks the file is still the one it read or last saved; if someone
1206
+ saved in between (or the file was deleted), it asks: Save as a copy,
1207
+ Overwrite or Cancel. Synced folders copy lock files late, so this check
1208
+ is what protects the data.
1209
+ - **Every test proven to pass and to fail.** `scripts/prove_tests.py` runs
1210
+ each test on the real code, then breaks the code it guards one change at
1211
+ a time and records the break that made it fail. `docs/test-assurance/`
1212
+ holds the result for all 3,420 tests (3,406 proven both ways) and the
1213
+ method. Tests that needed a developer's own sample files now generate
1214
+ their data, and tests that could never fail were strengthened.
1215
+ - Hover help on the auto-gate method choices, the embedding method boxes and
1216
+ the axis-settings parameter box.
1217
+ - **Offline user guide.** Help > User guide (F1) opens a guide and
1218
+ tutorial that ship with OpenFlo, so it works without a connection: ten
1219
+ topics from a first analysis to troubleshooting, with search, and the
1220
+ tutorial can load the example dataset. Help > Online documentation still
1221
+ opens the web page. A test checks that every menu path the guide names
1222
+ exists.
1223
+ - **Dean-Jett-Fox cell cycle.** Analyze > Cell cycle fits G1 and G2/M peaks
1224
+ and a polynomial S phase (Fox 1980) by default and reports their areas,
1225
+ so S-phase cells under the peaks count as S (synthetic truth 56/30/14
1226
+ read 55.8/30.5/13.6, against 53.4/29.3/17.4 by boundaries). The window
1227
+ draws the fit and lists the boundaries result beside it; Boundaries stays
1228
+ available, and `cell_cycle(method=...)` defaults to it.
1229
+ - **Proliferation.** Analyze > Proliferation fits a dye-dilution histogram
1230
+ as generations one halving apart and reports cells per generation,
1231
+ precursor frequencies, percent divided and the division, proliferation
1232
+ and expansion indices (Roederer 2011). The undivided peak is found, or
1233
+ fixed at an unstimulated control's median. Too few cells read NA.
1234
+ - **Spreading matrix for compensation.** Tools > Spreading matrix measures
1235
+ the spillover spread of conventional compensation from compensated
1236
+ single-stain controls, matched to their channels automatically; a
1237
+ channel without a control reads NA. CSV and figure export.
1238
+ - **Several autofluorescence signatures.** Spectral unmix can find up to
1239
+ five autofluorescence spectra in the unstained control; each cell is
1240
+ unmixed with the one that fits it best, and a U:AF index channel says
1241
+ which. Spectral QC adds Unmixed pairs, an NxN plot of every pair of
1242
+ unmixed fluors.
1243
+ - **Bead tracking.** Tools > Bead tracking keeps a Levey-Jennings log of QC
1244
+ bead runs (a CSV) and judges each run by the Westgard rules (1-3s, 2-2s,
1245
+ R-4s, 4-1s and 10x reject; 1-2s warns).
1246
+ - **Index-sort data.** Index-sort locations in the FCS keywords
1247
+ (INDEX SORTING LOCATIONS, $INDEX and similar) are read, and the plate view
1248
+ shows each sorted cell in its well, coloured by a channel.
1249
+ - **Group samples by keyword or table.** Edit > Group samples puts the
1250
+ samples into groups by an FCS keyword the files carry (tube, condition,
1251
+ plate, ...) or by a sample-to-condition table (CSV or tab separated: the
1252
+ sample or its file name, then its group), with a preview before Apply.
1253
+ The groups are the ones the sample tree and the group statistics use.
1254
+ - **Plate view.** Analyze > Plate view shows a 96-well plate heatmap of the
1255
+ events, or a gate's count or percent of its parent, per well. The well
1256
+ comes from the well keyword or the sample's name (Plate1_B07); a sample
1257
+ without the gate reads NA, and samples without a well are listed.
1258
+ - **Export populations, all samples.** File > Export populations, all
1259
+ samples (FCS) writes every sample's gated populations, a folder per
1260
+ sample.
1261
+ - **Concatenate samples** from File: joins several samples with the same
1262
+ channels and scales into one new sample, each event tagged with its
1263
+ source in a `__sample__` column.
1264
+ - **Downsampling the data can be undone.** Setting the downsample from
1265
+ Display + data back to Display only or Off restores every event, except
1266
+ in a sample whose data changed after the trim, which is named.
1267
+ - **More statistics.** Analyze > Statistics adds %Grandparent, SD, robust
1268
+ SD (half of P84.13 to P15.87), MAD, any percentiles you type (P5, P95,
1269
+ ...) and the MFI ratio: a population's median over the same population's
1270
+ median in a reference sample you pick, such as the unstained control.
1271
+ The column ticks sit on three rows so they fit a laptop screen, and Copy
1272
+ table puts the table on the clipboard, tab separated, for Excel or Prism.
1273
+ - **Editable figure text.** PDF and SVG exports keep their labels as text
1274
+ (TrueType in PDF, `<text>` in SVG), so Illustrator and Inkscape can edit
1275
+ them; Preferences can switch it off. Preferences also set the export
1276
+ resolution (150 to 1200 dpi, 300 by default) and the figure font, from the
1277
+ fonts installed. File > Copy plot to clipboard, and right-click on the
1278
+ figure in analysis windows, copy a figure as an image.
1279
+ - **Saturation report.** Analyze > Saturation report lists, per sample,
1280
+ the percent of events at the top of each channel's range (`$PnR` - 1, on
1281
+ the raw detector values) and across any channel, flags channels at 1 %
1282
+ or more, and exports or copies the table. A channel whose range the file
1283
+ does not give reads NA.
1284
+ - **Gating-ML 2.0 files.** File > Export Gating-ML 2.0 gates writes the
1285
+ active sample's gate tree in the ISAC standard other analysis tools read,
1286
+ with each channel's scale (logicle, hyperlog, asinh as `fasinh`, log as
1287
+ `flog`) and the sample's compensation matrix, so polygons keep their
1288
+ shape. Templates > From file… and `openflo-run --gates-from` read them;
1289
+ a polygon stays straight on the axes the file names. Gates Gating-ML
1290
+ cannot hold, or OpenFlo cannot read (quadrant gates, ratio dimensions),
1291
+ are left out with what is under them, and named.
1292
+ - **FlowJo groups come across.** Opening a `.wsp` puts each sample in an
1293
+ OpenFlo group named after its FlowJo group (not 'All Samples' or
1294
+ 'Compensation'), and the `.wsp` export writes OpenFlo's groups as FlowJo
1295
+ groups. A workspace whose FCS files moved finds them in the subfolders
1296
+ of its own folder or the FCS folder; two files of one name are told
1297
+ apart by their folder, never guessed.
1298
+ - **`openflo-run` reproducibility.** Each run writes `run_manifest.json`
1299
+ (command, settings, software versions, gates, and inputs and outputs
1300
+ with SHA-256). `--gates-from` gates every sample on one population
1301
+ (`--gate-population`) of a GUI session, template, `.wsp` or Gating-ML
1302
+ file. `--gate-units data` takes `--gates` in intensities instead of
1303
+ logicle units. The comparison CSV gains `p` (Mann-Whitney U) and `p_adj`
1304
+ (Benjamini-Hochberg), as the GUI's differential abundance has.
1305
+ - **Project archives.** File > Export project archive saves one .zip with
1306
+ the session, its saved data and every sample's FCS file, to send or move
1307
+ as one file. Open session opens it directly: it unpacks beside the zip
1308
+ and relinks the samples there. Samples whose FCS file was not found are
1309
+ listed when exporting.
1310
+ - **Save session as.** Save (Ctrl+S) now writes back to the session you
1311
+ opened or last saved, and asks for a name only the first time; Save
1312
+ session as picks a new one. The window title shows the session's name.
1313
+ - **Exploded view of a zoom or gate in layouts.** A layout plot whose zoom
1314
+ hides part of its data shows the whole plot with the zoomed region
1315
+ outlined, beside a magnified copy joined to it by connector lines, on the
1316
+ page and in Export page and Export batch. Edit… sets it per plot (Auto,
1317
+ Always when zoomed, Off), chooses the copy's side, and can magnify one of
1318
+ the plot's gates (its events, outline and %) instead of the zoom.
1319
+
1320
+ ### Changed
1321
+ - **Nothing written inside the installed package.** PhenoGraph's scratch
1322
+ files now go to a temporary folder per run that is removed afterwards, even
1323
+ when the run fails (they used to collect in a `.phenograph_cache` folder
1324
+ inside the package, which is read-only on a system Python). Saved gate
1325
+ templates now go to `~/.openflo/templates`; templates saved by an older
1326
+ OpenFlo inside the package still show in the Templates menu.
1327
+ - **The editor window appears sooner.** It paints before matplotlib loads
1328
+ and the plot area fills in right after, so the window shows in about
1329
+ 0.3 s instead of 0.7 s on a warm start (more on a cold disk).
1330
+ - **`openflo-run --help` answers at once.** Help and argument errors no
1331
+ longer load the scientific stack (0.07 s instead of 0.8 s); the console
1332
+ script's entry point is now `openflo.cli_args:main`.
1333
+ - **The window keeps answering while OpenFlo works.** Dragging a gate
1334
+ redraws only its outline, and its counts are made in the background
1335
+ after you let go (they used to be redone on every mouse move with gates
1336
+ applied). Zooming or changing axes no longer rebuilds the Samples & Gates
1337
+ tree. Trajectory, Marker expression, FMO gating, cell cycle, population
1338
+ export, Sample QC and compensation optimisation run in the background
1339
+ with a progress bar and Cancel; layout batch export shows how far it got
1340
+ and stops on Cancel. Results are unchanged.
1341
+ - **Gate rows in the Samples & Gates tree show the population.** Each row
1342
+ reads the gate's name, its event count and its frequency of the parent
1343
+ ('Cells 2,796 · 69.9%') instead of its bounds, which were cut off at the
1344
+ panel's width. An unnamed gate is named by its channels ('FSC-A x SSC-A')
1345
+ or its cut ('FL1-A >= 1K'). The tree widens to its longest row and
1346
+ scrolls sideways. A population that cannot be computed reads NA.
1347
+ - **Colouring by cluster gives each cluster its own colour**, the colour of
1348
+ its population in the tree, with a legend naming them. It used a
1349
+ continuous scale, so neighbouring clusters looked alike.
1350
+ - **Groups are named after the sample's own folder** when no 'Day N' folder
1351
+ is above it: `synth/pbmc/` and `synth/compensation/` are two groups, no
1352
+ longer both "synth", and the example data is no longer ".openflo". A
1353
+ generic 'Blank Experiment with…' folder still defers to the one above it.
1354
+ - **Samples are listed by name** once loading is done, not in the order
1355
+ they finished loading.
1356
+ - **Gates imported from a FlowJo workspace come in ticked**, like gates
1357
+ drawn in OpenFlo. They came in unticked.
1358
+ - **The Add FCS and Load CSV dialogs open in the folder last picked from,**
1359
+ else the current folder, else the home folder; never the installed
1360
+ package's own folder.
1361
+ - **Plots are drawn the way flow papers draw them.** Scatter (FSC, SSC)
1362
+ and Time open on a linear axis, so the cells no longer sit squeezed into
1363
+ one corner of a log plot. Fluorescence opens on logicle, so events at or
1364
+ below zero keep their own positions near 0 instead of piling up on the
1365
+ axis. Plots open in pseudocolor, coloured with Turbo, a rainbow that is
1366
+ close to perceptually uniform (View → Density colors switches back to the
1367
+ classic rainbow or to Viridis). A channel whose scale a session already
1368
+ saved keeps it.
1369
+ - **Logicle width and arcsinh cofactor can be set per axis.** The axis
1370
+ dialog (⚙ beside X and Y) offers Logicle, Arcsinh, Log and Linear, with
1371
+ logicle's width W and arcsinh's cofactor. These change only how the axis
1372
+ is drawn: gates select the same events, and the data is not
1373
+ re-transformed. The setting is saved with the session, its tabs and
1374
+ layout items.
1375
+ - **Gates show their name and frequency on the plot.** Each gate is drawn
1376
+ with a solid outline and labelled with its name (when it has one) and its
1377
+ percent of the parent, one decimal. Quadrants are labelled in their outer
1378
+ corners, and a population that cannot be computed reads NA. View → Gate
1379
+ labels on plots turns the labels off.
1380
+ - **One notation for axis ticks and percentages.** Every decade on an axis
1381
+ reads as a power of ten (10², 10³, not "100" next to "10³"). Linear axes
1382
+ read 50K and 100K. Frequencies in the Statistics table and the status bar
1383
+ have one decimal (93.0, not 93 or 92.96). A gate whose parent is empty
1384
+ reads NA instead of 0.00%.
1385
+ - **No window hides its buttons.** Statistics, Methods & provenance,
1386
+ Absolute counts, Preferences, FCS inspector, Expression, Frequencies,
1387
+ Calibration and Voltage optimization opened at fixed sizes too small for
1388
+ their content, and the button row (or Update, or Detect peaks) was simply
1389
+ not shown. Every window now opens big enough for all its controls (on a
1390
+ scaled display too), centred on the main window and on the screen; a
1391
+ window made smaller shrinks its table or plot, not its buttons.
1392
+ Preferences scrolls when the screen is too short and lists themes by
1393
+ name; the layout editor's toolbar wraps onto two rows when narrow.
1394
+ - **Statistics tells a parent from its child.** The Population column
1395
+ shows each gate's own name, indented under its parent, in a column as
1396
+ wide as the names; the exported CSV keeps the full path. Statistic
1397
+ columns are as wide as their headings.
1398
+ - **Sharp text on scaled Windows displays.** OpenFlo declares itself
1399
+ DPI aware, so at 125 % or 150 % Windows no longer stretches a blurry
1400
+ 96-DPI image of every window.
1401
+ - **One answer to "is this engine available?".** Help → Environment,
1402
+ Compare embeddings, the Cluster dialog and the Pipeline all test an
1403
+ engine by importing it (in a separate process, once per session), so a
1404
+ package installed without its own dependencies is shown as "fails to
1405
+ load", with the reason, everywhere, instead of being offered and then
1406
+ producing nothing.
1407
+ - **No stray untitled box over a window.** It was a hover tip left behind
1408
+ when its window closed, or was busy, under the pointer. Tips now close
1409
+ with their window and are not shown once the pointer has left.
1410
+
1411
+ ### Added
1412
+ - **Compensation can be applied to the whole acquisition batch.** Apply
1413
+ reached only the highlighted sample, so a set of files took one Apply
1414
+ each, and its tooltip said "every loaded sample". The Compensation window
1415
+ now has "Apply to: this sample / this batch". The batch is the samples
1416
+ acquired like the highlighted one, read from each file's keywords rather
1417
+ than its folder: the same instrument (`$CYTSN`, else `$CYT`), acquisition
1418
+ date (`$DATE`) and detector voltages (`$PnV`) on the matrix's channels,
1419
+ and the same stored spillover matrix when both files carry one. The
1420
+ window lists the batch, and hovering it says why each other sample was
1421
+ left out (other voltages, another date, no keywords). A file without
1422
+ these keywords is never guessed into a batch.
1423
+ - **Tabs in the Samples & Gates tray (beta).** A tab is a saved view: its
1424
+ samples and gates, their on/off states, and the plot's axes, type,
1425
+ colouring, axis types and zoom. Selecting a tab shows and plots only its
1426
+ members and restores its plot; "All" always holds every sample. + makes a
1427
+ tab from the selected rows (or empty); right-click a tab to rename,
1428
+ duplicate or delete it. In the tree's right-click menu, "Copy to tab" adds
1429
+ the selected samples or gates to a tab (a gate brings its ancestors and
1430
+ subtree), and "Apply to other tabs" (all, or chosen) pushes their on/off
1431
+ state to other tabs. "Within a tab (choose samples)…" applies it to only
1432
+ the samples you find and tick in one other tab, matching gates by
1433
+ population name. Samples and gates can also be dragged onto a tab on the
1434
+ strip: an arranged tab (samples from several groups, or its own order)
1435
+ holds them in a Presort zone at the top of its tree, which grows as needed
1436
+ and vanishes when empty; drag a row out of it to place it. Any other tab
1437
+ takes them straight to the top of its stack. Items dragged back from the
1438
+ Pipeline onto the tree or a tab join that tab. The main plot keeps your
1439
+ zoom through every redraw of the same view (a gate edit, a toggle, a tab
1440
+ switch) until Reset, and each tab keeps its own. Gates are shared: an edit shows in every tab holding
1441
+ that sample. The tabs pop out with the tray. The layout editor's "last
1442
+ displayed" view also reads the views saved in tabs.
1443
+ - **Sessions save the tabs and the layout (schema v2).** `.flowsession`
1444
+ gains `view_tabs` and `layout`; version-1 sessions upgrade on open (one
1445
+ All tab, no layout). A session saved now does not open in an older
1446
+ OpenFlo.
1447
+ - **Layout editor (beta).** The side panel's Layout button (was "to be
1448
+ added") opens a FlowJo-style editor of pages: plots of gated populations,
1449
+ stats tables (Count, %Parent, %Total and per-channel Median / Mean /
1450
+ GeoMean / CV / rCV), violin / box / bar charts across samples, and text
1451
+ boxes. Drag to move, drag the corner to resize, double-click to edit; an
1452
+ optional grid, and soft snapping to grid lines and to other items' edges,
1453
+ corners and centres, with guides. Samples and gates dragged from the
1454
+ Samples & Gates tree or the Pipeline (docked or popped out) land as plots,
1455
+ shown as that population was last displayed (Preferences can make it the
1456
+ current graph instead), or join the plot, table or chart they are dropped
1457
+ on, so several samples and populations overlay in one plot. A plot keeps
1458
+ the axis types and zoom it was taken with; when much of its data falls
1459
+ outside, a Zoom to fit badge offers (never applies) a refit. Cluster
1460
+ results (PhenoGraph, Leiden, FlowSOM) colour a plot by cluster and their
1461
+ populations can be plotted and tabulated. A View zoom (Fit, 50-200%,
1462
+ custom) sets the on-screen resolution; Export DPI defaults to 300 (150,
1463
+ 600, 1200 or custom). Pages can be added, renamed, duplicated, items copied
1464
+ or moved between them, and pages imported from or sent to another
1465
+ project's layout file. The layout re-applies to every sample, matching
1466
+ populations by name; Batch export… writes every page per sample. The
1467
+ mouse wheel scrolls the page (Ctrl+wheel zooms it, Shift+wheel scrolls
1468
+ sideways); over the selected plot it zooms that plot around the cursor.
1469
+ Holding the middle
1470
+ button drags its view, and right-click has Center zoom here, Zoom in,
1471
+ Zoom out and Zoom to fit; the view is kept with the plot. A cross-page
1472
+ host in the left margin keeps copies of items dragged onto it, to drag
1473
+ onto any page or place on every page; each is listed by number, kind,
1474
+ populations, axes and the page it came from, and can be renamed. Drag
1475
+ the host's edge to widen it; pointing at an entry outlines the items on
1476
+ the page that are still true duplicates of it (same populations, axes,
1477
+ zoom, colours...) and marks (◆) the tabs of other pages that hold one.
1478
+ Every item has a lasting id (P1, S2, C3, T4), shown in its corner on
1479
+ screen and in the host, so look-alike plots can be told apart. The toolbar reads left to
1480
+ right: insert, edit, the sample shown, Export (its dpi on the button, 300
1481
+ by default) and the Layout menu; the row below holds the page tabs, page
1482
+ size, grid, snap and the on-screen zoom. A plot dragged
1483
+ onto a stats table or chart supplies its samples (and its populations, if
1484
+ it has none). Each layer's colour can be set per item; a default colour
1485
+ list (layer 1 red, 2 blue, ...) fills in, and an override toggle (off by
1486
+ default) makes every layer use that list. A population that cannot be
1487
+ computed reads NA in tables, gets no bar (NA instead) and no violin, and
1488
+ its plot says so; a measured empty population still reads 0. Layouts
1489
+ save to `.oflayout` files (version 2; version-1 files open). Also under
1490
+ Tools → Layout editor (beta)….
1491
+
1492
+ ### Changed
1493
+ - **Saving and reopening sessions is much faster.** A session keeps each
1494
+ analysed sample's events beside it. They were CSV text, and formatting
1495
+ every value as text was most of a save: about 10 s per 300,000-event,
1496
+ 24-channel sample, holding the windows for the 5-minute autosave. They
1497
+ are now binary (`<session>_data/<sample>.npz`, every value and type kept
1498
+ exactly): four such samples save in 0.5 s instead of 44 s, reopen in
1499
+ 0.6 s instead of about 6 s, and take 58 MB each instead of 136 MB.
1500
+ Sessions with CSV sidecars still open, and are converted on their next
1501
+ save. A column the binary format can't hold (a nullable or categorical
1502
+ one) keeps that sample on CSV.
1503
+ - **The 5-minute autosave no longer freezes the windows.** It now writes on
1504
+ a background thread, as the close does; a close or Save started while it
1505
+ is writing waits for it.
1506
+ - **The first save after opening a session no longer rewrites every
1507
+ sample.** The sidecars the session was opened from are remembered, so the
1508
+ autosave and the close copy an unchanged one instead of encoding it again.
1509
+ - **Faster analysis with identical results.** Each of these gives the same
1510
+ numbers as before, bit for bit, checked against the previous code:
1511
+ - FlowSOM training (and CytoNorm, which uses it) runs compiled with numba,
1512
+ already installed with UMAP: about 14 s to 1.2 s per sample at the
1513
+ defaults.
1514
+ - The Statistics table splits large tables across CPU cores and computes
1515
+ only the statistics asked for: 31 s to 10 s for 2 million events, 20
1516
+ gates and 16 channels on 4 cores, faster with more.
1517
+ - Leiden and Louvain neighbour searches run on several cores: 7.8 s to
1518
+ 2.6 s for 50,000 events on 4 cores. Ties between equally near events
1519
+ resolve exactly as before.
1520
+ - Exporting a population as FCS converts the events in one step: 9.4 s to
1521
+ 1.5 s for 2 million events and 16 channels, with about 1 GB less memory,
1522
+ and a byte-identical file.
1523
+ - Integer FCS files that store fewer bits than their words hold (10-bit
1524
+ data in 16 bits, for example) load about 70x faster: 8.8 s to 0.12 s for
1525
+ 2 million events and 20 channels.
1526
+
1527
+ ### Removed
1528
+ - **Out-of-date developer files.** DESIGN_NOTES.md (last checked against the
1529
+ code in July, before most of the current interface) and five one-off
1530
+ scripts nothing used: four speed benchmarks and mutate_controls.py.
1531
+
1532
+ ### Fixed
1533
+ - **The public mirror's CI no longer fails its privacy gate.** That job
1534
+ and the release build's gate step ran scripts/publish_public.py, which is
1535
+ kept out of the mirror because it holds the private token list. Both
1536
+ now skip on the mirror; the gate still runs here before every snapshot
1537
+ and every PyPI upload.
1538
+ - **Compare embeddings shows Run and Cancel on a 150 % display.** The
1539
+ window took a fixed 420x360 and packed its buttons last, so larger text
1540
+ squeezed the button row to half height on Windows. It now takes the size
1541
+ its content asks for.
1542
+ - **New windows stay on a busy machine's screen.** On a loaded Windows
1543
+ PC a window could be fitted on opening and then put back where Windows
1544
+ first placed it, past the bottom of a laptop screen (seen with Plate
1545
+ view at 1366x768). The editor checks the place again just after.
1546
+ - **Help > Report a problem reaches the issue tracker.** It, and Help's
1547
+ documentation link, opened the private development repository, a 404 for
1548
+ everyone else. They open openflo-public now, and a test fails on any link
1549
+ to the private repository in what ships.
1550
+ - **FCS files with common writer quirks open.** A file whose $ENDDATA is
1551
+ one byte off, whose HEADER and TEXT disagree on where the data starts, or
1552
+ that chains a second data set with $NEXTDATA was refused and reported as
1553
+ "cut short" or "damaged". Each now loads (the first data set, for a
1554
+ chained file) with a load warning naming the quirk; the compensation
1555
+ readers and the FCS inspector read them too. A file that is really
1556
+ damaged still fails, with a message that names the quirk when one is the
1557
+ cause.
1558
+ - **Clustering and the FlowSOM tree repeat with their seed.** igraph has no
1559
+ number seed: `layout_fruchterman_reingold(seed=)` takes starting
1560
+ positions, so the tree layout was drawn differently every time. It now
1561
+ runs on a generator seeded from `seed`. Louvain left igraph on a fresh,
1562
+ unseeded generator after each run, so `random.seed()` stopped controlling
1563
+ igraph for the rest of the session; it now hands back the default.
1564
+ PhenoGraph's seeded Leiden backend stays the default, as before.
1565
+ - **Leiden noise reads as noise.** Its unassigned bucket imported as
1566
+ "leiden -1" in a palette colour and the Populations menu left Leiden out;
1567
+ it is grey and named "Unassigned (noise)", and the menu lists it (other
1568
+ Leiden populations keep their names, such as "leiden 3").
1569
+ - **`python -m openflo.inspect_fcs` lists the channel keywords** again
1570
+ (flowio 1.x returns them lower-case, without the `$`).
1571
+ - **The update check never offers a beta to someone on a final release**,
1572
+ and without the `packaging` module a beta no longer sorts above its own
1573
+ final release.
1574
+ - **The README's licence note** said OpenFlo does not import igraph and
1575
+ leidenalg (both GPL) directly; it does, and the note says so.
1576
+ - **The Cluster dialog starts FlowSOM at 20 metaclusters, and says why.**
1577
+ Asked for exactly as many metaclusters as there are populations, FlowSOM
1578
+ merged a rare (1.5%) population into a neighbour (F1 0.22; 0.99 with 12).
1579
+ The dialog now starts at 20 (was 10), notes "set above the populations
1580
+ you expect" beside the field, and its hover tip explains the reason and
1581
+ how to join extra metaclusters with an OR gate. **Behaviour change:**
1582
+ FlowSOM runs from the dialog at its default now make 20 metaclusters, not
1583
+ 10. The CLI and the Python API keep 10.
1584
+ - **A 30-channel recovery test for every clustering method.** Eight planted
1585
+ populations in a lineage tree, one of them rare (1.5%): PhenoGraph, Leiden
1586
+ and Louvain must recover all eight (ARI >= 0.95, each population F1 >=
1587
+ 0.9), FlowSOM the seven common ones, and the rare one when asked for more
1588
+ metaclusters than populations. The same checks must fail on data with no
1589
+ structure and on random labels.
1590
+ - **macOS is tested on the public mirror.** A macOS job in ci.yml runs only
1591
+ on openflo-public, where GitHub's macOS runners are free (they bill at ten
1592
+ times Linux on the private repository), so each mirror refresh runs the
1593
+ suite with the GUI on macOS.
1594
+ - **The PyPI release waits for the tests.** A version tag ran only the type
1595
+ check before building and publishing; it now runs the whole CI workflow
1596
+ (tests with the GUI, type check, privacy gate, install checks) first.
1597
+ - **A status message stays put after the gate tree rebuilds.** On Windows,
1598
+ adding FMO gates (or an FMO refusal) could have its message replaced by the
1599
+ selected gate's count, because the tree re-selecting its own row was taken
1600
+ for a click.
1601
+ - **The editor opens on Tk 9.** The drag-and-drop package pinned at 0.4.3
1602
+ stopped OpenFlo from starting on any Python that ships Tk 9 (uv's 3.14
1603
+ and 3.15 builds, python.org's macOS installer from 3.14.5). It is now
1604
+ 0.6.4, and if drag-and-drop still can't load, the editor opens without it
1605
+ instead of not at all.
1606
+ - **Identical groups keep p = 1 with SciPy 1.18.** SciPy 1.18 returns NaN
1607
+ for a Mann-Whitney test where every value is the same, so a cluster or
1608
+ population with the same share in both groups read NA in the comparison
1609
+ CSV, differential abundance and group statistics. It is 1 again on every
1610
+ SciPy.
1611
+ - **Cluster populations of a sample reopened without its clustering read
1612
+ NA, not 0.** A session whose processed data was not found reloads that
1613
+ sample from its FCS file, without the cluster column: every cluster read
1614
+ 0 events under a group holding all of them. They now read NA, and the
1615
+ load says which results were not found.
1616
+ - **A session keeps its previous save.** Saving over a session keeps the
1617
+ earlier version as `name.backup.flowsession`. A session that can't be read
1618
+ says so in plain words and offers that backup, and resuming after a crash
1619
+ skips a damaged autosave for the newest one that reads.
1620
+ - **Save and close updates the session you opened.** It used to write only
1621
+ the autosave, so the file you opened was left as it was at your last
1622
+ save.
1623
+ - **Problems found while loading are shown.** A renamed channel, an event
1624
+ count that disagrees with the data, a spillover matrix that could not be
1625
+ used, or a workspace population that was not imported went only to the
1626
+ log or to a status line the next sample overwrote. They now appear
1627
+ together in one "Notes from loading" window and are kept in the history.
1628
+ - **Damaged or unusual FCS files.** A parameter without `$PnN` is named
1629
+ `Ch<n>`; one without `$PnR` takes the range its `$PnB` allows (as FlowJo
1630
+ does); two channels with one name load as `FL1-A` and `FL1-A (2)`; and a
1631
+ cut-short or non-FCS file says so in plain words instead of showing a
1632
+ parser error.
1633
+ - **Spillover keywords that name markers.** A `$SPILLOVER` listing marker
1634
+ names (`$PnS`) instead of detectors is matched to the detectors when each
1635
+ marker belongs to exactly one, instead of leaving the sample
1636
+ uncompensated. Values past the end of the matrix are reported.
1637
+ - **CSV files in common layouts.** Load CSV detects semicolon, tab and bar
1638
+ separators and decimal commas, skips `#` comment lines at the top, keeps
1639
+ the first event of a file with no header row (columns `Ch1`, `Ch2`, ...),
1640
+ and refuses a file with no numeric column in plain words.
1641
+ - **Undo after compensation or a transform change.** Undo can't reverse
1642
+ those, and it used to undo the gate change made before them without a
1643
+ word. It now says it can't, and the next Undo goes on to the gate change.
1644
+ Undo no longer brings back the gates of a removed sample.
1645
+ - **Exported FCS files keep their keywords.** Every keyword from the source
1646
+ that does not describe the file's layout is carried over (`$SRC`, `$OP`,
1647
+ tube and well names and more; the source's `$FIL` as
1648
+ `OPENFLO_SOURCE_FIL`), and the matrix the values were compensated with is
1649
+ recorded as `OPENFLO_APPLIED_MATRIX`. Keywords about compensation are not
1650
+ carried, so no reader compensates the values twice.
1651
+ - **A session sample whose FCS file is missing is no longer deleted by the
1652
+ next save.** Opening a session skipped a sample whose file had moved, and
1653
+ the save or autosave after it wrote the session without it and its gates.
1654
+ The sample is now kept offline and saved back unchanged, and after opening
1655
+ OpenFlo offers to look for the missing files in a folder (and its
1656
+ subfolders) and relinks them by file name.
1657
+ - **A session's saved data cut short no longer loads part of a sample.** The
1658
+ per-sample data beside a session is written whole or not at all, and one
1659
+ that is damaged or holds fewer rows than recorded is rebuilt from the
1660
+ sample's FCS file, with a warning.
1661
+ - **Log-amplified and gain-adjusted FCS files load as intensities**
1662
+ (behaviour change). Integer data with `$PnE` log amplification was read as
1663
+ raw channel numbers and `$PnG` was ignored; both are now applied on load
1664
+ and recorded in the history. A session holding such a file now reads it
1665
+ as intensities. Float data is left as stored.
1666
+ - **A gate on a channel a sample lacks reads NA with the reason**
1667
+ (behaviour change). It used to admit every event, so the population read
1668
+ 100%. It now admits none and Statistics, the report and the tree show NA
1669
+ ("channel X is not in this sample").
1670
+ - **FlowJo AND / OR / NOT gates are imported from and written to .wsp.**
1671
+ They were dropped on import without a word, with everything under them.
1672
+ An operand that can't be found reads NA with a warning.
1673
+ - **Pipeline (beta) says when an embedding can't run.** TriMap was ticked by
1674
+ default, and without the `[embed]` extra it was skipped with nothing but a
1675
+ log line: the run ended "Done: 4 ok" with no TriMap output. PHATE, TriMap
1676
+ and PaCMAP are now unticked and disabled in the panel when not installed
1677
+ (the tooltip says how to install them), and a run that asks for one anyway,
1678
+ or whose embedding fails or makes no plot, lists it in the run's summary.
1679
+ - **The highlighted sample is always one on the plot.** After loading
1680
+ several files the first was plotted but the last to arrive was
1681
+ highlighted, so a gate drawn with "all shown" went to both and
1682
+ compensation went to a sample not on screen. Now the plotted sample is the
1683
+ highlighted one after a load, highlighting an unticked sample ticks it,
1684
+ clicking a ☑/☐ box no longer moves the highlight, and unticking the
1685
+ highlighted sample moves the highlight to one still plotted.
1686
+ - **The Copy gates dialog shows its Copy button.** Its four buttons sat in
1687
+ one row about 560 px wide inside a 420 px dialog, so Copy, packed last,
1688
+ got no room and only appeared if you widened the window. Select all /
1689
+ Deselect all and Copy / Cancel now sit on two rows of a 460 px dialog.
1690
+ - **Copying gates no longer stacks duplicates.** Copying appended every
1691
+ gate, so a sample that already held it (copied before, or drawn there with
1692
+ "all shown") got a second, identical-looking row. A gate the target
1693
+ already has is now reused; the same population drawn differently (same
1694
+ kind, axes and name, under the same parent, and only one such) takes the
1695
+ copied geometry in place, keeping its id, on/off state and children; the
1696
+ status line counts both. Anything less certain is appended as before.
1697
+ - **A FlowJo export names each unnamed gate the same in every sample.**
1698
+ An unnamed gate was written under the editor's per-sample gate id, so
1699
+ the same gate could be `g2` in one sample and `g3` in the next (any
1700
+ sample that had once had a gate deleted numbered on from there). FlowJo
1701
+ matches populations across samples by name, so its tables and group gates
1702
+ didn't line up. An unnamed gate is now named for its axes (`FSC-A x
1703
+ SSC-A`, or the channel of a 1-D gate), and siblings with the same name get
1704
+ `(2)`, `(3)`. Before: 6-sample export, populations `g1`/`g2`/`g3` with
1705
+ ctrl_d1 off by one. After: `FSC-A x SSC-A` / `FL1-A x FL2-A` in all six.
1706
+ - **Closing a large session no longer freezes OpenFlo.** "Save and close"
1707
+ saved on the window thread, rewriting every sample's processed-data CSV:
1708
+ about a minute for a large session, with every window marked "Not
1709
+ Responding" and no message, and each extra click on ✕ queued another full
1710
+ save. The save now runs in the background behind a "Saving session… n of
1711
+ N samples" window, extra clicks only bring that window forward, and a
1712
+ sample's CSV is rewritten only when its data changed since the last save,
1713
+ which also makes the 5-minute autosave quick when nothing changed.
1714
+ - **Closing asks whether to save the session.** ✕ autosaved every time,
1715
+ and a large session (every sample's data rewritten) froze every window
1716
+ for up to a minute with no message, so the windows looked as if they
1717
+ would not close; each further ✕ queued another save. It now asks: Save
1718
+ and close, Close without saving, or Cancel. A second ✕ while closing is
1719
+ ignored. (A stop-gap: making the save itself fast is its own change.)
1720
+ - **The update offer asks Update or Cancel.** It asked Yes / No / Cancel,
1721
+ and No opened the release page: three ways to say yes. The release notes
1722
+ link is now in the question's text.
1723
+ - **Export to FlowJo keeps gates on a channel of unknown scale.** A column
1724
+ of an older OpenFlo CSV with no transform record has no transform to undo,
1725
+ so its gates reached FlowJo in display units with only a log line (an
1726
+ asinh cut at 4.0 became an intensity of 4.0). When the sample's raw FCS
1727
+ can be read, the channel is now matched to it rank for rank, as FlowJo
1728
+ will gate it, and the gate keeps its events (a cut exactly, a polygon to
1729
+ within its re-traced edges). Without the FCS the gate is still written,
1730
+ as stored. Either way a warning after the export names each such gate.
1731
+ - **A sample on the same transform with other settings is put on the
1732
+ editor's.** A library export on asinh cofactor 5 loaded beside samples on
1733
+ the editor's asinh (cofactor 150) kept cofactor 5: the same events plotted
1734
+ at median 4.45 beside 1.16, its axis labels were wrong, and the Expression
1735
+ window pooled both scales into one violin. The parameters are now compared,
1736
+ not just the method, for the first sample too. A restored session's gates
1737
+ move with the data: a cut drawn on cofactor 5 held 380 of 800 events, and
1738
+ without converting it held none after reopening.
1739
+ - **A population that cannot be computed reads NA, not 0.** A gate whose
1740
+ region is unknown (a FlowJo ellipse not converted, a boolean that lost an
1741
+ operand, an ellipse whose covariance describes no ellipse) selects
1742
+ nothing, and so does every gate under it or built on it. Statistics,
1743
+ Frequencies and its comparisons, the report and the gate's status line
1744
+ showed that as a measured 0 events. They now show NA with the reason,
1745
+ Frequencies leaves it out as missing, and the tidy CSV writes NA.
1746
+ - NOT or OR over such an ellipse now admits no events too; NOT over it
1747
+ admitted every event.
1748
+ - **A FlowJo gate imported onto a channel of unknown scale reads NA.** Its
1749
+ coordinates are in FlowJo's linear units and cannot be converted to a
1750
+ channel whose display scale is unknown (an older OpenFlo CSV with no
1751
+ transform record). It was kept as is, comparing a linear cut against
1752
+ values on another scale, with only a log warning. It is now kept but
1753
+ selects nothing, and its population, everything under it and every
1754
+ boolean built on it read NA with the reason.
1755
+ - **`openflo-run`'s shared clustering and comparison files.**
1756
+ - A run with one group removes an earlier two-group run's comparison
1757
+ files (`compare_<A>_vs_<B>.csv`/`.png`, `compare_shared_clusters.csv`)
1758
+ and says so, instead of leaving them as if it had made them. A file
1759
+ that cannot be removed is named and the run exits 1.
1760
+ - Channel labels match across samples despite common real spellings: a
1761
+ fluor name at either end ('CD3 FL5' on FL3-A), Greek letters
1762
+ ('CD8α' = 'CD8a') and detector names that differ only in case
1763
+ ('FL7-A' / 'fl7-a'). Such channels were dropped before.
1764
+ - Pooled and unpooled events follow one noise rule: each finite event
1765
+ takes the label of its nearest labelled pooled event, so a sample's
1766
+ "Unclustered (noise)" share no longer scales with how much of it was
1767
+ pooled. Only rows with a non-finite marker value stay noise.
1768
+ - The shared PhenoGraph's RAM check applies with the default
1769
+ `--workers 1` too; with too little free RAM it runs on one core and
1770
+ says why.
1771
+ - **The compensation editor's Apply keeps what was computed on the
1772
+ sample.** Apply replaced the sample's data with a fresh load of its FCS.
1773
+ Clusters, embeddings, U: abundances and _pos flags were lost, so their
1774
+ populations selected no events and gates on them selected every event. A
1775
+ downsampled sample came back with all its events.
1776
+ - Apply now re-compensates the sample's own events in place: every event
1777
+ and column stays, and each channel keeps its transform.
1778
+ - The status line names what was computed before the change and asks
1779
+ you to re-run it.
1780
+ - A sample opened from a CSV with no record of its compensation and no
1781
+ FCS behind it is refused, with the reason. It used to fail with "could
1782
+ not parse FCS".
1783
+ - **A session with Leiden or FlowSOM populations can be saved.** Saving,
1784
+ and the autosaves, failed with "Object of type int64 is not JSON
1785
+ serializable".
1786
+ - **Gates that cannot be evaluated select nothing, and say so.**
1787
+ - A boolean gate over a boolean that lost an operand (or over a gate
1788
+ under one) now admits no events. With the inner boolean empty, NOT over
1789
+ it admitted every event, and OR with it was the other operand alone.
1790
+ Clear, Cut, session restore, .wsp ingest and template apply now also
1791
+ name the booleans built on a broken one.
1792
+ - An ellipse whose covariance is singular or not positive definite
1793
+ admits no events instead of every event. Exported to FlowJo it is still
1794
+ the all-pass rectangle, and the warning says FlowJo will count every
1795
+ event.
1796
+ - **Open-sided gates are drawn in full on every view.** A quadrant, an
1797
+ imported max-only range, or a polygon opened at an axis end drew only part
1798
+ of its outline on a fluorescence channel's log and symlog views (no
1799
+ outline at all on symlog). An open interval bound no longer draws a line
1800
+ or stretches a linear view to ±1e12.
1801
+ - **Editing overlapping polygons grabs the nearest vertex.** A click on one
1802
+ polygon's corner took the first-drawn polygon's corner ~9 px away; the
1803
+ nearest vertex (then the nearest edge) of all polygons now wins.
1804
+ - **FMO gates are placed at the FMO's linear cutoff.** The FMO dialog took
1805
+ the percentile in the FMO tube's own display coordinates and used it as a
1806
+ gate value on the stained sample. An FMO on asinh cofactor 5 beside a
1807
+ stained sample on cofactor 150 called 5.36% positive where 50.52% is
1808
+ right; an FMO of unknown scale gave 0.0%.
1809
+ - The cutoff is the FMO's percentile on its linear values, mapped into the
1810
+ stained sample's display coordinates. With a parent gate selected, the
1811
+ FMO is read through it in the stained sample's coordinates.
1812
+ - An FMO or stained channel of unknown scale is refused, and the status
1813
+ names the channel, the FMO and why.
1814
+ - **A logicle scale inferred from the value range is marked as a guess.**
1815
+ A column of an older OpenFlo CSV with no transform record, whose values
1816
+ all lie in the logicle range, was taken as logicle. A dim channel on asinh
1817
+ falls in that range too: Statistics reported a linear median of 1793.94
1818
+ where 80.56 is right.
1819
+ - The entry is kept as logicle but marked guessed, through re-saves,
1820
+ session restore, re-transforms and pooled Workspace runs.
1821
+ - Its statistics are titled "(linear, guessed logicle)" in the Statistics
1822
+ window and its CSV. Expression and Group comparison name the samples
1823
+ whose medians rest on a guess.
1824
+ - **`marker_expression` names the samples it could not use.** A sample of
1825
+ unknown scale went into `differential_test` as NaN and was dropped
1826
+ without a word. The groups now list omitted and guessed samples, and each
1827
+ result row carries `omitted_a/b` and `guessed_a/b`.
1828
+ - **Compensation-control names.** The rule took any word ending in "comp"
1829
+ ("recomp", "precomp", "Decomp_1", "Autocomp", "TComp_1") as a compensation
1830
+ control. It now names the bead brands written as one word (UltraComp,
1831
+ OneComp, VersaComp, AbC, ArC). "Non-comp" and "No-comp" are samples.
1832
+
1833
+ ## [2.6.2] - 2026-10-09
1834
+
1835
+ ### Changed
1836
+ - **Example markers and cutoffs are generic too.** The example panel,
1837
+ synthetic data, CLI defaults, docs and tests now use the markers `CD901b`,
1838
+ `CD902` and `CD903` (they still parse as CD markers) and round example
1839
+ cutoffs (0.5 / 0.4 / 0.45). The privacy gate blocks the real markers.
1840
+ - **Examples, fixtures and docs use generic detector names.** Channels in
1841
+ the example panel, synthetic data, CLI defaults, README, tests and golden
1842
+ fixtures are now `FL1-A`, `FL2-A`, ... instead of real fluorochromes, and
1843
+ the comp-control matcher tests use generic dyes (`FLB`, `FLB-Cy7`, ...)
1844
+ with the same tandem structure. The privacy gate now blocks the three
1845
+ most common real ones, except in a functional fluor catalog
1846
+ (`src/openflo/fluor_catalog.py`), which names dyes without revealing a
1847
+ panel. Note: a session or template saved against the old example channel
1848
+ names will not match the renamed example panel.
1849
+ - **Repo tidy for the beta.** Removed four empty files at the repo root
1850
+ (`UMAP`, `data`, `events`, `i'`) left by stray shell redirects, and
1851
+ CONTRIBUTING.md now points to `src/openflo/pipeline.py` for
1852
+ `OpenFloError` instead of the nonexistent `flow_pipeline.py`.
1853
+ - **The privacy gate now runs on every PR and before every PyPI upload.**
1854
+ It used to run only when the public mirror was refreshed, and the sdist
1855
+ is built from this repo, so nothing scrubbed what PyPI would get. That
1856
+ gap was real: the sdist shipped `scripts/publish_public.py`, whose
1857
+ pattern list names the private tokens the gate exists to block.
1858
+ MANIFEST.in now excludes it, `publish_public.py` gains `--check` and
1859
+ `--check-dist DIR`, and CI's new `privacy gate` job and the release
1860
+ workflow run them.
1861
+ - **The side panel is compact, and the plot's controls fit the default
1862
+ window.** At 1500×840 the Samples & Gates panel took 747px, half the
1863
+ window, and the plot's control bar was cut off ("Color" showed one
1864
+ letter).
1865
+ - The panel is now 372px and the plot pane 1116px.
1866
+ - The action buttons use tighter padding and sit in three columns.
1867
+ Undo and Redo share one cell.
1868
+ - The Pipeline button reads "Pipeline (beta)".
1869
+ - A "Layout (to be added)" button marks the place of a FlowJo-style
1870
+ layout editor, which is not built yet. Clicking it says so and offers
1871
+ File → Figure layout…, which builds multi-panel figures from preset
1872
+ arrangements.
1873
+ - Showing the log/console no longer widens the panel.
1874
+ - The tree's "All" heading is no longer clipped.
1875
+ - **Acquisition QC on load is off by default.** It is a new Preferences
1876
+ option: Loading → "Run acquisition QC on load". QC removes events from
1877
+ time bins where the signal drifts or the flow rate jumps, and events piled
1878
+ at a channel's ceiling, before any gate. FlowJo has no such step, so with
1879
+ QC on, the same gates count fewer events in OpenFlo than in FlowJo.
1880
+ - Turning the option on applies to files loaded afterwards.
1881
+ - A sample keeps the choice it was loaded with when it is re-compensated,
1882
+ reopened from a session, moved to another window or reloaded by the
1883
+ Workspace.
1884
+ - Sessions saved before this change reload their samples with QC, as they
1885
+ were loaded.
1886
+ - `openflo-run` is unchanged.
1887
+ - **The type check gives the same answer locally and in CI** (development
1888
+ only; nothing a user runs changes).
1889
+ - CI and the pre-push hook now run the same `scripts/typecheck.py`, with
1890
+ the same packages: CI also installs the optional backends
1891
+ (`requirements-typecheck.txt`), which it used to leave unresolved.
1892
+ - The script refuses to run (exit 2) when pyright would read a different
1893
+ environment than CI: wrong package versions, a different pyright, or
1894
+ pyright resolving imports from another interpreter.
1895
+ - In a git worktree it uses the main checkout's `.venv`; no junction needed.
1896
+ - A release tag is built only after the same type check passes.
1897
+
1898
+ ### Fixed
1899
+ - **Reference spectra of dim dyes on autofluorescent cells.** The positive
1900
+ population of a dim single-stain control was picked by total brightness.
1901
+ It then included bright-AF cells, which mixed autofluorescence into the
1902
+ dye's spectrum. Example: a dye at 400 on cells where 30% have 10x AF read
1903
+ cosine 0.53–0.98 to its true spectrum, and unmixing error was 0.38–0.72.
1904
+ - Positives are now picked by the dye's own excess in its primary
1905
+ detector. Each event's autofluorescence is subtracted using its nearest
1906
+ unstained neighbours in the detectors where AF dominates. The same case
1907
+ now reads 0.9999 and unmixing error 0.044–0.061, which matches the
1908
+ true-spectrum floor.
1909
+ - Guardrail: the old estimate is kept unless the two estimates disagree
1910
+ (cosine < 0.99) or the control is poorly separated. Over a 576-cell
1911
+ simulation grid, no cell got worse than before. Every bright dye
1912
+ (≥10k) and every control without AF variation gives a bit-identical
1913
+ result.
1914
+ - Each spectrum now says how it was built. Controls too dim to trust are
1915
+ named with a recommendation (beads or a brighter fluor), in the CLI
1916
+ report, the editor status bar and the Spectral QC window.
1917
+ - New `method='auto'|'total'|'matched'` and `beads=` options, and the CLI
1918
+ flags `--unmix-spectra` and `--unmix-beads`.
1919
+ - **Selecting a sample or gate on large files is fast.** At 20M events,
1920
+ clicking a sample row took 1.4 s and clicking a gate row 13 s. Every replot
1921
+ shuffled all events to pick the drawn subsample, re-scanned them for
1922
+ missing values, and recounted each gate's chain twice.
1923
+ - Now 0.2 s for either once a sample has been prepared. The first view of
1924
+ a sample takes 0.4 s, and 1.8 s for a gate row.
1925
+ - Filter mode went from 8.1 s to 0.18 s, Highlight from 3.0 s to 0.27 s,
1926
+ and Histogram from 1.1 s to 0.17 s.
1927
+ - After a change, the displayed sample is drawn first. The other loaded
1928
+ samples are prepared in the background, at about 1.7 s per 20M sample,
1929
+ without blocking the window.
1930
+ - No count, percentage or histogram value changes. The drawn subsample is
1931
+ still fixed per sample, axes and cap, and now also across restarts.
1932
+ - Caches are dropped whenever a sample's data or a gate on its chain
1933
+ changes, and are bounded at 192 MB.
1934
+ - **Gate bounds are shown as intensities, not stored logicle values.** A cut
1935
+ at an intensity of about 1,000 read `CD3-A >= 0.453` and now reads
1936
+ `CD3-A >= 987`, in the same style as the axis ticks.
1937
+ - Applies to the gate list, unnamed population names in Statistics,
1938
+ Frequencies and their CSVs, the histogram slider, the auto-threshold
1939
+ status, the backgate legend, the Expression, FMO and Workspace windows,
1940
+ and the `openflo-run` log.
1941
+ - Open sides are said as open (`CD3-A < 2.5K`) instead of printing the
1942
+ ±1e12 sentinel. A channel of unknown scale keeps its stored value,
1943
+ marked "(stored, scale unknown)".
1944
+ - Colour bars and the CLI scatter PNGs are ticked at round intensities
1945
+ (`0, 100, 10^3, 10^4`) instead of `0.2 … 1.0`.
1946
+ - Populations are still matched across samples by gate structure, so
1947
+ Frequencies is unaffected by the new names.
1948
+ - **Frequencies: a population missing from a sample is missing, not 0.**
1949
+ - Populations were matched across samples by their displayed path. An
1950
+ unnamed gate's path includes its coordinates, so a cut nudged in one
1951
+ sample, or a marker on another detector, split one population into
1952
+ several.
1953
+ - Differential abundance then counted each gap as 0 events. A gate drawn
1954
+ only in one group's samples read as a population that had vanished from
1955
+ the other group.
1956
+ - Populations now match by gate name, or by kind of gate, markers and
1957
+ position among siblings.
1958
+ - A sample that still lacks a population shows it as missing in the
1959
+ table, the tests, the overview and the exports. Differential abundance
1960
+ fits it on the samples that have it and reports `n_missing`.
1961
+ - %Parent of an empty parent is NaN, not 0%.
1962
+ - **Compensation is applied once, with each sample's own matrix.** Reported
1963
+ as "either ignored or applied twice"; both happened.
1964
+ - **Applied twice.** A second compensation call (`auto_compensate` then
1965
+ `compensate_from_wsp`, the same call twice, or `FMOGater.prepare` run
1966
+ twice) compensated values that were already compensated: a channel with
1967
+ 0.30 spillover moved by about 15,000 units. A new matrix now replaces the
1968
+ one applied before. Compensating a channel that is already transformed
1969
+ raises `CompensationError` instead of mixing logicle values.
1970
+ - **Wrong matrix or none.** `compensate_from_wsp` applied the workspace's
1971
+ first matrix to every file, and raised on a workspace with no matrix.
1972
+ That affected Compare workspace, `FMOGater` and `Experiment.run_all`. It
1973
+ now uses the matrix inside the file's own `<Sample>`, or the file's
1974
+ `$SPILL` (FlowJo's Acquisition-defined) when the workspace gives it none.
1975
+ Several same-named matrices are no longer collapsed into the last one.
1976
+ - **Editor.** Opening a `.wsp` applied each FCS's `$SPILL` instead of the
1977
+ workspace's matrix for that sample. A reopened session, a sample moved
1978
+ to another window and a workspace re-run silently went back to `$SPILL`,
1979
+ discarding a matrix the user had applied. Each now uses the matrix the
1980
+ sample was compensated with. A sample restored from a processed CSV keeps
1981
+ its matrix and the link to its FCS.
1982
+ - **FlowJo gate coordinates are converted to and from OpenFlo's display
1983
+ scale.** A `.wsp` holds gates in linear intensity, while the editor and
1984
+ `openflo-run` keep gates in the stored (e.g. logicle) coordinates. Nothing
1985
+ converted between them:
1986
+ - an imported `FL1-A > 1000` selected 0 events where FlowJo counts 7,653;
1987
+ - an exported logicle threshold of 0.454 reached FlowJo as an intensity of
1988
+ 0.454, so FlowJo counted 14,087 events where OpenFlo counts 7,653.
1989
+
1990
+ Import (workspace and template) and export (editor and `--export-wsp`) now
1991
+ convert per channel. Bounds convert exactly. Polygon edges are subdivided,
1992
+ and ellipses are traced as polygons. A channel of unknown scale is left
1993
+ unconverted, with a warning. Importing also keeps sibling gates on one
1994
+ channel: one workspace's 5 gates arrived as 3, with a child moved under
1995
+ the wrong parent.
1996
+ - **GeoMean and rCV follow FlowJo's definitions.** These change the
1997
+ statistics table and its CSV.
1998
+ - rCV is now 100 × ½ (P84.13 − P15.87) / median. The MAD form used before
1999
+ read 74% where FlowJo reports 89% on a skewed population.
2000
+ - GeoMean on a transformed channel is the mean in display space,
2001
+ back-transformed. It used to drop every non-positive event, so a dim
2002
+ population read 212 where FlowJo reports 84.
2003
+ - **Differential abundance is calibrated.** The negative-binomial test gave
2004
+ a false positive in 24% of no-difference 3-vs-3 comparisons at the 5%
2005
+ level, and in 64–84% for nested gates. It solved the Poisson score and
2006
+ used one dispersion for every population, and nested parent gates pulled
2007
+ that dispersion to almost zero. Each population now has its own NB fit
2008
+ and dispersion, with a likelihood-ratio test on an F(1, df) reference.
2009
+ Simulations give 5–6% false positives, and rows gain a `df` column.
2010
+ One sample per group now gives no p-value.
2011
+ - **One fold-change direction: log2(B / A) everywhere.** `differential_test`
2012
+ alone gave A over B, so one comparison showed opposite signs in two
2013
+ windows. A group mean of 0 now gives ±inf, not ±29.2 from a 1e-9
2014
+ pseudo-count.
2015
+ - **Absolute counts use the bead and sample volumes and the dilution.**
2016
+ 50 µL of beads at 1,000/µL in 100 µL of blood read 2,000 cells/µL where
2017
+ 1,000 is right. With no volumes entered, the result is unchanged.
2018
+ - **`compare_conditions` compares populations, not cluster numbers.**
2019
+ Samples clustered one by one number their clusters independently, so
2020
+ cluster 0 could be CD4 cells in one sample and CD8 cells in another: a
2021
+ CD4 split of 80% vs 20% read 80% vs 80%. The compared samples are now
2022
+ clustered together first. `openflo-run` already did this.
2023
+ - **Compensation matrix files.**
2024
+ - A `$SPILL` that names its channels by parameter number now compensates.
2025
+ Before, nothing matched and the sample stayed uncompensated.
2026
+ - A matrix in percent (diagonal ≈100) is converted, with a warning, from
2027
+ any source. Before, it left every channel at 0.01×.
2028
+ - CSV rows are matched to the header by their labels.
2029
+ - Header-only CSVs from FlowJo now read.
2030
+ - A `.wsp` with no matrix returns `(None, None)`, as documented.
2031
+ - **Optimize compensation is unbiased.** It now uses the difference of the
2032
+ positive and negative medians and drops saturated events. On bead peaks
2033
+ the estimate was 0.158 for a true 0.20, and 0.63 with 3% of events
2034
+ saturated; it is now 0.200.
2035
+ - **Compensation QC** ranks spillover by magnitude relative to each source's
2036
+ own signal. A −0.35 entry ranked below +0.04.
2037
+ - **Spectral unmixing.**
2038
+ - The editor unmixes what the detectors measured, i.e. the values before
2039
+ compensation and transform. It used to read compensated logicle values
2040
+ (median abundance error 36–54%).
2041
+ - The spillover spread matrix uses the standard formula, net of the
2042
+ negative population. It overstated spread 4–6× when the control was
2043
+ mostly negative, and understated it at 0.58× with non-negative
2044
+ unmixing.
2045
+ - `run_qc` drops the same events from `raw`, so `--unmix` no longer
2046
+ unmixes events that QC removed.
2047
+ - **Values reported on the linear scale.** Several outputs carried logicle
2048
+ coordinates as intensities:
2049
+ - CLI `<sample>_stats.csv` and `compare_shared_clusters.csv` medians read
2050
+ 0.43 / 0.75 where 796 / 20,104 is right;
2051
+ - the trajectory trends export;
2052
+ - the AnnData export, which keeps X and adds a `linear` layer and each
2053
+ marker's transform;
2054
+ - the FMO and auto-gate status lines, which now give the linear cut;
2055
+ - the heatmaps, which now say they are on the display scale.
2056
+ - **GPU logicle** no longer clamps values above ~7.4×10⁵ (error 0.16 →
2057
+ 2×10⁻⁸).
2058
+ - **Scatter, Time and instrument channels are recognised in any case**
2059
+ (`TIME`, `fsc-a`, Beckman `FS INT`/`SS INT`, Sony `BSC-A`, CyTOF
2060
+ `Event_length`…). They were logicle-transformed and clustered as markers.
2061
+ - **Switching a channel to 'log'** now asks first and names how many events
2062
+ at or below zero it will empty.
2063
+ - **Gating.**
2064
+ - A threshold counts `x >= value`, as Gating-ML and FlowJo do. On integer
2065
+ data events exactly on the cut were lost: 50 vs FlowJo's 60. This
2066
+ includes FMO `_pos` flags.
2067
+ - A gate drawn to the end of an axis is open there, so the negative
2068
+ events piled on the axis floor are inside. A rectangle from the corner
2069
+ counted 271 of 2,543 events.
2070
+ - Axis ticks and the range dialog are in intensity, not stored logicle
2071
+ units.
2072
+ - A `.wsp` template applies one sample's gate tree, not every sample's.
2073
+ - **Compare workspace.**
2074
+ - A gate on a bare (uncompensated) parameter is counted on uncompensated
2075
+ values, as FlowJo does: 5,216 → 8,413.
2076
+ - A population under a gate OpenFlo cannot read is reported as an error.
2077
+ It used to be re-rooted and counted.
2078
+ - **Cancel stops a Workspace run's whole process tree on Linux and macOS.**
2079
+ Before, it stopped only the job process, so PhenoGraph's worker pool and
2080
+ the Louvain binaries kept running. Windows already killed the tree.
2081
+ - **Events removed by the editor's load QC are reported** in the load status
2082
+ and the `.wsp` export warning. FlowJo has no such step, so its counts
2083
+ include them.
2084
+ - **Ellipses drawn in FlowJo are imported.** FlowJo stores such an ellipse
2085
+ as foci and edge points in its display space. The import (workspace,
2086
+ template, Compare workspace) now converts it with the sample's own axes:
2087
+ exactly on linear axes, and as a 64-vertex polygon on logicle and biex axes
2088
+ (FlowJo's biex lookup table). Before, it was dropped and each population
2089
+ under it counted every event the ellipse excluded.
2090
+ - An ellipse on any other axis (log, arcsinh, a linear gain other than 1)
2091
+ becomes a new gate kind, `flowjo_ellipse`. It counts no events, its
2092
+ populations stay under it, the gate list and status bar flag it, and an
2093
+ export to FlowJo writes it back as FlowJo drew it, with its axes.
2094
+ - **Session files:** the `.flowsession` keys and schema version are
2095
+ unchanged (v1), and older sessions load as before. A session can now
2096
+ contain a `flowjo_ellipse` gate. An older OpenFlo does not know the kind
2097
+ and lets every event through it, so open such sessions in this version
2098
+ or later.
2099
+ - **Single-stain auto-detect** (Optimize compensation). A marker name in the
2100
+ channel no longer decides the match (`CD4 FL3-A` took `FL5.fcs`), `-A` and
2101
+ `-H` channels of one fluor share its control, and joined names match
2102
+ (`AF647`, `CD3FLA`, `eF450`, `compFLA`). A file that two fluors fit
2103
+ equally is reported, not assigned.
2104
+
2105
+ ## [2.6.1] - 2026-10-02
2106
+
2107
+ ### Fixed
2108
+ - **FlowJo round trip, measured against FlowJo 10.10.2.** FlowJo opened,
2109
+ computed and saved synthetic probe workspaces and OpenFlo exports with no
2110
+ manual steps; each FlowJo behaviour below comes from those saved files.
2111
+ - **Ellipse gates.** A workspace with an ellipse gate lost every sample that
2112
+ carried it, with no message. Ellipses are now exported as 64-vertex
2113
+ polygons.
2114
+ - **Quadrants.** A quadrant was exported as a QuadrantGate, which FlowJo
2115
+ counts as the whole parent. It is now four rectangle gates in FlowJo's own
2116
+ form, with each open side left unbounded, and a child gate stays under its
2117
+ own sector. Importing FlowJo's own quadrants also failed: their sectors
2118
+ have one bound per side, and the reader skipped every such rectangle.
2119
+ - **Fluorescence gates were applied to uncompensated data in FlowJo.** They
2120
+ were exported under the raw channel name, which FlowJo reads as the raw
2121
+ parameter. They are now written as `Comp-<channel>` for each channel the
2122
+ sample's own matrix compensates.
2123
+ - **Per-sample `<Keywords>`.** Without the FCS TEXT in each sample, FlowJo
2124
+ keeps the sample but computes none of its populations.
2125
+ - **Time gates were 100× off** across the `.wsp` boundary. FlowJo works in
2126
+ seconds: ticks × the FCS file's own `$TIMESTEP`, or ($ETIM − $BTIM) / last
2127
+ tick without one, even when the workspace says otherwise. Export and
2128
+ import now convert the same way.
2129
+ - **Compensation.** `read_compensation_matrix` found no spillover in any FCS
2130
+ file, because flowio lower-cases keywords. It now reads SPILL, and prefers
2131
+ it over `$SPILLOVER`, as FlowJo does. Each exported sample carries its own
2132
+ matrix, so a run that mixes files with different spillover no longer gets
2133
+ one file's matrix applied to all.
2134
+ - **Import of a lower edge near the bottom of an axis.** FlowJo treats such
2135
+ an edge as open. Imported gates now do the same, so an imported gate no
2136
+ longer undercounts by the events below the axis. The writer warns about
2137
+ each exported edge FlowJo will open.
2138
+ - Polygon counts still differ: by about 2% for scatter polygons and 3-7%
2139
+ for exported ellipses. See "Where FlowJo and OpenFlo count differently" in
2140
+ the README.
2141
+
2142
+ ## [2.6.0] - 2026-09-10
2143
+
2144
+ ### Added
2145
+ - **Seed phrases.** The run seed accepts a word or phrase as well as a number,
2146
+ so a run can be named (`pilot run 3`) instead of remembered. Hashed with
2147
+ blake2b rather than `hash()`, because Python randomises string hashing per
2148
+ process: a `hash()`-based phrase reproduces perfectly inside one session and
2149
+ silently stops between sessions, which surfaces as unexplained drift in
2150
+ results rather than as an error. The workspace seed field previously ran
2151
+ through an int-only validator that discarded unparseable input without
2152
+ saying so.
2153
+
2154
+ - **Embedding kernels compile off the critical path.** UMAP sets `cache=True`
2155
+ on none of its 114 numba kernels, so every process that embeds recompiles
2156
+ them — 18 to 29 seconds — and the workspace runs each unit in a fresh child.
2157
+ A run clusters first, and clustering is compiled C that releases the GIL, so
2158
+ the compile now happens alongside it. Measured on 30k events, Leiden + UMAP
2159
+ in one child: 52.2s to 46.3s, with clustering time unchanged and every
2160
+ result checksum identical. The embedding dialog warms on open too, so
2161
+ choosing methods pays for the compile instead of Run stalling.
2162
+
2163
+ - **"Fast graph" — an optional pruned neighbour graph, default off.** Links
2164
+ only kNN pairs, as PhenoGraph and Seurat do, instead of every pair sharing a
2165
+ neighbour. On separable populations it wins on every measure and is 6x
2166
+ faster (8 clusters vs 6 where 8 were planted, ARI 0.9545 vs 0.9487,
2167
+ homogeneity 0.9220 vs 0.8964, Leiden 0.8s vs 5.0s). On heavily overlapping
2168
+ populations the two disagree substantially and pruned scores worse on ARI —
2169
+ it splits populations the dense graph merges. Off by default because
2170
+ switching would change every clustering result already produced;
2171
+ `scripts/bench_snn_prune.py` regenerates the comparison.
2172
+
2173
+ ### Changed
2174
+ - **`run_louvain` default `restarts` 20 → 5.** The old default was justified
2175
+ by a docstring table that did not survive re-measurement. ARI at 1, 5 and 20
2176
+ restarts sits within 0.0014 on 20,000 events, so no count in that range is a
2177
+ measured optimum; 5 keeps some of the exploration restarts exist for at a
2178
+ quarter of what 20 cost. The docstring now says outright that 1 loses almost
2179
+ nothing measurable.
2180
+
2181
+ ### Fixed
2182
+ - **The public-mirror scrub gate could pass a leak.** A file that was not
2183
+ valid UTF-8 was skipped rather than scanned, so a forbidden token in a
2184
+ latin-1 file passed untouched; and the email allowlist was matched against
2185
+ the whole line, so one allowlisted token shadowed every address beside it.
2186
+ Both closed, with binary-by-content files skipped deliberately so the new
2187
+ fallback cannot decode a `.pyd` into spurious hits.
2188
+
2189
+ - **The GUI guard only protected tests that remembered to call it.**
2190
+ `OPENFLO_REQUIRE_GUI=1` exists so a CI leg cannot go green having tested no
2191
+ GUI. Measured with Tk deliberately broken, 55 GUI tests still skipped
2192
+ silently. It is now a pytest hook that sees every skip however the test
2193
+ wrote it, and probes Tk before converting anything so an xdist race still
2194
+ skips.
2195
+
2196
+ - **README told everyone to clone a private repository.** The install
2197
+ instructions pointed at `ChironTheCentaur/openflo`, which is private, and
2198
+ the public mirror carried that line verbatim. Now points at the public
2199
+ mirror, with relative links made absolute so they resolve on PyPI.
2200
+
2201
+ - The "Fast graph" checkbox pushed the run panel's widest row past its width
2202
+ cap and clipped the toolbar on Linux, where font metrics are wider than on
2203
+ Windows.
2204
+
2205
+ - `run_louvain`'s docstring reported measurements taken on a prototype that
2206
+ partitioned a different graph; every figure was wrong. `_snn_jaccard_graph`
2207
+ described itself as "the PhenoGraph / Seurat construction", which it is not
2208
+ — those bound the edge set at n*k, while this links every pair sharing a
2209
+ neighbour (6,106,544 edges against 297,138 at 20k events, k=30).
2210
+
2211
+ - The comment justifying the threaded `job.pkl` write claimed it was
2212
+ disk-bound. It is roughly half CPU, and that half holds the GIL. The fix
2213
+ works for a different reason — pickle releases the GIL for numpy's large
2214
+ buffer writes — now measured directly instead of inferred.
2215
+
2216
+ ### Reverted
2217
+ - **Parallel UMAP neighbour search.** Building the kNN with
2218
+ `pynndescent(n_jobs=-1)` and passing `precomputed_knn` is about 10% faster
2219
+ and was reverted: pynndescent's parallel NN-descent is deterministic only at
2220
+ a fixed numba thread count, so a seeded embedding became machine-dependent.
2221
+ An exact kNN keeps both properties but measured 1.03x at 100k events — the
2222
+ neighbour search is not the bottleneck, the seeded layout optimisation is.
2223
+
2224
+ ## [2.5.0] - 2026-09-09
2225
+
2226
+ ### Added
2227
+ - **`FlowSample.run_louvain()` — Louvain community detection that is actually
2228
+ reproducible.** PhenoGraph's Louvain shells out to the Blondel binaries,
2229
+ which seed themselves `srand(time(NULL) + getpid())`; the PID half is
2230
+ assigned by the OS, so that path cannot be pinned from outside the process.
2231
+ igraph — already a pinned dependency — implements Louvain in-process with a
2232
+ settable RNG, so here the seed is ours.
2233
+
2234
+ `restarts` preserves what makes PhenoGraph's Louvain good rather than merely
2235
+ fast. A single Louvain run lands in whatever local optimum its node ordering
2236
+ leads to; PhenoGraph re-runs it ~28 times and keeps the best modularity. This
2237
+ does the same with a **deterministic seed sequence** (`random_state`,
2238
+ `random_state + 1`, …) — the exploration is preserved and the answer is still
2239
+ identical every time.
2240
+
2241
+ Measured on 20k events, 8 overlapping populations, k=30, against the
2242
+ PhenoGraph binary (57.3s, Q=0.5887, 11 clusters, ARI 0.3187 vs planted
2243
+ truth, **not** reproducible):
2244
+
2245
+ | restarts | time | Q | ARI vs truth |
2246
+ |---|---|---|---|
2247
+ | 1 | 2.6s | 0.5770 | 0.3217 |
2248
+ | 5 | 13.4s | 0.5841 | 0.3393 |
2249
+ | **20** (default) | **47.5s** | **0.5877** | **0.3576** |
2250
+ | 40 | 92.9s | 0.5877 | 0.3576 (converged) |
2251
+
2252
+ At the default it matches the binary's modularity, finds the same number of
2253
+ clusters, scores **better** against the planted truth, runs slightly faster —
2254
+ and is reproducible. No subprocess, no clock, no PID, and no per-platform
2255
+ native shim.
2256
+
2257
+ ### Changed
2258
+ - The shared-nearest-neighbour / Jaccard graph construction is extracted into
2259
+ `_snn_jaccard_graph`, used by both `run_leiden` and `run_louvain`. They now
2260
+ differ only in how the graph is *partitioned*; a difference in how it was
2261
+ *built* would have made their results quietly incomparable.
2262
+
2263
+ ### Fixed
2264
+ - The workspace-sizing test asserted pixel counts that held only on Windows —
2265
+ the same toolbar needs 343px on Linux against 336px on Windows, so it failed
2266
+ CI on py3.11. It now asserts what the width floor is responsible for (the
2267
+ panel receives everything the 60% cap allows) rather than a measurement taken
2268
+ on one machine.
2269
+
2270
+ ## [2.4.9] - 2026-09-09
2271
+
2272
+ ### Changed
2273
+ - **Clustering is now reproducible by default.** PhenoGraph's default community
2274
+ detection is Louvain, run through the Blondel reference *binaries* shipped
2275
+ inside the package. Those expose no seed at all — `community.exe` accepts only
2276
+ `-w -p -q -l -v -h`, and are driven in a restart loop whose stopping rule
2277
+ includes `(time.time() - tic) < time_limit`. So the result depended on both
2278
+ an unpinnable RNG **and how fast your machine was**.
2279
+
2280
+ The seed is `srand(time(NULL) + getpid())` (`main_community.cpp:121` in the
2281
+ Blondel v0.2 source the binaries are built from), feeding a shuffle of the
2282
+ node traversal order in `one_level()` (`community.cpp:295-302`). **The PID
2283
+ half is assigned by the OS**, which rules out pinning the result by injecting
2284
+ a fixed or stepped sequence of clock values: **20 runs on identical input,
2285
+ all finishing inside one second, produced 20 distinct partitions** — same
2286
+ `time(NULL)`, different PIDs.
2287
+
2288
+ Measured on 50 000 events with overlapping populations: two runs of the
2289
+ default, same data and same `random_state`, agreed only to **ARI 0.76** —
2290
+ the labels genuinely differed. An analysis that changes when you re-run it is
2291
+ hard to put in a methods section.
2292
+
2293
+ The default is now PhenoGraph's *seeded Leiden* backend, which calls
2294
+ `leidenalg.find_partition(..., seed=...)` in-process: deterministic by
2295
+ construction, no subprocess, no timing-dependent loop. It was also **~1.5x
2296
+ faster** (63s vs 95s on that fixture) and equivalent against planted ground
2297
+ truth (ARI 0.4905 vs 0.4990).
2298
+
2299
+ **This changes results.** Leiden is the corrected form of Louvain — it cannot
2300
+ emit internally disconnected communities — not a reimplementation of it: on
2301
+ ambiguous data the two agreed only to ARI 0.74, giving 7 clusters against 9.
2302
+ Pass `reproducible=False` (or `--no-reproducible`) to restore Louvain for
2303
+ reproducing prior output, matching published PhenoGraph results, or using the
2304
+ GPU path (cuGraph's Louvain is likewise unseeded). `--reproducible` is kept as
2305
+ an accepted no-op so existing scripts keep working.
2306
+
2307
+ ### Fixed
2308
+ - **`import openflo.cli` no longer reconfigures `sys.stdout`.** It applied the
2309
+ UTF-8 console guard as an *import side effect*, changing the stream's error
2310
+ handler for the whole process. That broke pytest-xdist outright — workers talk
2311
+ to the controller over stdout via execnet, so importing this module inside a
2312
+ worker corrupted the channel and killed it with `EOFError: expected 1 bytes,
2313
+ got 0`, taking an unrelated test down with it. The same hazard applied to any
2314
+ notebook or embedding host importing it as a library. The guard now runs only
2315
+ in `main()`, which owns the process's streams — and still covers `--help`,
2316
+ which the import-time call did not.
2317
+
2318
+ ## [2.4.8] - 2026-09-09
2319
+
2320
+ ### Changed
2321
+ - **Loading is ~3x faster and uses half the memory.** Every sample was copied
2322
+ twice on load — once when pandas materialised the event array, and again to
2323
+ give `data` its own copy of `raw`. Neither was needed. Measured on 8 files of
2324
+ 2M events x 18 channels: **2.25s / +2304 MB → 0.75s / +1152 MB**. Projected
2325
+ to a 25-file session: **7.0s / 7.2 GB → 2.3s / 3.6 GB**. A single 2M x 18
2326
+ file loads in 0.08s, down from 0.34s.
2327
+
2328
+ The memory halving matters more than the seconds. Holding 7 GB of resident
2329
+ frames is where a machine starts paging, and paging is what makes loading
2330
+ *feel* slow long after the CPU work is done.
2331
+
2332
+ How: flowio returns an `array.array` supporting the buffer protocol, so
2333
+ `np.asarray` was already a free view — `pd.DataFrame(..., copy=False)` stops
2334
+ pandas duplicating a buffer nothing else references. And `data` is now a
2335
+ SHALLOW copy of `raw`, which under pandas ≥ 3 copy-on-write is a distinct
2336
+ object that duplicates a block only when written.
2337
+
2338
+ `raw` still holds the pristine detector values the `.fcs` export depends on:
2339
+ `data` is written in place in a dozen places (compensation and transforms
2340
+ among them) and none of those writes reach `raw`. Pinned by
2341
+ `tests/test_load_no_copy.py`, which fails if anyone later "simplifies" the
2342
+ shallow copy into an alias.
2343
+
2344
+ No numbers change: golden 8/8, controls 22/22, dtypes still float32.
2345
+
2346
+ - **Ellipsoid gates use flowutils' compiled `gating_c`** — bit-identical
2347
+ (0 of 1,000,000 events differ, same inclusive-edge rule) and ~1.4x faster,
2348
+ and it puts polygon and ellipsoid geometry in one library.
2349
+
2350
+ ## [2.4.7] - 2026-09-09
2351
+
2352
+ ### Fixed
2353
+ - **Adjacent interval and rectangle gates silently LOST every event on their
2354
+ shared boundary.** Both kinds used fully-open bounds (`lo < x < hi`), so two
2355
+ neighbouring gates — `[0,10]` and `[10,20]` — each excluded a value of
2356
+ exactly 10 and those events vanished from *both* populations. Frequencies
2357
+ then failed to sum. Measured on integer `$DATATYPE I` data: **14.3% of
2358
+ events lost** between two adjacent intervals, and a pair of rects covering
2359
+ `x ∈ [0,20]` captured only 77.5% of it. Bounds are now half-open `[lo, hi)`,
2360
+ matching Gating-ML's RectangleGate convention and the polygon rule fixed in
2361
+ 2.4.6, so adjacent gates partition with no gaps and no overlaps.
2362
+ - **The same region gated differently as a rect than as a polygon** — 823 of
2363
+ 20 000 integer events on a 10×10 square. Both now use the same convention
2364
+ and agree exactly.
2365
+ - **Quadrant gates no longer need the `nextafter` workaround.** The Gating-ML
2366
+ quadrant importer nudged each divider down by one ULP so a strict `>` would
2367
+ still keep events sitting exactly on it. Half-open bounds do that properly,
2368
+ so the hack is removed — and it would now actively *cause* a double-count.
2369
+ Verified with 3 410 events on a divider (including a pile-up at exactly
2370
+ zero, i.e. `arcsinh(0)`): all four quadrants tile exactly, summing to 100%.
2371
+
2372
+ **Who is affected:** continuous float data is unchanged (0 of 200 000 events
2373
+ differ), so ordinary FCS gating and the golden baseline are untouched. This
2374
+ only moves events that land on exact coordinates — integer `$DATATYPE I`
2375
+ channels — and there they were previously being dropped. `threshold` gates
2376
+ are deliberately unchanged (`> value`): a threshold is a cut, not a region,
2377
+ and its NOT-complement already partitions exactly.
2378
+
2379
+ ### Changed
2380
+ - `tests/test_gating.py` — the test that pinned "interval is strictly between"
2381
+ now pins the half-open rule, and gains three property tests: adjacent gates
2382
+ partition, a rect equals the identical polygon, and quadrants tile without
2383
+ the removed hack. Reverting the fix turns all four red.
2384
+ - GUI tests now distinguish a genuinely broken Tk from a transient one. The
2385
+ fail-closed guard added alongside 2.4.6 could turn an intermittent Tk
2386
+ start-up race under `pytest-xdist` into a red build; it now retries once and
2387
+ only fails when Tk is really unavailable, so a silent coverage hole is still
2388
+ caught without crying wolf.
2389
+
2390
+ ## [2.4.6] - 2026-09-09
2391
+
2392
+ ### Fixed
2393
+ - **Polygon gates no longer double-count events on a shared boundary.** Gating
2394
+ used `matplotlib.path.Path.contains_points`, whose treatment of points lying
2395
+ exactly ON an edge is wrong for gating in three ways, all measured rather
2396
+ than argued:
2397
+ - **It double-counts a shared edge.** Two gates meeting on a line both
2398
+ claimed every point on it. Over a quadrant split of integer-valued events,
2399
+ 902 of 20 000 landed in two populations at once and the four quadrants
2400
+ summed to more than 100%.
2401
+ - **It depended on winding.** The same square drawn clockwise and
2402
+ anticlockwise gave different answers on its boundary — so the same gate
2403
+ could include different events depending on the direction the user happened
2404
+ to drag the mouse.
2405
+ - **Its rule was not self-consistent**: for an axis-aligned square it
2406
+ included three edges and excluded the bottom; for the same square rotated
2407
+ 45° it included all four.
2408
+
2409
+ Gating now uses flowutils' compiled `gating_c`, which implements the standard
2410
+ half-open convention — a point on a "lower/left" edge is inside, one on an
2411
+ "upper/right" edge is outside — so adjacent gates tile with no gaps and no
2412
+ overlaps, which is the property quadrant gating depends on.
2413
+
2414
+ **Who is affected:** on continuous float data the two rules agree *exactly*
2415
+ (0 of 200 000 events differed), so ordinary FCS gating is unchanged and the
2416
+ golden baseline did not move. They diverge only where events land on exact
2417
+ coordinates — integer `$DATATYPE I` channels, where 6.96% of events sit on a
2418
+ gate line and matplotlib was the rule double-counting them. If you gate
2419
+ integer-valued channels with adjacent or quadrant gates, your population
2420
+ counts change, and they change to the correct values.
2421
+
2422
+ ### Changed
2423
+ - Polygon gating is **5–12x faster** as a side effect: at 1M events a normal
2424
+ gate went 172 → 32 ms and a tight gate 163 → 14 ms. The bounding-box
2425
+ prefilter added in 2.4.5 has been REMOVED — with the faster crossing test,
2426
+ building the prefilter mask costs more than it saves (48 ms vs 32 ms).
2427
+ - `tests/test_polygon_prefilter.py` is now `tests/test_polygon_gating.py` and
2428
+ asserts the correctness *properties* — partitioning, winding invariance,
2429
+ vertex-order and translation invariance — rather than agreement with any
2430
+ particular library, so it stays meaningful if the backend changes again.
2431
+
2432
+ ## [2.4.5] - 2026-09-09
2433
+
2434
+ ### Changed
2435
+ - **Polygon gating is ~3x faster on a real gate hierarchy, with bit-identical
2436
+ results.** Polygon gates were the most expensive step on the default path
2437
+ (~172 ms per gate per million events), and a hierarchy evaluates one per gate
2438
+ per descendant. A point outside the polygon's bounding box cannot be inside
2439
+ it, so those are now rejected with two comparisons per axis instead of a full
2440
+ crossing test; everything that survives goes through the same matplotlib
2441
+ call. Measured on a 12-gate hierarchy over 500k events: **873 ms → 265 ms**,
2442
+ masks identical. The gain scales with how tight the gate is — 13x for a gate
2443
+ covering ~4% of the plot, 1.6x at 50% — which is the right way round, since
2444
+ tight child gates deep in a tree are evaluated most often.
2445
+
2446
+ This is a pure speed change: `tests/test_polygon_prefilter.py` pins the
2447
+ results against matplotlib on vertices, edge midpoints, collinear points,
2448
+ concave notches, clockwise winding and flow-scale magnitudes, because a
2449
+ polygon gate decides which cells are in a population.
2450
+
2451
+ ### Notes
2452
+ - **`flowutils.gating.points_in_polygon` was evaluated and deliberately NOT
2453
+ adopted**, despite being ~5x faster again and already compiled and already a
2454
+ dependency. It disagrees with matplotlib on points lying exactly ON a
2455
+ boundary — vertices, edge midpoints, collinear points — which is not
2456
+ hypothetical for integer `$DATATYPE I` channels, whose events land on exact
2457
+ coordinates. Adopting it would silently move boundary events between
2458
+ populations and change every downstream frequency, so it needs a deliberate
2459
+ re-baseline rather than a drop-in swap. The reasoning is recorded at the call
2460
+ site and in the tests so the trade-off is visible to whoever next looks for
2461
+ speed here.
2462
+
2463
+ ## [2.4.4] - 2026-09-09
2464
+
2465
+ ### Fixed
2466
+ - **The secondary windows had no hover help at all.** The editor gained
2467
+ tooltips early; the thirty `ui_*` dialogs never did — an audit found **zero**
2468
+ tooltip calls against ~105 interactive controls, so Preferences,
2469
+ Compensation, Frequencies, Auto-gate, Statistics and the rest offered nothing
2470
+ on hover. That is the wrong way round: the main window's controls are seen
2471
+ constantly and learned by repetition, while a dialog's options are met once a
2472
+ month and forgotten in between. Every actionable control in every dialog now
2473
+ carries help that says what it *does* and what changes if you use it —
2474
+ %Parent vs %Total, Transparent vs Translucent export backgrounds, why the
2475
+ non-parametric test is the default, what the FMO percentile trades off.
2476
+ Nothing was wrong with the tooltip machinery itself; it was simply never
2477
+ called outside the editor.
2478
+ - **The docked Pipeline Workspace opened clipped, and squeezed the plot.** Its
2479
+ `Group ▾` menubutton lost its arrow, and `res`, `PHATE` and `Reproducible`
2480
+ were cut off at the panel edge. Three causes: the status line had no
2481
+ `wraplength`, so it reported its whole 456 px of text as its required width
2482
+ and — being the widest child — made that the panel's requirement; the toolbar
2483
+ was gridded without `sticky`, so it centred and lost content off *both* edges
2484
+ at once; and the reveal used a guessed 320 px floor when the toolbar needs
2485
+ 336. The floor is now measured from the panel itself, so it stays correct as
2486
+ controls change.
2487
+ - **Docking the workspace no longer takes its width out of the plot.** At the
2488
+ default window size that left the plot column ~512 px against control rows
2489
+ needing ~754, clipping the axis combos, `Auto-gate` and `Show cleaned-out
2490
+ events` — controls the user never asked to give up. The window now grows when
2491
+ the screen has room, and shares only when it does not.
2492
+
2493
+ ### Added
2494
+ - `tests/test_dialog_tooltips.py` — asserts every actionable dialog control has
2495
+ hover help, and that revealing the workspace clips neither its own toolbar nor
2496
+ the plot's controls. All three tests were verified to fail against the code
2497
+ they replaced.
2498
+
2499
+ ## [2.4.3] - 2026-09-08
2500
+
2501
+ ### Fixed
2502
+ - **`openflo-selftest` crashed instead of reporting on a legacy Windows
2503
+ console.** Its results table prints check marks, which cp1252 cannot encode,
2504
+ so the command whose whole purpose is to say whether an install reproduces
2505
+ reference behaviour raised `UnicodeEncodeError` and reported nothing. The
2506
+ same fault was found in `openflo-doctor --help` **despite** that module
2507
+ having a guard — the guard ran after `parse_args`, and argparse prints help
2508
+ and exits from inside it — and in `openflo-compare`, `openflo-voltage` and
2509
+ `openflo-synth`, which had no guard at all. Every entry point now calls a
2510
+ shared `force_utf8_streams()` before building its parser.
2511
+ - **The declared `psutil` floor could not be installed on Linux.** psutil
2512
+ 5.9.0-5.9.3 publish Linux wheels only up to cp310, so Python 3.11+ had to
2513
+ build it from source, which needs a C compiler. The floor is now 5.9.4, the
2514
+ first release shipping a `cp36-abi3` wheel that covers every supported
2515
+ Python.
2516
+ - **The declared `matplotlib` floor could not be resolved at all.** matplotlib
2517
+ 3.8.0-3.8.3 declare `numpy<2`, which cannot be satisfied alongside this
2518
+ project's `numpy>=2.0`. The floor is now 3.8.4, the release that dropped the
2519
+ cap.
2520
+
2521
+ ### Changed
2522
+ - **The eight shared core dependencies are declared as ranges, not exact
2523
+ pins** (numpy, pandas, scipy, scikit-learn, matplotlib, seaborn, openpyxl,
2524
+ psutil), so `pip install openflo` can coexist with an environment that
2525
+ already has a scientific stack rather than downgrading it. Each floor was
2526
+ found by testing, and the evidence is recorded per-dependency in
2527
+ `docs/PYPI_READINESS.md`. `requirements.txt` still pins the exact tested
2528
+ stack — that is what CI and `openflo-doctor` check against, and it is
2529
+ unchanged. The six leaf dependencies nobody else competes for (FlowIO,
2530
+ FlowUtils, PhenoGraph, umap-learn, igraph, leidenalg) remain exact pins,
2531
+ because their versions move OpenFlo's numbers.
2532
+
2533
+ ### Added
2534
+ - **CI installs the built wheel at both edges of every declared range** — a new
2535
+ `install` job across {ubuntu, windows} x {floor dependencies on Python 3.11,
2536
+ newest resolvable on 3.12}. It builds the wheel, installs it the way a user
2537
+ does (no editable install, no `--no-deps`), and then requires the installed
2538
+ artifact to reproduce the golden baseline and pass the response controls.
2539
+ This closes two gaps: an editable install imports from `src/`, so missing
2540
+ package data was invisible, and nothing previously installed the floor of any
2541
+ declared range. The floor pins are generated from `pyproject.toml` by
2542
+ `scripts/lowest_direct_requirements.py`, so they cannot drift from what is
2543
+ declared. All three bugs fixed above were found by this job.
2544
+
2545
+ ## [2.4.2] - 2026-09-08
2546
+
2547
+ ### Fixed
2548
+ - **Starting a run no longer stalls on labelling events.** `prepare_unit` tags
2549
+ every event with its source group and sample — one short string repeated
2550
+ across a whole member. Assigning those as object-dtype columns materialised
2551
+ millions of Python string references and was, on a 3M-row unit, **415 ms of
2552
+ the function's 531 ms** (the concat itself was 80 ms), then paid again when
2553
+ the frame is pickled for the child process. As categoricals: 604 → 111 ms of
2554
+ UI-blocking work, 730 → 414 MB in memory, and the job file 3044 → 803 ms to
2555
+ write. The tag values are unchanged.
2556
+ - **Downsample-propagate gave a different answer depending on which sample
2557
+ finished loading first.** The option trims every loaded sample to the smallest
2558
+ enabled one — but each arriving sample was compared against the floor of
2559
+ whatever had loaded *so far*, and that order comes from a thread pool.
2560
+ Measured on four samples of 50k/40k/30k/20k events: loading them
2561
+ largest-first trimmed **nothing at all**, smallest-first trimmed **everything
2562
+ to 20k**, and an interleaved order gave a mix. The same session resumed twice
2563
+ could hold different event counts, and every frequency computed from them
2564
+ would differ. The trim now happens once over the whole set, sharing its
2565
+ definition with the toggle that documents the behaviour.
2566
+ - **A log histogram threw away events it could have shown.** The log floor was
2567
+ `max(lo, hi * 1e-6)`, which raised a perfectly good *positive* lower bound to
2568
+ six decades below the top of the range — on an all-positive channel spanning
2569
+ 1e-3 to 1e6 that put **half the events off the bottom of the plot**, silently.
2570
+ The clamp is only needed when there is no positive lower bound to use. What a
2571
+ log axis genuinely cannot show — non-positive values, routinely a third of a
2572
+ compensated channel — is now stated on the figure itself, so an exported panel
2573
+ carries it too, and the note stays silent when nothing is hidden.
2574
+
2575
+ ### Changed
2576
+ - **Starting a pipeline run no longer freezes the window while it stages.**
2577
+ `_launch_next` built the run unit AND serialised it to `job.pkl` on the Tk
2578
+ thread — the code even called it "cheap (main)". The write dominates:
2579
+ measured here, 569 ms for a 20-sample x 200k-event run (336 MB) and 1264 ms
2580
+ for 8 x 1M (672 MB), all of it blocking the UI. The write now happens on a
2581
+ worker and the child process launches when it completes.
2582
+ **Only the write moved.** `prepare_unit` deliberately stays on the Tk thread:
2583
+ it reads the editor's loaded samples, which the user can mutate, so running
2584
+ it concurrently would introduce the same read-during-write race that was
2585
+ fixed in the plot path. Its share is the smaller one (concat + tagging, 250 of
2586
+ the 1720 ms at 8 x 1M) and removing it needs a snapshot design, not a thread.
2587
+
2588
+ ## [2.4.1] - 2026-09-08
2589
+
2590
+ ### Upgrading — results that legitimately change
2591
+
2592
+ This release corrects analyses, so re-running work done on an earlier version
2593
+ can give different numbers. Nothing needs migrating: the session format is
2594
+ unchanged and old sessions open normally. What follows is where to expect a
2595
+ difference, and in which direction.
2596
+
2597
+ - **Event counts go UP where auto-clean was removing events for no reason.**
2598
+ Four separate causes, each measured on a sample with no actual fault: an
2599
+ integer-valued channel (`$DATATYPE I`) lost **38%** to reported drift; one
2600
+ dead or unused detector removed **100%**; a sample QC'd after clustering or
2601
+ FMO positivity lost **12.4%** (a cluster label) to **49.8%** (a positivity
2602
+ flag); and a coarsely quantised viability channel cost **46%** of an all-live
2603
+ sample. Those events are now kept, so any frequency computed from them
2604
+ shifts.
2605
+ - **Gates on a LOG-scaled channel select FEWER events.** A non-positive value
2606
+ has no logarithm and is now `NaN` rather than `0.0`. Previously such events
2607
+ were placed at raw intensity 1.0 — above every genuinely positive event
2608
+ dimmer than 1.0 — so a negative-side gate was collecting them. They now drop
2609
+ out of gates, medians and plots. If a marker-positive frequency falls, this
2610
+ is the likely reason, and the old number was the wrong one.
2611
+ - **Condition-comparison group means go DOWN for patchy populations.** A
2612
+ cluster absent from a sample now contributes 0% instead of being left out of
2613
+ the average. A population found in one of three samples at 10% was reported
2614
+ as 10%; it is now 3.3%. The table also gains an `n_samples` column, and
2615
+ `sd_pct` is now a real number where it used to be `NaN`.
2616
+ - **MEM scores go UP when a very small cluster was present.** A cluster with no
2617
+ measurable spread was taking the top of the scale and compressing every other
2618
+ population; measured, three real populations recovered from 4, -5, -5 back to
2619
+ 8, -10, -9. Marker labels can change as a result.
2620
+ - **Doublets are removed from transformed scatter.** If FSC-A/FSC-H had been
2621
+ arcsinh- or logicle-transformed before auto-clean ran, the doublet filter was
2622
+ a no-op and kept them all.
2623
+ - **A "NOT" gate no longer contains events it could not measure.** The
2624
+ complement population shrinks, and a marker's positive and negative
2625
+ populations no longer sum to the whole sample when some events have no
2626
+ reading — which is the honest arithmetic.
2627
+
2628
+ Sessions saved by an earlier version still load, including their processed-data
2629
+ sidecars: sample names that had to be sanitised now get a different sidecar
2630
+ filename, and the loader looks for both.
2631
+
2632
+ ### Changed
2633
+ - **Five more, in file formats and the auto-clean filters.** A FlowJo
2634
+ workspace import *discarded every population name* — the name lives on the
2635
+ enclosing `<Population>` element, not the `<Gate>` inside it — so a
2636
+ 40-population workspace arrived fully unnamed while its geometry survived
2637
+ intact, which is why a round trip looked like it worked. A compensation
2638
+ matrix whose *channel name contained a comma* (a `$PnS` description like
2639
+ "CD4 FL3, clone SK3" is ordinary) wrote a file this same module then
2640
+ refused to read. Standards-conformant Gating-ML `-INF` bounds reached the
2641
+ session file as `-Infinity`, which *RFC 8259 forbids* — valid to Python,
2642
+ rejected by `JSON.parse` and `jq`. An *unusable reference spectrum* (a
2643
+ single stain dimmer than the unstained control clips to all-zero) was
2644
+ reported as cosine similarity 0.0 against every other fluor, i.e.
2645
+ "maximally distinct, trivially unmixable", so the panel was declared free of
2646
+ problematic pairs *because* one fluor had no spectrum. And the *auto-clean
2647
+ doublet window was a no-op on transformed scatter*: it is a fixed fraction
2648
+ of the median ratio, but that ratio's spread collapses from 14.5% to 1.7%
2649
+ after an arcsinh bake, so all 4,000 planted doublets were retained where
2650
+ linear scatter lost all 4,000.
2651
+ - **Seven more places where the wrong events were selected, deleted or
2652
+ counted.** Every one was reproduced and measured before it was touched.
2653
+ *Acquisition QC read our own analysis output as a detector*: the denylist had
2654
+ fallen behind the columns the app writes, so a `leiden` column made the margin
2655
+ filter delete the whole top-numbered cluster (12.4% of a clean sample) and an
2656
+ FMO `<channel>_pos` flag deleted every marker-positive event (49.8%).
2657
+ *Two samples could be handed the same session sidecar* — `Tube 01 Rep
2658
+ A` and `Tube_01_Rep_A` sanitised to one filename — and because
2659
+ restore prefers the sidecar over the raw FCS, one sample silently came back
2660
+ holding the other's events, clusters and compensated values.
2661
+ *A cluster absent from a sample was treated as unmeasured rather than 0%*, so
2662
+ a population found in 1 of 3 samples at 10% was published as the group's 10%
2663
+ instead of 3.3%.
2664
+ *A boolean gate that could not resolve its operands admitted everything*,
2665
+ turning a "NOT X" population into the entire sample (20,000 events where
2666
+ 10,004 was correct); it now fails closed, and `apply_region_gates` passes the
2667
+ gate list so operands actually resolve.
2668
+ *A NOT gate adopted every event it could not measure* — 2,000 events with no
2669
+ reading made up 20% of a CD3-negative population, while the two gates still
2670
+ summed to the sample total so the result looked self-consistent.
2671
+ *The `log` transform folded every non-positive value onto 0.0*, i.e. raw
2672
+ intensity 1.0, ranking the dimmest events above every genuine positive below
2673
+ 1.0 and mis-assigning 32,620 of 100,000 events; its inverse separately
2674
+ destroyed every value at or below raw 1.0 on a round trip, which runs
2675
+ whenever a channel's scale is changed.
2676
+ *A comb was mistaken for a bimodal distribution*: an integer channel spread
2677
+ over 256 bins leaves most of them structurally empty, and a zero bin between
2678
+ two comb teeth satisfies the "is this really bimodal" test, so the auto-clean
2679
+ viability filter deleted 46% of an all-live sample — whether it fired at all
2680
+ decided by how the bins happened to line up.
2681
+ - **Batch-correction QC reports what it cannot measure (hardening, not a live
2682
+ bug).** `CytoNorm.qc` returned `0.0` — the *best* score on a lower-is-better
2683
+ scale — for a channel with no usable events, and raised `Distribution can't
2684
+ be empty` when no event was finite in every channel. Both are now NaN with a
2685
+ warning. To be precise about the scope: neither was reachable from the
2686
+ shipped call paths, because both callers pass `qc()` exactly the events they
2687
+ just fitted and `fit()` rejects that input first. Measured distances for real
2688
+ data are unchanged.
2689
+ - **A cluster with no spread no longer sets the scale for the whole MEM
2690
+ table.** MEM scores a population as `|median shift| + IQR_ref/IQR_pop - 1`
2691
+ and then rescales so the largest |score| maps to 10. A population with no
2692
+ measurable spread — a single event, or every event identical — has an IQR of
2693
+ exactly 0, which the epsilon guard turned into "perfectly tight", the most
2694
+ that term can reward. Since the table is divided by its own largest value,
2695
+ one such cluster compressed every real call in it: a single stray event given
2696
+ its own label pushed three well-separated populations from 10, -10, -7 down
2697
+ to 4, -5, -5. The spread term now drops out when there is no spread to
2698
+ measure, and such a population is scored on its median shift alone.
2699
+ Deliberately unchanged: a population sitting exactly at the median of a
2700
+ multimodal reference has a genuinely unstable sign under "vs all other
2701
+ cells", which is the metric's own ambiguity rather than a defect.
2702
+ - **Three more places where a degenerate input was answered with a
2703
+ confident number.** The FMO
2704
+ preview substituted `-999` for a missing per-axis threshold and then drew
2705
+ four quadrant percentages from it, two of them defined by a boundary nobody
2706
+ measured — and on a compensated channel whose dim population sits near
2707
+ -2000, that sentinel fell *inside* the data and split it 67.9 / 7.0 / 23.2 /
2708
+ 2.0. And a FlowJo workspace declaring spillover for only some detectors was
2709
+ accepted with the rest silently left uncompensated, because the acceptance
2710
+ check counted coefficients rather than channels. All three now report what
2711
+ they do not know, and the uncovered detectors are named.
2712
+ - **Acquisition QC no longer punishes integer-valued channels, dead detectors,
2713
+ or quiet time bins.** All three came from one line: the robust outlier band
2714
+ was `median(|v - med|) + 1e-10`, and that epsilon silently redefined "no
2715
+ measurable spread" as "infinitely sensitive". On an integer channel
2716
+ (`$DATATYPE I`, very common) more than half the per-bin medians tie with the
2717
+ overall median, so the MAD is exactly 0 — and a clean, drift-free sample lost
2718
+ **38% of its events** to reported "drift", while the identical data stored as
2719
+ float lost none. The scale floor now comes from each caller's own data (a
2720
+ channel's quantisation step; sqrt(N) counting noise for an event count)
2721
+ rather than from the deviations being judged, which also stops ordinary
2722
+ Poisson scatter in low-count bins reading as a flow-rate fault. Where no
2723
+ scale exists in either place, nothing is called an outlier.
2724
+ Alongside it, the margin/saturation filter skipped its own degenerate case:
2725
+ a channel that is constant across the file has no ceiling to pile up
2726
+ against, but every event counted as "at the max" — and because margins are
2727
+ OR-ed across the panel, **one unused or disabled detector deleted 100% of an
2728
+ otherwise healthy sample**. Integer and float storage of the same data now
2729
+ produce identical QC verdicts in both directions, and real faults — a 400-unit
2730
+ clog, a saturated channel, a 7x count burst — are still caught.
2731
+ - **Differential abundance now shows the sample count per group, and says so
2732
+ when the design cannot support its p-values.** The GLM borrows a dispersion
2733
+ shared across populations, so it still returns a p-value with ONE sample per
2734
+ group — and the table rendered that as significance stars with no `n`
2735
+ anywhere on screen, making a 1-vs-1 comparison look identical to a replicated
2736
+ one in a figure users export directly. The numbers are unchanged; `n` per
2737
+ group is now a column, and fewer than three samples in the smaller group
2738
+ draws an explicit note that the result is exploratory rather than evidence of
2739
+ a difference.
2740
+
2741
+ ### Added
2742
+ - **MEM annotation has a response control.** A planted marker-high population
2743
+ must score that marker positive, swapping which population carries the
2744
+ phenotype must swap the sign, a marker drawn identically for both populations
2745
+ must score near zero, and a stray one-event cluster must not compress the
2746
+ real scores.
2747
+ - **The response controls are now attacked, not just run.**
2748
+ `scripts/mutate_controls.py` breaks each subsystem on purpose — compensation
2749
+ never applied, a gate ignoring its vertices, clustering collapsed to one
2750
+ cluster, QC removing a fixed 10% regardless of input — and checks that some
2751
+ control notices. All twelve attacks are caught. It paid for itself
2752
+ immediately by catching a containment check that could not fail, and the
2753
+ allow-list guard was re-keyed by enclosing function after a new fallback was
2754
+ found inheriting an unrelated entry's excuse.
2755
+ - **Gating has a response control.** A polygon must recover the fraction
2756
+ planted inside it, an empty region must come back empty, a child gate must be
2757
+ the intersection with its parent, and the same events must be selected when
2758
+ the rows are shuffled or the index does not start at zero.
2759
+ - **Acquisition QC now has a response control.** The golden baseline and the
2760
+ response controls between them never touched `AcquisitionQC` — the stage that
2761
+ runs on every sample, and the one that turned out to hold three live bugs.
2762
+ Four relational checks were added: a clean run is left alone, events removed
2763
+ tracks the span of a planted drift excursion, and the verdict is unchanged
2764
+ both by rounding the data and by the presence of a dead detector. The last
2765
+ two come back 45% and 100% apart against the previous release.
2766
+ - **A regression guard for the "plausible number for an unknown" bug.** Six
2767
+ shipped defects in this codebase share one shape: a guard correctly notices
2768
+ the input is degenerate, then substitutes a value that reads as a successful
2769
+ measurement rather than one meaning *unknown* — r² = 1.0 for an undefined
2770
+ fit, a corrupt event at the end of a trajectory, `sd = 0` for one replicate,
2771
+ an entire sample deleted. A test now scans the analysis modules for that
2772
+ shape and fails on anything new, with a reviewed allow-list carrying a
2773
+ specific reason per site. It states its own limits (it only catches
2774
+ comparison-style degeneracy checks and literal fallbacks), checks that the
2775
+ scanner still matches the shape it was written for, and fails if an
2776
+ allow-list entry goes stale.
2777
+
2778
+ ### Fixed
2779
+ - **A trajectory is no longer rooted on an arbitrary cell.** `robust_root`
2780
+ returned index 0 when the root channel had no finite values, so pseudotime
2781
+ still came back looking like a trajectory while its origin and direction were
2782
+ meaningless. It now refuses, matching the `empty X` check beside it; the
2783
+ dialog already reports the error.
2784
+ - **Unusable scatter no longer deletes the whole sample.** `filter_doublets`
2785
+ built its acceptance window from the median FSC-A/FSC-H ratio and substituted
2786
+ `0.0` when no event had a usable one — making the window `[0, 0]`, which
2787
+ matches nothing. A sample whose FSC-H is non-positive throughout was reduced
2788
+ to **zero events**, with no error and no warning. Doublets cannot be
2789
+ identified without the ratio, so the filter is now skipped and reported,
2790
+ which is what the missing-channel branch directly above it already did. The
2791
+ cell-cycle singlet gate had the identical fallback, where the effect was every
2792
+ cell scored `NA` and an empty cell-cycle result.
2793
+ - **A group of one replicate no longer reports `sd = 0`.** The sample standard
2794
+ deviation divides by n-1, so it is undefined for a single value —
2795
+ `compare_groups` returned `0.0`, a confident claim of "this group had no
2796
+ variability", rendered straight into the group summary as `sd=0`. It is now
2797
+ NaN. A group that genuinely is constant still reports 0, because that is
2798
+ measured rather than assumed.
2799
+ - **A corrupt event is no longer placed at the end of the trajectory, and no
2800
+ longer contaminates the trend curve.** Two halves of the same problem.
2801
+ `_geodesic_pseudotime` fills unreachable cells with the MAXIMUM distance — a
2802
+ deliberate, documented choice for a genuinely disconnected component — but an
2803
+ event with a `NaN` or `inf` coordinate is also unreachable, so it inherited
2804
+ that fill and plotted at pseudotime 1.0, indistinguishable from the most
2805
+ differentiated cell in the sample. Such events are now excluded and returned
2806
+ as NaN; a real disconnected component keeps its documented fill. Then
2807
+ `pseudotime_trends` binned them anyway: `searchsorted` sorts NaN **last**, so
2808
+ every unknown-pseudotime event was clipped into the FINAL bin, dragging the
2809
+ terminal point of the published "expression vs pseudotime" curve toward
2810
+ whatever they happened to express — measured, three such events at 1000 moved
2811
+ a last-bin mean from 1.0 to 600.4. Empty bins are still NaN rather than zero.
2812
+ - **A degenerate MESF calibration no longer reports a perfect fit.** When every
2813
+ bead peak is assigned the same value — a data-entry mistake — r² is
2814
+ mathematically undefined, and `fit_mesf_calibration` returned **1.0**: the
2815
+ strongest possible "this calibration is good" signal, attached to a fit whose
2816
+ slope is ~0 and which would turn every converted value into the intercept.
2817
+ The dialog showed `R²=1.0000` and the user could apply it to every loaded
2818
+ sample. r² is now NaN in that case, and the calibration dialog refuses the
2819
+ fit and explains why. A real calibration is unchanged (the golden's
2820
+ `calibration.r2` still reads exactly 1.0), and a genuinely poor fit still
2821
+ reports r² well below 1 — the job the docstring claims for it.
2822
+
2823
+ ## [2.4.0] - 2026-09-05
2824
+
2825
+ ### Added
2826
+ - **Response controls — `openflo-selftest --controls`.** A second synthetic
2827
+ suite that answers the question the golden baseline cannot. The golden pins
2828
+ seven numbers from one fixed dataset, so it proves *the answer did not
2829
+ change* — an implementation that ignored its input and returned those seven
2830
+ constants would pass all seven. The response controls instead assert
2831
+ *relationships between an input you vary and the output produced*: a
2832
+ spill-free negative control that must come back at zero, a spillover
2833
+ titration that must track its planted dose, a doublet titration, a planted
2834
+ composition shift that must be found with the correct sign, a ctrl-vs-ctrl
2835
+ null comparison that must find nothing, a compensation-APPLIER pair
2836
+ (identity is a no-op; the true signal is recovered at every spill level —
2837
+ the estimator checks cannot see an applier bug, which is where the 2.2.1
2838
+ transpose lived), and a clustering pair that scores
2839
+ purity against the true populations then shows it collapsing under a
2840
+ permutation control. **Nothing in the suite is a pinned number**, so there is
2841
+ no baseline to update and nothing to re-freeze — a failure is always a real
2842
+ behaviour change. `--all` runs both suites.
2843
+ Demonstrated to catch two sabotages the golden reports as 7/7 green: a
2844
+ transposed spillover matrix (the v2.2.1 bug class, which corrupted every real
2845
+ compensated analysis) and a hardcoded estimator.
2846
+
2847
+ ### Added
2848
+ - **GPU backend picker** (Preferences -> Performance). `gpu_backend` was read
2849
+ at startup and in the Preferences dialog but written by NOTHING, so the
2850
+ portable PyTorch backend (NVIDIA / AMD / Intel / Apple, incl. DirectML on
2851
+ native Windows) could only be selected through the undocumented
2852
+ `OPENFLO_GPU_BACKEND` environment variable — a shipped feature wired at one
2853
+ end and unreachable from the UI. The picker offers auto / CuPy / PyTorch /
2854
+ off, persists the choice, re-probes immediately and reports the resolved
2855
+ device. A meta-test now fails if any preference is read-but-never-written or
2856
+ written-but-never-read, so this class cannot recur.
2857
+
2858
+ - **Geometric mean and robust CV in the statistics table.** Fluorescence is
2859
+ approximately log-normal, so the geometric mean — not the arithmetic mean —
2860
+ is the central tendency FlowJo reports and papers cite; its absence meant
2861
+ anyone reconciling OpenFlo numbers against FlowJo was comparing different
2862
+ statistics. `GeoMean` is computed on positive values only (compensated data
2863
+ legitimately contains negatives, and clamping them to a floor would bias the
2864
+ result upward). `rCV` is the MAD-based robust CV, which unlike the ordinary
2865
+ CV survives the outliers real data carries. Both are opt-in; the default
2866
+ column set is unchanged.
2867
+
2868
+ ### Fixed
2869
+ - **The golden baseline's compensation metric never ran compensation.** It
2870
+ wrote the generator's own planted spillover matrix to a CSV, read it straight
2871
+ back, and asserted the value it had just written — a metric labelled
2872
+ "Compensation FL2 -> FL8 spill" that could not observe compensation
2873
+ being broken, and which therefore stayed green through the 2.2.1 transpose
2874
+ bug that corrupted every real compensated analysis. It now applies
2875
+ compensation to data with a known spillover baked in and reports two numbers
2876
+ that cannot both be satisfied by a degenerate answer: `residual_err` (max
2877
+ error against the true signal, relative to signal scale) and
2878
+ `signal_retained` (which a "zero everything" implementation would fail while
2879
+ flattering the residual). Verified to catch a transposed matrix, a zeroing
2880
+ implementation, and a no-op. **Baseline keys changed**: `compensation.fl2_leak`
2881
+ is replaced by `compensation.residual_err` + `compensation.signal_retained`.
2882
+ - **`--update` could bless a bug as the baseline.** Re-pinning rewrites every
2883
+ golden value from the current run, so running it while a bug was live
2884
+ recorded the breakage as correct — permanently. It now runs the response
2885
+ controls first and REFUSES if any fail, since a relational control failing
2886
+ means an output has stopped tracking its input and there is nothing
2887
+ legitimate to re-pin. `--force` overrides for a verified intentional change.
2888
+ - **Auto-gate proposals were fit on the display subsample, so they changed on
2889
+ every app restart.** All three methods called `_get_df` with its default
2890
+ `downsample=True`, which caps the frame at the SMALLEST loaded sample's size
2891
+ and draws it with `random_state = hash((name, x, y, cap))`. Python randomises
2892
+ `str` hashing per process, so the same file yielded a different gate each
2893
+ launch — measured seeds for one identical expression across three processes:
2894
+ 1618484138 / 2221425230 / 432000015 (stable only under `PYTHONHASHSEED=0`).
2895
+ Worse, loading a small compensation control alongside a 500k-event sample
2896
+ collapsed the cap to the control's size, fitting the big sample on ~1% of its
2897
+ events; `auto_singlet_gate`'s `frac_kept` / `ratio_cv`, which drive the
2898
+ clean-vs-REVIEW verdict, described that subsample rather than the sample.
2899
+ All three now fit on the full frame — `gmm_ellipse_gates` already does its
2900
+ own seeded 20k draw, and the other two are cheap O(n) statistics that are
2901
+ more accurate on complete data.
2902
+ - **A cross-instance move could delete samples the destination never took.**
2903
+ Pasting a staged move wrote the `<move_id>.done` completion marker the moment
2904
+ the FCS loads were *queued*, and the source treats that marker as authority
2905
+ to delete its copies. Any path the destination skipped or failed to load
2906
+ still cost the source the sample AND its gate tree, while the destination
2907
+ applied nothing. The common trigger was benign: sending a sample to a window
2908
+ that already had that `.fcs` open, which the loader skips as "already
2909
+ loaded" — and the skip notice was itself overwritten by "Pulling N
2910
+ sample(s)", so the move looked successful. The marker is now written only
2911
+ once every accepted path has landed and lists the PATHS actually taken
2912
+ (names are per-instance; collision disambiguation renames them per window);
2913
+ the source removes only those, reports what it kept, and treats an
2914
+ unreadable marker as "took nothing" rather than deleting on ambiguous
2915
+ evidence. Samples already open at the destination are left in place and
2916
+ named in the status line.
2917
+ - **A non-finite value no longer becomes a real, very negative measurement.**
2918
+ The biexponential backends map every non-finite input to the BOTTOM of the
2919
+ scale (-1.0) — `NaN`, `+inf` and `-inf` alike. That is silent corruption
2920
+ twice over: a saturated reading is rendered and gated as the DIMMEST event in
2921
+ the sample, and, worse, a `NaN` becomes a *finite* coordinate, which defeats
2922
+ the `dropna` that protects the plot and the gate masks — so the event goes on
2923
+ to count as a real measurement in every population, median and frequency.
2924
+ `asinh` always propagated `NaN` correctly, so the behaviour also differed by
2925
+ transform. `logicle` and `hyperlog` now match it. Finite values are
2926
+ bit-identical, and the golden baseline is unchanged.
2927
+ - **A non-dict session file no longer crashes startup.** The resume read was
2928
+ guarded, but the `len(data.get('samples', []))` immediately after it was not
2929
+ — so a file whose JSON top level is an array, string or number parsed fine
2930
+ and then raised an uncaught `AttributeError` while the editor was still
2931
+ starting. Latent (nothing in the tree writes such a file), but
2932
+ `_find_resumable_session` appends the legacy `last_session.flowsession`
2933
+ unconditionally, which is how a foreign-format file would arrive.
2934
+ - **A refused resume no longer leaves the session directory pointing at it.**
2935
+ `_session_dir` / `_session_data_dir` were assigned *before* the schema check,
2936
+ so an autosave rejected as "written by a newer OpenFlo" left both addressing
2937
+ the rejected file. They are now set only after the session is accepted.
2938
+ - **`~/.openflo/transfer` is pruned.** A `<move_id>.done` marker is normally
2939
+ consumed by the source instance, but not when that instance has exited or
2940
+ cancelled the move — and nothing ever removed the leftovers, so they
2941
+ accumulated for the life of the install. Now pruned after a day, mirroring
2942
+ `_prune_autosaves`. Zero-byte files, so this is file-count hygiene rather
2943
+ than space.
2944
+ - **The FMO gating dialog states its precondition instead of raising.** Opened
2945
+ without an active sample it did `editor._samples[None]` and died with a bare
2946
+ `KeyError: None` from its constructor. The launcher already refuses in that
2947
+ case, so this was defence in depth rather than a live path — found by a new
2948
+ sweep that drives every dialog in a nearly-empty editor, which is where the
2949
+ other two dialog crashers this cycle also lived.
2950
+ - **One bad event no longer destroys a whole sample's spectral unmixing.**
2951
+ `unmix` solves every event in a single least-squares factorisation, so a
2952
+ single non-finite detector reading contaminated the entire solution: one
2953
+ `inf` event turned **all** events' abundances into NaN. (A `NaN` event
2954
+ happened to stay contained — an accident of how the SVD propagates, not a
2955
+ guarantee.) Non-finite events are now held out and returned as NaN, the
2956
+ remaining events solve exactly, and the count is logged. v2.2.3 hardened the
2957
+ reference spectra against non-finite values; this is the matching guard on
2958
+ the event data, which was overlooked. Non-finite reference spectra now also
2959
+ raise a clear error instead of numpy's opaque "SVD did not converge".
2960
+ - **`.h5ad` export was completely broken and nothing caught it.** Writing an
2961
+ AnnData file raised `RuntimeError: allow_write_nullable_strings is False`
2962
+ before producing anything: pandas 3.x backs string columns with
2963
+ `StringArray`, and anndata >= 0.11 refuses to write those without an opt-in —
2964
+ so the export failed on the observation index alone. Every use of Sample QC →
2965
+ "Export AnnData (.h5ad)…" and of `openflo.write_h5ad` was affected. String
2966
+ columns and indices are now coerced to plain object dtype, which is the
2967
+ representation anndata has always written and does not depend on a global
2968
+ setting or an anndata version. The gap was structural: `to_anndata` was
2969
+ tested in memory while the file write had no coverage at all, so a value that
2970
+ serialised correctly and one that could not be serialised looked identical.
2971
+ There is now a real round-trip test — write a file, read it back with
2972
+ anndata, and check the matrix, sample labels, cluster columns, marker names
2973
+ and recorded `dropped_markers` all survive.
2974
+ - **An unreadable audit entry no longer vanishes from the Methods paragraph.**
2975
+ `methods_paragraph` builds text meant to be pasted into a manuscript, and it
2976
+ silently skipped any audit entry that was not a record (and silently replaced
2977
+ malformed `details` with an empty one). The result was a step the user
2978
+ actually performed being absent from their published description of what they
2979
+ did, with nothing saying so. The drop still happens — a corrupt record holds
2980
+ nothing to recover — but it is now reported inside the paragraph itself,
2981
+ where it cannot be published without being noticed. Clean trails produce
2982
+ clean prose, unchanged.
2983
+ - **`_normalise_groups` is now idempotent.** Normalising an already-normalised
2984
+ group silently discarded per-sample FMO overrides: the first pass flattens
2985
+ `[{'name': 'd', 'fmo_set': 'X'}, 'e']` to `['d', 'e']` and records the
2986
+ override in `sample_fmo`, and a second pass — seeing plain strings —
2987
+ overwrote that map with the group default. The sample would then be gated
2988
+ against the WRONG FMO control, silently. No shipped path normalised twice, so
2989
+ this was latent; but three call sites carry a comment telling the next
2990
+ developer to "normalise first" (the fix for a different bug in the same
2991
+ area), which actively invites a fourth. Overrides now survive a repeat pass,
2992
+ while the group default still reaches samples that lack one.
2993
+ - **Auto-gate now fits on the population it attaches to.** In Display=filter
2994
+ mode `apply_gates_var` is set, so the frame `_get_df` returns is the union of
2995
+ every ENABLED gate's chain — while `_add_gate` parents the result on the
2996
+ SELECTED row, which is `None` (root) when a sample row is selected. Filtering
2997
+ down to lymphocytes and running auto-gate therefore fit ellipses on
2998
+ lymphocytes and attached them at root, where they also captured monocytes and
2999
+ debris, with nothing reporting that the fit domain differed from the applied
3000
+ domain. Auto-gate resolves the destination parent once and fits on exactly
3001
+ that population. `_get_df` gained an explicit `gate_parent` argument for
3002
+ callers that need a specific population rather than "what is on screen".
3003
+ - **A failed autosave on exit is now reported instead of vanishing.** `_on_close`
3004
+ detached the in-app console's log sinks and *then* wrote the autosave,
3005
+ reporting failure with a `print()` into a sink nothing could receive, moments
3006
+ before `destroy()`. Disk full at exit meant the session was lost, the next
3007
+ launch had nothing to resume, and no message ever reached the user. The
3008
+ autosave now runs before the sinks are detached and raises a dialog on
3009
+ failure — the last moment the state still exists. The sidecar-failure flag
3010
+ that `File → Save session` already warns about is honoured here too, so an
3011
+ autosave that silently degraded to raw-FCS-only (losing clusters, embeddings
3012
+ and FMO gates on resume) now says which samples were affected.
3013
+ - **A second Send no longer strands the first batch's move flags.** Staging a
3014
+ new move overwrote `_pending_move` without clearing the previous one, and
3015
+ `_mark_pending_move` only ADDS to the flagged set while both clearing paths
3016
+ clear the *new* batch's names — so the first batch's rows kept the ✄
3017
+ pending-move indicator for the rest of the session, with two 700 ms poll
3018
+ loops running in parallel. A new Send now supersedes the old one properly and
3019
+ cancels its poll.
3020
+ - **A gate replicated to other samples now measures the same population.**
3021
+ With "-> all shown" on (the default), `_add_gate_multi` forced every replica
3022
+ to `parent_id = None` while the active sample's copy kept the selected
3023
+ parent. So with a parent gate selected, sample A got `Singlets` nested under
3024
+ `Lymphocytes` and samples B..N got `Singlets` at ROOT — one gate name
3025
+ measuring a different population per sample, which silently invalidates the
3026
+ cross-sample frequency comparison the fan-out exists to enable. Replicas are
3027
+ re-parented by population PATH (ids are per-sample and meaningless on a
3028
+ target); a sample that genuinely lacks that population still gets a root
3029
+ gate, but is now named in the status line instead of being quietly wrong.
3030
+ - **Auto-gate provenance recorded only the active sample** while the proposal
3031
+ was applied to every displayed one; the audit entry now carries the full
3032
+ applied-to list.
3033
+ - **FlowJo-comparison sweeps share one gate-mask cache.** `compare_wsp` walks
3034
+ every population in a workspace against the same sample, each re-evaluating
3035
+ its whole ancestor chain — the same redundancy already fixed in
3036
+ `population_stats` and `_get_df`.
3037
+ - **Sample QC's `_xy` was assigned only in `_compute`**, after `_D`, so an
3038
+ exception between them left `_D` set and `_xy` missing while `_draw` read
3039
+ both — the same class as the `ui_voltage._chan_lookup` landmine. Initialised
3040
+ at construction and guarded in `_draw`.
3041
+ - **Sample QC crashed when opened with nothing selected.** `_markers()` reads
3042
+ `names[0]` and both of its callers invoked it *before* their own length
3043
+ check, so an empty selection raised an unguarded `IndexError` inside an
3044
+ `after()` callback — swallowed by Tk, leaving the window blank with no error.
3045
+ - **Voltage dialog's `_chan_lookup`** was only assigned inside
3046
+ `_populate_channels`, which returns early when the first FCS is unreadable;
3047
+ `_selected_channel()` reads it and was safe only by call order. Now
3048
+ initialised at construction.
3049
+
3050
+ ### Changed
3051
+ - **Population statistics are 5-7x faster on real gating hierarchies.**
3052
+ `cumulative_gate_mask` gained opt-in memoisation, and `population_stats` now
3053
+ shares one cache across the collection pass, so a gate reuses its parent's
3054
+ cumulative mask instead of re-walking and re-evaluating the whole ancestor
3055
+ chain. This matters because a polygon gate costs ~250x a threshold gate
3056
+ (`Path.contains_points` over every event) and real hierarchies are
3057
+ polygon-based: over 200k events a 32-gate panel drops from 1.82 s to 0.36 s
3058
+ and a 60-gate panel from 4.72 s to 0.67 s. Masks are unchanged — verified
3059
+ identical for nested trees, cycles, missing parents and injected overrides.
3060
+ Callers that omit `cache` keep the previous allocate-fresh behaviour exactly.
3061
+ - **Dragging a 1-D gate no longer rebuilds the gate tree on every mousemove.**
3062
+ The line-gate drag and the histogram slider both rebuilt the whole
3063
+ samples-and-gates tree per tick. That rebuild is O(total rows) across every
3064
+ *loaded FCS file*, not just the gate being dragged — about 8 ms for a
3065
+ 20-file workspace with 30 gates each (621 rows), which at drag rate is
3066
+ roughly half of wall-clock spent rebuilding a tree nobody was looking at.
3067
+ Both now refresh once at end-of-drag, which is what every other drag kind
3068
+ already did (and what the code's own comment said it did). The gate itself
3069
+ still updates every tick, so the plot tracks the cursor as before.
3070
+ - **Replotting is ~23% faster on a multi-file workspace.** `_get_df` OR-ed the
3071
+ cumulative mask of every enabled gate, each re-walking and re-evaluating its
3072
+ whole ancestor chain; it now shares one call-local cache. Profiling an 8-file
3073
+ x 50k-event x 12-gate workspace, `_replot()` drops from 1034 ms to 797 ms and
3074
+ gate-mask time halves. (Profiling also showed matplotlib's `tight_layout` and
3075
+ `loc='best'` legend placement dominate what remains — see the note in
3076
+ DESIGN_NOTES rather than a code change, since legend position is a visual
3077
+ decision.)
3078
+ - **New preference: legend position** (Preferences -> Appearance), and with it
3079
+ a ~1.9x faster replot by default. matplotlib's automatic (`loc='best'`)
3080
+ placement tests candidate corners against EVERY plotted point, so it costs
3081
+ ~1.1 s on 8 samples x 50k events — and it ran on every replot. Three modes:
3082
+ *Automatic (reuse position)* — the new default — solves the placement once
3083
+ per plot shape, remembers where it landed in axes coordinates and reuses it
3084
+ until the axes or sample set change; *Automatic (recompute each time)* — the
3085
+ previous behaviour; and *Always top-right* — cheapest, but can sit on top of
3086
+ data. Measured on an 8-file x 50k-event x 12-gate workspace: 808 ms per
3087
+ replot recomputing, 435 ms reusing, 440 ms fixed — so the default now costs
3088
+ what a fixed corner costs while keeping the placement that automatic
3089
+ mode chose.
3090
+ - `make_compensation_controls` accepts a `leaks` mapping so the planted
3091
+ spillover can be varied (or removed entirely) rather than being fixed.
3092
+ - `immunophenotyping_sample(return_labels=True)` additionally returns the TRUE
3093
+ per-event population, so a clustering result can be *scored* rather than
3094
+ merely counted. The labels ride through the existing shuffle as an extra
3095
+ column — a row-wise Fisher-Yates cannot observe a column, so the seeded RNG
3096
+ stream is untouched and every generator output stays bit-identical (asserted
3097
+ by a test, including the `batch_gain` and `fmo` paths). Reading this as a
3098
+ ground truth revealed that the golden's `leiden_n = 18` against 6 true
3099
+ populations is benign over-clustering, not error: homogeneity is 1.00, so the
3100
+ extra clusters are pure sub-clusters that never mix populations.
3101
+
3102
+ ## [2.3.0] - 2026-09-03
3103
+
3104
+ ### Added
3105
+ - **Theme audit harness** (`scripts/theme_audit.py`). Builds the editor under a
3106
+ chosen theme on a withdrawn root, walks every widget, combobox popdown and
3107
+ figure facecolor, and flags near-white surfaces — so a light-leaking widget is
3108
+ caught mechanically rather than by eye. Wired into the test suite as a Midnight
3109
+ no-white regression.
3110
+
3111
+ ### Changed
3112
+ - **Refreshed GUI chrome ("sleek").** The `light` / `dark` / `midnight` themes
3113
+ are restyled to a hand-tuned, dependency-free look: a near-black / off-white
3114
+ palette with a cool accent, flat surfaces, hairline borders tied to the
3115
+ palette, more generous padding, muted headings and tabs, and accent-on-focus.
3116
+ Check and radio indicators are flat and understated. Theme names and the
3117
+ Preferences picker are unchanged.
3118
+ - **Auto-clean now follows the standard gating order.** The doublet FSC-A/FSC-H
3119
+ ratio median was computed over *all* events, debris included. When debris
3120
+ cleaning is also enabled, the ratio window is now centred on the
3121
+ debris-removed (cell-sized) population, matching the conventional hierarchy.
3122
+ AND-of-methods semantics are unchanged. **This can shift doublet-gate results
3123
+ on runs that enable both methods.**
3124
+ - **The unclustered / noise bucket is kept and named everywhere.** The `-1`
3125
+ sentinel (non-finite-channel events, PhenoGraph outliers, unassignable
3126
+ sub-sample rows) previously surfaced as a bare `-1`, as a `C-1` bar label, or
3127
+ was dropped outright depending on the path. It is now retained and labelled
3128
+ "Unclustered (noise)" — via one shared `pipeline.cluster_label()` used by both
3129
+ the pipeline stats and the GUI — across per-sample frequency exports, the
3130
+ cluster heatmap, group-vs-group comparisons and the annotate dialog. Frequency
3131
+ exports gained a leading `population` label column. Real clusters' counts and
3132
+ percentages are unchanged. The imported noise population is drawn in a neutral
3133
+ grey so it does not read as a real population.
3134
+ - **FlowSOM and cell-cycle unassigned buckets** are likewise kept and labelled
3135
+ ("Unassigned (noise)"; `NA` for cell cycle) instead of being silently dropped,
3136
+ and category populations use friendly names ("FlowSOM metacluster 3") rather
3137
+ than the raw storage token ("flowsom_meta 3").
3138
+ - **Workspace compensation column**: an item with no compensation is flagged
3139
+ with "⚠" alone, now clearly distinct from a group's neutral "—" (no override).
3140
+ - **Prism grouped export** keeps replicates with a missing row/column factor
3141
+ under an explicit `(unassigned)` level instead of dropping them from the table.
3142
+ - **Faster grouped runs.** Preparing a run unit copied every member's full event
3143
+ frame just to tag it with its group and sample — N large copies before the
3144
+ concatenation's own copy, a multi-second UI freeze on a grouped run. The tags
3145
+ are now applied once, after the concatenation. The output is identical and the
3146
+ source frames are no longer touched.
3147
+
3148
+ ### Fixed
3149
+ - **Exported population `.fcs` files could contain the wrong events.** The
3150
+ cumulative-gate walk in the population export swallowed every exception, so a
3151
+ gate that failed to evaluate was silently dropped and the exported file held
3152
+ the *un-restricted superset* — silent data corruption in a file users take
3153
+ downstream. A failed gate now skips that population, and the skip is surfaced
3154
+ in the status line.
3155
+ - **A gate that errored admitted every event.** `_evaluate_gate_on` returned an
3156
+ all-True mask on failure, an inflated superset that read as success. It now
3157
+ fails closed (all-False), so the population empties visibly. Valid gates and
3158
+ the "channel missing → skip" paths are unaffected.
3159
+ - **A failed per-channel transform left that channel in raw scale** among its
3160
+ logicle siblings, so gates and plots compared against raw values. Failures are
3161
+ now recorded and logged prominently, listing the affected channels.
3162
+ - **Preferences could be wiped.** `write_pref` did a read-modify-write, and
3163
+ `read_prefs()` returns `{}` on *any* error — so a single transient file lock
3164
+ (antivirus, search indexer) persisted a one-key file, discarding every other
3165
+ setting. A read error now aborts the write instead of clobbering; only a
3166
+ missing or corrupt file resets. Writes are atomic.
3167
+ - **Sessions are written atomically.** A mid-write failure previously truncated
3168
+ the existing good session or autosave.
3169
+ - **Session sidecar failures were invisible.** A processed-data sidecar write
3170
+ failure was printed to stdout while the save reported clean success — the
3171
+ sample then silently reloaded from raw FCS with its clusters, UMAP and FMO
3172
+ gates gone. The save now warns and names the samples that lost computed
3173
+ results.
3174
+ - **Session display state was lost on a slow resume.** The restore was a single
3175
+ 600 ms one-shot that cleared unconditionally, so any sample finishing later
3176
+ kept its default (unchecked) plot state and lost its saved axis. It now
3177
+ re-applies while samples are still loading.
3178
+ - **`PI3K` was mis-detected as the PI DNA stain.** The `pi` token boundary
3179
+ allowed an adjacent digit, so a PI3K signalling channel was auto-selected as
3180
+ the DNA channel.
3181
+ - **CLI `--gates` threshold overrides were silently dropped** for any group with
3182
+ an empty FMO set — that is, every by-day auto-group — while `--export-wsp`
3183
+ still recorded them, so the two paths disagreed. Overrides now apply to every
3184
+ sample. Malformed or incomplete gate dicts are reported rather than discarded
3185
+ in silence.
3186
+ - **CLI `--gates` help documented a form the parser rejects.** Copying the
3187
+ documented `{"FL1-A":0.5}` example ran the analysis **ungated**; the help now
3188
+ shows the accepted JSON-list form.
3189
+ - **`--export-wsp` resolved zero samples** for a comma-string or per-sample-dict
3190
+ `samples` spec (it iterated the string or dict). It now normalises groups
3191
+ first, honours each group's `trial_dir` so by-day filename collisions pick the
3192
+ right day's FCS, and allocates gate ids in two passes so a child-before-parent
3193
+ gate nests correctly instead of re-rooting. The panel probe had the same
3194
+ un-normalised-groups bug and silently dropped the panel.
3195
+ - **Compliance records verified as valid when a signed file was deleted** — the
3196
+ missing file was dropped from the checked set, and `all([])` is `True`. The
3197
+ check is now fail-closed.
3198
+ - **KDE density crashed** on a single-event or constant-channel gate (it had no
3199
+ degenerate-input guard); it now falls back to flat density so the scatter still
3200
+ draws.
3201
+ - **t-SNE was silently skipped for 3–5 events** — the perplexity clamp floor of
3202
+ 5.0 exceeded *n*, so the call still raised.
3203
+ - **Stale grouping in the frequency dialog.** The Group-by combo, Tokens entry
3204
+ and Parametric checkbox did not trigger a rebuild, so "Diff. abundance…" and
3205
+ "Compare all…" could run the GLM on the previous grouping while the UI showed a
3206
+ new one.
3207
+ - **All-NaN channels produced garbage p-values** presented as valid in group
3208
+ stats; non-finite values are now filtered before the per-group median.
3209
+ - **A clustering worker could be read mid-write.** The plot redraw could observe
3210
+ sample DataFrames while a clustering or embedding worker mutated them off the
3211
+ Tk thread (transient `KeyError` / torn read); the redraw now defers until the
3212
+ run completes.
3213
+ - **Undo could not revert a gate toggle.** Toggling a gate or auto-clean method's
3214
+ enabled checkbox changes real gating output but took no undo snapshot.
3215
+ - **A 1-D gate re-commit silently re-enabled a disabled gate** or dropped its
3216
+ name — the replace path preserved only the colour, not the enabled/name/open
3217
+ state.
3218
+ - **FlowJo comparison could read as valid on uncompensated data.** A
3219
+ compensation failure left every gate evaluated against uncompensated events
3220
+ with no per-row flag; rows now carry an explicit "uncompensated" error.
3221
+ - **Sample-distance MDS collapsed onto the origin.** An undefined pair (no shared
3222
+ usable markers) scored 0.0, which reads as "identical"; it is now `NaN`, and
3223
+ the embedding places those samples at the edge while still returning finite
3224
+ coordinates.
3225
+ - **AnnData and distance-matrix exports** use the shared-marker intersection
3226
+ consistently (heterogeneous samples no longer raise), record which markers were
3227
+ omitted (in `adata.uns['dropped_markers']`) instead of silently shrinking the
3228
+ panel, and add a labelled sibling `obs` column alongside the raw integer id.
3229
+ - **Cross-platform paths and encoding.** Session `rel_path` and `processed_csv`
3230
+ are stored forward-slash and normalised on read, so a Windows-authored
3231
+ `.flowsession` relinks its FCS on Linux and macOS; the voltage-titration CSV is
3232
+ opened as UTF-8 (it was the only text `open()` without it, giving cp1252
3233
+ mojibake on non-ASCII channel names); the watch-folder seen-set is keyed on the
3234
+ exact filename, so two case-distinct files no longer collapse on Linux.
3235
+ - **A legacy workspace with deletion gaps could overwrite an existing item** —
3236
+ the `_mseq` / `_gseq` fallback resumed from the item *count* rather than the
3237
+ maximum id suffix, re-minting a live id.
3238
+ - **Frequency and heatmap exports were hardcoded to the `cluster` column**, so
3239
+ Leiden and FlowSOM label columns got no noise-labelled export; both now take a
3240
+ `label_col` (the default preserves the previous behaviour).
3241
+ - **Smaller dialog and export guards**: Log scale with Min ≤ 0 is rejected
3242
+ instead of producing an empty axis; the FMO percentile is validated in [0, 100]
3243
+ up front (out of range previously surfaced as a misleading "nothing added");
3244
+ quadrant counts guard an empty frame (`ZeroDivisionError`); a 0.0 annotation
3245
+ score with a non-positive threshold no longer raises `KeyError`; a zero-drop
3246
+ gate no longer shows a "drops 0" suffix; non-separable voltage SI/rCV render
3247
+ `n/a` rather than a bare `nan` in both stdout and the CSV; `write_fcs` logs a
3248
+ count of the non-finite cells it zeroed; and a group comparison that produced
3249
+ no rows now says so instead of writing nothing silently.
3250
+ - **Theming completeness.** About 25 hint, status and caption labels moved from
3251
+ hardcoded greys to a palette-tracked `Muted.TLabel` token (dark greys were
3252
+ invisible on the dark panel); free matplotlib artists that the figure theming
3253
+ cannot reach — significance brackets, empty-state text, FlowSOM spokes, and the
3254
+ gating-tree diagram's connectors and node labels — now take theme-correct ink;
3255
+ Treeview bad/warn row tints are theme-matched (previously an unreadable light
3256
+ band under Midnight); a colourless gate row no longer renders black-on-dark;
3257
+ and the populations context menu, the voltage dialog's initial paint and the
3258
+ progress-bar trough no longer leak light. A live theme switch now re-themes
3259
+ already-open dialogs instead of leaving them on the old palette.
3260
+ - **Space utilisation.** The frequency summary gained a vertical scrollbar — long
3261
+ multi-group / BH-pairwise output was previously truncated invisibly in a fixed
3262
+ six-line box — and the compensation spillover matrix stretches to the canvas
3263
+ width instead of jamming into the top-left corner.
3264
+ - **A closed tool window retained its figure and cached DataFrame** until it was
3265
+ reopened; the registry slot is now released on destroy.
3266
+ - **A long-lived log-drain timer** on a destroyed window could reschedule itself
3267
+ forever (a latent leak).
3268
+ - Corrected 14 docstrings and comments that contradicted the code — including
3269
+ `gate_to_mask`'s missing-channel behaviour (all-True for geometric gates,
3270
+ all-False for cluster and category gates), `WspWriter`'s round-trip scope (only
3271
+ boolean is out of scope), the compensation editor's four-step auto-import order
3272
+ and non-modal behaviour, `multi_group_test`'s Friedman truncation, and
3273
+ `_sample_group_label`'s longest-match token rule.
3274
+
3275
+ ### Removed
3276
+ - The unused `tqdm` dependency, which was declared as a core requirement but
3277
+ never imported.
3278
+
3279
+ ### Security
3280
+ - **FlowJo `.wsp` parsing rejects a DTD or `ENTITY` declaration** before parsing,
3281
+ closing the stdlib ElementTree entity-expansion denial of service ("billion
3282
+ laughs" / quadratic blowup). No XXE file read was possible, and no new
3283
+ dependency was added.
3284
+ - **Session relink is contained to the session directory.** `processed_csv` is
3285
+ always app-generated and relative, so absolute or `..`-escaping values are now
3286
+ rejected, closing an arbitrary-file-read vector. `rel_path` still permits
3287
+ legitimate cross-directory projects but warns when it resolves outside the
3288
+ session tree.
3289
+
3290
+ ## [2.2.4] - 2026-07-02
3291
+
3292
+ ### Changed
3293
+ - **Test-coverage hardening — no runtime changes.** A ~40-test sweep from a
3294
+ "looks-tested-but-core-untested" audit now pins the *core behavior* of
3295
+ components that previously had only periphery coverage (parse / IO round-trip /
3296
+ sign / monotonicity) — the same class of gap that let the 2.2.1
3297
+ compensation-transpose bug ship. Newly pinned: FCS per-cell row pairing +
3298
+ external-read fallback, gate interval + multi-gate AND masks +
3299
+ cumulative/override semantics, density event x/y alignment + KDE-vs-SciPy +
3300
+ smoothing floor, spectral autofluorescence + bright-event selection + detector
3301
+ ordering, effect-size magnitude, workspace↔FlowJo comparison arithmetic + XML
3302
+ gate parsing + inventory walk, clustering purity + subsample assignment +
3303
+ non-finite handling, trajectory geodesic property + bin means, CytoNorm
3304
+ per-metacluster proportion preservation + QC magnitude, calibration r² + peak
3305
+ fallbacks, voltage stats (robust CV, arcsinh split, channel resolution). **No
3306
+ source changed; every closure confirmed the existing code correct (zero bugs
3307
+ found).** The compensation-transpose shape of bug would now be caught across
3308
+ the codebase.
3309
+
3310
+ ## [2.2.3] - 2026-07-02
3311
+
3312
+ ### Fixed
3313
+ - **Defensive robustness hardening (code-review sweep).** A batch of guards that
3314
+ turn edge-case crashes / silent-wrong results into graceful handling:
3315
+ - Compensation: a singular spillover matrix is skipped with a warning instead
3316
+ of raising; compensation-QC rejects non-finite (NaN/inf) matrices; a
3317
+ malformed FCS `$SPILL` keyword is reported + ignored rather than silently
3318
+ read as "no compensation".
3319
+ - Spectral unmixing: non-finite events in a single-stain control no longer
3320
+ poison its reference spectrum; a zero-norm spectrum reads self-similarity 1.
3321
+ - CytoNorm: clear errors for `n_quantiles < 2` and for a batch missing a
3322
+ normalized channel (were cryptic downstream crashes).
3323
+ - Session restore: a corrupt per-channel display range is skipped instead of
3324
+ aborting the whole load; auto-saved sessions are schema-migrated on resume
3325
+ (same as a manual open).
3326
+ - FlowJo import: gates on natively-compensated channels (FlowJo's `<FL5-A>`
3327
+ bracket form) now match the data instead of selecting everything; the
3328
+ region/CLI gate filter evaluates all gate kinds (ellipsoid / cluster /
3329
+ category / boolean), not just lines and polygons.
3330
+ - Auto-clean / cell-cycle: no more spurious divide-by-zero warnings; the
3331
+ cell-cycle width-channel lookup matches lowercase `-w`/`-h` columns.
3332
+
3333
+ ## [2.2.2] - 2026-07-02
3334
+
3335
+ ### Fixed
3336
+ - **Compensation-QC "strong pairs"** now includes spillover exactly at the 0.10
3337
+ threshold (a strict comparison previously excluded it).
3338
+ - **Spectral unmixing condition number** reports `inf` for an underdetermined
3339
+ panel (more fluorophores than detectors) instead of a misleadingly-small value.
3340
+ - **Differential abundance** proportions use each sample's true total event count
3341
+ as the library size; nested/overlapping populations previously inflated it and
3342
+ understated the displayed percentages.
3343
+ - **Workspace ↔ FlowJo comparison** no longer collapses populations that share a
3344
+ name (e.g. quadrant Q1–Q4, copied gates) — each is compared against its own gate.
3345
+ - **CytoNorm** models whose batch label contains `|` (e.g. a POSIX path) reload
3346
+ correctly instead of failing.
3347
+ - **Batch run outputs** no longer silently overwrite one another when two run
3348
+ units sanitise to the same filename (colliding labels are suffixed); a failed
3349
+ subprocess launch no longer leaks a temp directory.
3350
+ - **Friedman test** no longer suppresses a perfectly-concordant (maximally
3351
+ significant) result.
3352
+ - **Histogram (symlog view)**: bin spacing and the axis scale now share one
3353
+ anchor, removing uneven bins.
3354
+
3355
+ ## [2.2.1] - 2026-07-01
3356
+
3357
+ ### Fixed
3358
+ - **Compensation was applying the transposed inverse spillover.** The internal
3359
+ apply step computed `data @ inv(M).T` instead of `data @ inv(M)`, which left
3360
+ asymmetric spillover uncorrected and corrupted otherwise-clean channels —
3361
+ affecting every compensated dataset with a non-symmetric spillover matrix (i.e.
3362
+ essentially all real data). **This is a correctness fix: compensated values,
3363
+ and everything derived from them (transforms, clustering, gating, exported
3364
+ populations), will change — for the better.** Re-run compensation on affected
3365
+ analyses; the stored spillover matrix itself is unchanged, so sessions reload
3366
+ fine and simply recompute.
3367
+ - **Differential abundance could report enrichment backwards.** The results table
3368
+ labelled the two groups (and the log2FC sign) by sample-load order while the
3369
+ GLM fitted them in the count matrix's (alphabetical) column order; the table
3370
+ now matches the fitted direction.
3371
+ - **Histogram highlight mode could crash** (`ValueError`) when highlighting two or
3372
+ more gates/samples at once.
3373
+
3374
+ ## [2.2.0] - 2026-07-01
3375
+
3376
+ ### Added
3377
+ - **Reproducible clustering.** PhenoGraph's default Louvain community detection
3378
+ is not seed-reproducible (its community binary is time-seeded), so cluster
3379
+ labels could differ run-to-run. A new opt-in **Reproducible** mode routes
3380
+ PhenoGraph to its seeded Leiden backend so a re-run gives identical clusters —
3381
+ available in the Cluster dialog, the Pipeline Workspace Run bar, and the CLI
3382
+ (`--reproducible`). Off by default, so default Louvain results are unchanged;
3383
+ Leiden, FlowSOM, and all embeddings were already deterministic.
3384
+
3385
+ ### Fixed
3386
+ - **FlowJo import — max-only 1-D gates are no longer dropped.** A `RectangleGate`
3387
+ with only an upper bound (`x < hi`, no min) was silently discarded and its
3388
+ child populations re-parented to the grandparent, quietly loosening every
3389
+ descendant; it now imports as a bounded interval, keeping the constraint.
3390
+ - **Quadrant gates tile the plane.** Events landing exactly on a divider (e.g. a
3391
+ value at 0 on an arcsinh axis) fell into no quadrant; the four quadrants are
3392
+ now a true partition.
3393
+ - **Gate editor — invisible handles are no longer grabbable.** In histogram mode
3394
+ (or for a degenerate polygon) a 2-D gate's hidden handles could be dragged,
3395
+ silently corrupting its bounds; hit-testing is now tied to what's actually
3396
+ drawn. Quadrant shift-click-add also honors compound modifiers (e.g. Ctrl+Shift).
3397
+ - **Auto-clean — freezing a valley-mode debris gate keeps granulocytes.** Freezing
3398
+ or copying it collapsed the 2-D FSC×SSC rescue into a 1-D floor, dropping
3399
+ low-FSC/high-SSC granulocytes; it now pins both thresholds and reproduces the
3400
+ full 2-D cut identically across samples.
3401
+ - **Autosave no longer leaks disk.** Orphaned autosave `_data` sidecar folders
3402
+ (processed event-table CSVs) were never pruned; they are now removed alongside
3403
+ their session file.
3404
+ - **Batch runs report empty clustering as a failure.** A FlowSOM/Leiden run below
3405
+ its event floor produced no labels yet was reported as a successful 0-cluster
3406
+ run; it is now surfaced as an error.
3407
+ - **Loader no longer grows `sys.path`.** Each FCS load prepended a duplicate path
3408
+ entry from the worker threads; the redundant insert was removed.
3409
+
3410
+ ## [2.1.0] - 2026-07-01
3411
+
3412
+ ### Added
3413
+ - **Portable session relink.** Saved sessions now record each sample's file
3414
+ basename and a path relative to the session file, so a project that's been
3415
+ moved, copied, or opened on another machine re-finds its raw FCS once it sits
3416
+ beside the `.flowsession` — resolving stored absolute path → session-relative
3417
+ path → basename in the session folder. Additive to the session format (older
3418
+ sessions keep opening; the fields live inside each sample entry).
3419
+ - **Choose clustering markers.** The Cluster dialog now has a marker picker that
3420
+ restricts PhenoGraph / Leiden / FlowSOM to a chosen subset of fluorochrome
3421
+ channels; leaving all selected keeps the previous "all markers" behaviour.
3422
+ - **Cell-cycle gating controls.** The cell-cycle dialog exposes the doublet-cut
3423
+ strength (`k`) and the singlet-tolerance window, which were previously fixed.
3424
+
3425
+ ### Changed
3426
+ - **Tighter self-test reproducibility contract.** The golden self-test
3427
+ tolerances were tightened to match each metric's real reproducibility — the
3428
+ bit-exact metrics (auto-clean debris/doublets and the compensation spill) to
3429
+ exact, and MESF calibration slope/R² to 1e-3 — so behavioural drift is caught
3430
+ far sooner. Desktop and web baselines are kept in lockstep. Platform-variable
3431
+ metrics (viability, Leiden count) intentionally stay loose.
3432
+
3433
+ ## [2.0.1] - 2026-07-01
3434
+
3435
+ ### Fixed
3436
+ - **Compensated samples now reattach correctly on session reopen.** A sample
3437
+ that was compensated but not clustered/embedded wasn't persisted to the
3438
+ processed-data sidecar, so reopening a session reloaded the raw (uncompensated)
3439
+ FCS and any gate/population drawn in compensated space selected the wrong
3440
+ events. Compensated samples are now persisted like any other computed result.
3441
+
3442
+ ## [2.0.0] - 2026-06-30
3443
+
3444
+ ### Added
3445
+ - **Vendor-portable GPU acceleration (PyTorch backend).** The opt-in load-math
3446
+ acceleration (compensation matmul, logicle LUT interp, arcsinh) now runs on a
3447
+ PyTorch backend alongside CuPy, reaching **AMD (ROCm), Intel (XPU), Apple
3448
+ (MPS), and any Direct3D-12 GPU on Windows via DirectML** — not just NVIDIA.
3449
+ Backend preference `auto|cupy|torch|off` (env `OPENFLO_GPU_BACKEND`);
3450
+ Preferences shows the detected device. Extras: `[gpu-torch]`, `[gpu-dml]`.
3451
+ Still OFF by default with the exact numpy fallback, so the golden baseline is
3452
+ unchanged.
3453
+ - **RAPIDS 26.06 GPU-clustering image.** `docker/Dockerfile.rapids` moves to the
3454
+ RAPIDS 26.06 base, which ships numpy 2.4.6 (our exact pin); cuDF/cuML/cuGraph
3455
+ import, `leidenalg==0.11.0` keeps Leiden golden-exact, golden 7/7 in-container.
3456
+ pandas 3.0.3 remains an upstream cuDF wall (documented in `docs/RAPIDS_SHIM.md`).
3457
+ - **Configurable loader concurrency + priority.** The background file-load pool
3458
+ now sizes itself from your CPU & RAM (instead of a fixed 2), with an override
3459
+ in **Edit ▸ Preferences ▸ Performance** (“Concurrent file loaders”, Auto or
3460
+ 1–8, persisted). The pool is a priority queue: the active / first-rendered
3461
+ sample loads first so its plot appears soonest, and loader threads run at
3462
+ lower OS priority so a big batch doesn't starve the UI.
3463
+
3464
+ ### Changed
3465
+ - **Loading feedback now covers session resume, not just Add-FCS.** Reopening a
3466
+ session paints a muted “⏳ name” row for *every* sample up front (grouped by
3467
+ trial) before any file is read, so a large session fills the tree immediately
3468
+ instead of looking frozen. Processed samples (workspace results with
3469
+ clusters/UMAP) now load on the same background pool as raw FCS — the window
3470
+ stays responsive even with large sidecars — each row swapping to its real
3471
+ entry as it lands. Still-loading rows prefer the trial recorded in the session.
3472
+
3473
+ ### Fixed
3474
+ - **Auto-clean now applies to every selected sample**, not just the
3475
+ active/last-selected one.
3476
+ - **Toggling an auto-clean gate/method refreshes the cleaned-out-events overlay**
3477
+ — previously, in “all” display mode it only redrew gate lines, leaving the red
3478
+ removed-events dots stale.
3479
+ - **Right-click menu rendering on dark themes** — the context menu is now themed
3480
+ (incl. `disabledforeground`), so a greyed-out Paste no longer looks garbled.
3481
+ - **Auto-clean removal is clearer** — the menu item reads “Remove auto-clean
3482
+ gate”, and the debris method row shows its effective cut
3483
+ (`[beads]` / `[valley]` / `[beads→valley: no bead file]`) so switching modes
3484
+ with no bead file in the run isn't a silent no-op.
3485
+
3486
+ ## [1.6.0] - 2026-06-25
3487
+
3488
+ ### Added
3489
+ - **Help ▸ Run diagnostics…** — an install health check for when something
3490
+ behaves oddly or an install looks corrupted. Reports whether core
3491
+ dependencies match their pinned versions, which optional engines are present,
3492
+ and whether the seeded behavioural self-test still reproduces the golden
3493
+ baseline. It runs in a *separate* process, so a genuinely broken install (or a
3494
+ native-library crash) is reported instead of taking the editor down. Also
3495
+ available standalone for when the GUI won't start: `openflo-doctor`,
3496
+ `python -m openflo.diagnostics` (`--json` / `--quick`), or
3497
+ `scripts/diagnose.bat` / `scripts/diagnose.sh`.
3498
+
3499
+ ### Changed
3500
+ - **Faster startup.** The Pipeline Workspace panel (~100 ms to build) is now
3501
+ constructed lazily on first reveal instead of eagerly at window open, so a
3502
+ session that never opens it doesn't pay for it.
3503
+ - **Loading feedback for multi-file / large loads.** Every queued FCS now shows
3504
+ as a muted “⏳ name” row in the samples tree *immediately* and stays there
3505
+ until that file finishes — previously the placeholders were wiped on the first
3506
+ file's load and samples popped in one-by-one, which made a big load look
3507
+ stalled. The first sample still renders as soon as it's ready; the rest remain
3508
+ visibly “loading” (with the existing N/N progress bar) until each lands.
3509
+
3510
+ ### Tested
3511
+ - New `tests/test_editor_mixins.py` guards the editor mixin decomposition:
3512
+ every `editor_*` module imports standalone, every `*Mixin` is actually mixed
3513
+ into `ViewGateEditorWindow`, no two mixins shadow each other's method names,
3514
+ and the `gui` back-compat re-exports stay present. New `tests/test_diagnostics.py`
3515
+ covers the health check.
3516
+
3517
+ ## [1.5.0] - 2026-06-25
3518
+
3519
+ ### Changed
3520
+ - **Internal: `ViewGateEditorWindow` fully decomposed into mixins (no behaviour
3521
+ change).** The remaining ~190 editor methods moved out of `gui.py` into 24
3522
+ focused `editor_*` mixin modules (analysis, tools, autoclean, autogate,
3523
+ clipboard, update, figure, slider, template, drag-drop, chrome, gate-tools,
3524
+ compute, stats, populations, load-pool, downsample, mode, grouping, channels,
3525
+ lifecycle, audit; plot renderers folded into `PlotMixin`, undo checkpoints
3526
+ into `UndoMixin`). All subclass the shared `editor_base.EditorMixin`. `gui.py`
3527
+ is now ~1.8k lines (was ~7.4k): the constructor, two class-attribute
3528
+ classmethods, and tooltip glue. `messagebox`/`filedialog`/`_LOAD_POOL_SIZE`
3529
+ are re-exported from `gui` for back-compat. Pyright 0 errors, ruff clean, full
3530
+ suite green (745 passed), golden baseline 7/7.
3531
+
3532
+ ## [1.4.5] - 2026-06-25
3533
+
3534
+ ### Changed
3535
+ - **Internal: editor decomposition continues (no behaviour change).** The
3536
+ log/console pane and the Help-menu dialogs moved out of `ViewGateEditorWindow`
3537
+ into `editor_console` / `editor_help` mixins. (Further editor mixins in
3538
+ progress.)
3539
+
3540
+ ## [1.4.4] - 2026-06-25
3541
+
3542
+ ### Changed
3543
+ - **Internal: `gui.py` decomposed (no behaviour change).** The ~16k-line GUI
3544
+ monolith was reduced ~26% by extracting pure logic and self-contained
3545
+ windows into focused modules: `ui_logic`, `gating`, `tree_ids`, `plotmath`,
3546
+ `density`, `scales`, `paths` (pure, headless-tested), `prefs` + `theme`
3547
+ (shared palette/figure/prefs helpers), and ~24 `ui_*.py` dialog/window
3548
+ modules. Dialog modules now depend on the small shared modules rather than
3549
+ importing the whole GUI — faster imports, cleaner dependency graph. Pinned
3550
+ ruff/pyright so local and CI lint identically.
3551
+
3552
+ ### Fixed
3553
+ - Voltage-optimization dialog rendered a white plot on first open under the
3554
+ Midnight theme (now themed at build time).
3555
+ - Compensation-matrix values were near-illegible on Midnight (theme-aware
3556
+ cell colours; zeros muted, used values carry the header colour).
3557
+
3558
+ ## [1.4.3] - 2026-06-25
3559
+
3560
+ ### Fixed
3561
+ - **Help → Environment no longer freezes the UI.** The engine probe used to
3562
+ *import* each backend (umap/phate/… are slow to import) on the Tk thread; it
3563
+ now checks presence with `find_spec` and reads versions from metadata — no
3564
+ heavy import. Git SHA for the provenance stamp is cached too.
3565
+ - **Pop-up figures honour the Midnight theme.** Voltage, Trajectory, and the
3566
+ other analysis dialogs rendered a white plot under the dark Midnight theme;
3567
+ dark pop-ups now follow either the "Dark figures in pop-ups" toggle *or* the
3568
+ Midnight theme.
3569
+ - **Display modes greyed out without real gates.** "Highlight gated" / "Filter
3570
+ to gated" are disabled (and a stale selection falls back to "All events")
3571
+ when the active sample has no positive gates, so they can't keep drawing
3572
+ gates that were deleted. Auto-clean gates don't count as real gates.
3573
+
3574
+ ### Added
3575
+ - **Help → Environment** — lists which analysis engines (FlowIO, UMAP,
3576
+ PhenoGraph, Leiden, TriMap, PaCMAP, PHATE, AnnData, drag-and-drop, …) are
3577
+ installed, with version and a copy-paste `pip install 'openflo[extra]'` hint
3578
+ for anything missing — so a greyed-out method or skipped run is explained.
3579
+ - **Provenance footer on exported figures** — every saved figure carries a
3580
+ subtle `OpenFlo <version> (<git sha>)` stamp for reproducible, paper-ready
3581
+ output. Toggle in Edit → Preferences → Export.
3582
+
3583
+ ## [1.4.2] - 2026-06-24
3584
+
3585
+ ### Added
3586
+ - **Session format versioning + auto-migration.** `.flowsession` files now
3587
+ carry a schema version; opening an older one auto-upgrades it (with a status
3588
+ note), and one written by a newer OpenFlo is refused rather than mis-read.
3589
+ New **File → Upgrade saved session…** and a headless
3590
+ `scripts/migrate_session.py` upgrade files without opening them.
3591
+ - **Save-format continuity test** (`tests/test_session_continuity.py`) locks
3592
+ the session schema (keys + version) against `openflo.session_format`, so a
3593
+ downstream-visible format change fails the suite until it's made
3594
+ intentional: bump the version, add a migration, and note it here.
3595
+ - **Newer-version alerts for workspaces & recipes.** Saved workspaces and run
3596
+ recipes now record the OpenFlo version that wrote them; loading one produced
3597
+ by a newer build (newer schema or newer app version) warns that some
3598
+ features may not load and suggests updating OpenFlo (Help → Check for
3599
+ updates…).
3600
+
3601
+ ## [1.4.1] - 2026-06-24
3602
+
3603
+ ### Added
3604
+ - **Keyboard shortcuts across the gating loop** — `Ctrl+F` find,
3605
+ `Ctrl+0` reset view, `Ctrl++`/`Ctrl+-` zoom, `F5` replot, `Esc` cancel zoom
3606
+ tool, `Ctrl+1/2/3` display mode (all / highlight / filter), `Ctrl+,`
3607
+ Preferences, `Ctrl+Shift+S` save plot image, `F9` Pipeline Workspace,
3608
+ `` Ctrl+` `` log/console, `Ctrl+T` Statistics. Menu accelerators and the
3609
+ Help → Keyboard shortcuts reference updated to match.
3610
+ - **Swap X↔Y axes** — a `⇄` button between the axis pickers; each axis keeps
3611
+ its own scale/range.
3612
+ - **Selected-gate readout** — selecting a gate shows its event count and
3613
+ **% of parent** (% of all events for a root gate) in the status bar;
3614
+ selecting a sample shows its total event count.
3615
+ - **Type-to-filter channel pickers** — the X / Y / Color combos narrow as you
3616
+ type and snap to the matching channel on commit (helps with large panels).
3617
+ - **First-run empty state** — the empty canvas offers clickable starting
3618
+ points (Add FCS / Load example / Open session) and a drag-and-drop hint.
3619
+
3620
+ ### Fixed
3621
+ - **Clear gate** now hints "Ctrl+Z to undo" in the status bar.
3622
+
3623
+ ## [1.4.0] - 2026-06-24
3624
+
3625
+ ### Added
3626
+ - **Pipeline Workspace v2** — batch co-embedded clustering over groups of
3627
+ samples, now with:
3628
+ - **Clustering method** dropdown (PhenoGraph / Leiden / FlowSOM) with a
3629
+ **Leiden resolution** control and FlowSOM meta-cluster count.
3630
+ - **Full embedding set** — UMAP / t-SNE / PHATE / TriMap / PaCMAP, run
3631
+ together or selectively, with optional concatenation across the group.
3632
+ - **Save / Load recipe** — persist a run configuration as JSON and reload it.
3633
+ - **Import results as populations** — load a processed run's events CSV back
3634
+ into the editor as a sample.
3635
+ - **Batch over folder** — run the active recipe across every FCS in a folder.
3636
+ - **Marker picker** and **per-group parameter overrides**.
3637
+ - **Watch folder** — auto-load new FCS files as they appear.
3638
+ - **CLI clustering parity** — `--cluster-method`, `--resolution`,
3639
+ `--n-metaclusters` for Leiden / FlowSOM / PhenoGraph runs.
3640
+
3641
+ ### Changed
3642
+ - **UI polish** — toolbar buttons stack into two rows (no overflow on wide
3643
+ screens), shortened the workspace tree's sample/population column, renamed
3644
+ the plot-controls **Workspace** button to **Pipeline**, and re-laid-out the
3645
+ group-parameters dialog on a clean grid.
3646
+
3647
+ ### Fixed
3648
+ - Right-click menu **Paste** label rendered garbled on dark menus (themed
3649
+ `disabledForeground`).
3650
+ - Interior pane-resize lag — the matplotlib canvas no longer re-rasters on
3651
+ every pixel of a sash drag; it freezes during the drag and does one clean
3652
+ replot on release.
3653
+ - Dragging a pane no longer exposes a white strip — the canvas backing matches
3654
+ the chrome background.
3655
+
3656
+ ## [1.3.0] - 2026-06-24
3657
+
3658
+ ### Added
3659
+ - **Backend workflows surfaced in the GUI**
3660
+ - **Voltage optimization** (Tools) — PMT / stain-index titration with
3661
+ per-channel recommendations.
3662
+ - **Compare FlowJo workspace** (Tools) — re-apply a `.wsp` and compare gate
3663
+ counts vs FlowJo, with CSV export.
3664
+ - **Generate dataset** (File) — synthetic datasets (PBMC / differentiation /
3665
+ cell-cycle / spectral / beads), loaded in-app.
3666
+ - **Quick preview** (File) — raw single-sample density-scatter QC.
3667
+ - **FCS inspector** (Tools) — raw channels / keywords / spillover viewer.
3668
+ - **Plot navigation** — Zoom-to tool (drag a rectangle; greys the gating
3669
+ tools while active), centered ⌂/⛶/+/- bar, middle-drag pan, wheel zoom.
3670
+ - **Dark figures in pop-ups** (View) — preview + export of every analysis
3671
+ figure window on a dark background; plus a **Midnight** dark-plot theme and a
3672
+ **New windows open at** corner toggle.
3673
+ - **Flow-cytometry tools**
3674
+ - **Singlet gate** (Edit → Add singlet gate) — FSC-A vs FSC-H singlet
3675
+ polygon from the robust height/area band.
3676
+ - **FMO gating** (Edit → FMO gating…) — map markers to FMO controls; places
3677
+ threshold gates at the FMO percentile.
3678
+ - **Compensation QC** (Tools → Compensation QC…) — spillover heatmap +
3679
+ metrics for the active sample's matrix.
3680
+ - **Absolute counts** (Tools → Absolute counts…) — counting-bead cells/µL.
3681
+ - **Gating-tree diagram** (Tools → Gating tree diagram…).
3682
+ - **Embedding comparison** (Analyze → Compare embeddings…) — UMAP / t-SNE /
3683
+ PHATE side by side.
3684
+ - **Research / stats**
3685
+ - **Group comparison** (Analyze → Group comparison…) — Kruskal-Wallis +
3686
+ pairwise Mann-Whitney (BH) + Cliff's δ across trial groups.
3687
+ - **Methods & provenance** (Analyze → Methods & provenance…) — a paper-ready
3688
+ methods paragraph (from the audit trail + citations) and a reproducibility
3689
+ run manifest.
3690
+ - **Export populations as FCS** (Tools → Export populations (FCS)…) — each
3691
+ gated population to its own FCS 3.1 file.
3692
+ - **App**
3693
+ - **Preferences** dialog (Edit → Preferences…); **Documentation** and
3694
+ **Keyboard shortcuts** in Help.
3695
+ - **Plot pan/zoom** — middle-drag pans, scroll-wheel zooms (left-click stays
3696
+ gating); View → Reset plot view.
3697
+ - **Find box** above the sample/gate tree; **periodic autosave** (5 min).
3698
+
3699
+ ## [1.2.4] - 2026-06-24
3700
+
3701
+ ### Added
3702
+ - **File → Load example dataset** — generates and loads a small synthetic
3703
+ PBMC dataset (2 groups × 2 donors), so OpenFlo can be tried with no FCS
3704
+ files of your own.
3705
+ - **File → Save plot as image…** — export the current plot directly to
3706
+ PNG / SVG / PDF (white background, 300 dpi).
3707
+
3708
+ ## [1.2.3] - 2026-06-24
3709
+
3710
+ ### Added
3711
+ - **Global error handling.** Unhandled UI errors now flag the status bar and
3712
+ auto-reveal the log/console (instead of failing silently). A **tokenised**
3713
+ error report (Help → Report a problem…) is written for submission: file
3714
+ paths, sample names, usernames and emails are replaced with stable tokens,
3715
+ and the token→value key is kept in a separate LOCAL file that is never meant
3716
+ to be submitted.
3717
+ - **Keyboard shortcuts** with menu accelerators: Ctrl+O (open session),
3718
+ Ctrl+S (save), Ctrl+E (export .wsp), Ctrl+W (close), Ctrl+Shift+A (add FCS),
3719
+ F1 (About).
3720
+ - **Window size/position** is remembered across launches (validated on-screen).
3721
+ - **File → Open Recent** — the last sessions you opened or saved.
3722
+
3723
+ ## [1.2.2] - 2026-06-24
3724
+
3725
+ ### Added
3726
+ - **One-step setup scripts** (`setup.bat` / `setup.sh`) that create the `.venv`
3727
+ and install OpenFlo + all dependencies; the `openflo-gui` launchers run them
3728
+ automatically on first launch if the environment is missing.
3729
+
3730
+ ### Fixed
3731
+ - Startup session-restore no longer hard-crashes when the data dependencies
3732
+ (FlowIO, etc.) aren't installed — it reports the missing dependency clearly
3733
+ and opens an empty session instead of failing the whole window.
3734
+
3735
+ ## [1.2.1] - 2026-06-24
3736
+
3737
+ ### Added
3738
+ - **Light / Dark / Midnight themes** (View → Theme), persisted across
3739
+ sessions. Light and Dark keep the scatter/plot light (flow-cytometry norm);
3740
+ **Midnight** darkens the plot canvas too — figure, axes, ticks, labels,
3741
+ spines, grid, legend and the backgate legend.
3742
+ - **App icon** — a flow-cytometry density-scatter mark replaces Tk's default
3743
+ feather in the title bar / taskbar.
3744
+ - **Dropdown-menu help** in the status bar (per entry, as you navigate).
3745
+ - **Resizable, pop-out panels.** Samples & Gates | Plot | Pipeline Workspace
3746
+ are draggable panes, and both side panels float into their own window and
3747
+ re-dock.
3748
+ - **Hover tooltips** on the plot controls, gate tools, and action buttons,
3749
+ toggleable via View → Show hover tips.
3750
+ - **Per-population density scaling** for overlays, with a clickable backgate
3751
+ legend (on/off · density · colour) that is draggable and collapsible.
3752
+ - Cell-cycle results group under a collapsed container and persist across
3753
+ session save/restore.
3754
+
3755
+ ### Changed
3756
+ - **Mode** and **Downsample** are now dropdowns; mode-specific options
3757
+ (KDE / contour scatter & outliers / Hist-Y) appear only when relevant, and
3758
+ **Max points** is shown and applied only while downsampling is enabled.
3759
+ - Gate-tree heading expands/collapses all groups; control bars regrouped so
3760
+ each section aligns to its column.
3761
+
3762
+ ### Fixed
3763
+ - Session results (clusters / UMAP) recover after a dropped processed-data
3764
+ sidecar pointer; backgating clustered populations works through the new
3765
+ collapsed group containers.
3766
+ - Downsampling **Off** now truly draws every event (Max points no longer
3767
+ silently caps when downsampling is off).
3768
+ - Refreshed the README/limitations (Auto-gate offers reviewable scored
3769
+ proposals; it is not disabled) and added a prominent citation request for
3770
+ research use (README banner + About dialog; MIT unchanged). Renamed
3771
+ LICENSE → LICENSE.txt so it opens with a double-click.
3772
+
3773
+ ## [1.1.0] - 2026-06-23
3774
+
3775
+ ### Added
3776
+ - **Built-in template library picker (ease-of-use).** The editor's template
3777
+ button is now a **Templates ▾** menu that lists every bundled template by its
3778
+ friendly name (the `cleanup_*` recipes first) plus your own saved templates —
3779
+ apply one in a click, no file navigation. The curated library now ships
3780
+ *inside* the package (`openflo/template_library/`, package data) so it's
3781
+ available to installed users, not just source checkouts; user-saved templates
3782
+ still live in the editor's writable dir and shadow same-named shipped ones.
3783
+ - **One-click cross-group comparison + volcano plot.** The Frequencies window
3784
+ gains a **Compare all…** button that compares *every* population across the
3785
+ current grouping in a single pass (instead of stepping through populations one
3786
+ at a time), Benjamini-Hochberg-correcting across populations. Results open in a
3787
+ new window with a sortable table (per-group means, log2 fold-change, adjusted
3788
+ p, stars) beside a **volcano plot** (log2FC vs −log10 adjusted-p, significant
3789
+ populations highlighted and labelled), with full-table CSV and figure export.
3790
+ New pure `openflo.stats.compare_all_features` (runs `compare_groups` over all
3791
+ features, BH across them) and `volcano_data`, both exported. The volcano needs
3792
+ the two-group case; with >2 groups the table still shows the omnibus
3793
+ Kruskal-Wallis / ANOVA result.
3794
+ - **End-user self-test + seeded data generator (regression baseline for
3795
+ everyone).** Two new console entry points let users — not just contributors —
3796
+ reproduce and regression-check behavior on data they don't have to provide:
3797
+ `openflo-synth` writes the full seeded synthetic dataset (now including the
3798
+ `beads/` size-calibration file), and **`openflo-selftest`** runs that data
3799
+ through the core feature paths (auto-clean debris/viability/doublets, Leiden
3800
+ clustering, MESF calibration, compensation) and compares each metric to a
3801
+ committed golden baseline (`openflo/_golden.json`), printing a PASS/FAIL table
3802
+ — so after pulling an update or editing code you can instantly see whether any
3803
+ feature's behavior changed (`--update` refreshes the baseline after an
3804
+ intended change; `--json` dumps raw metrics). The same golden file backs the
3805
+ pytest continuity tests, so the CLI and CI share one source of truth. Tests
3806
+ ship in the sdist (`MANIFEST.in`). New `openflo.selftest`.
3807
+ - **Bead-calibrated debris removal + dead-cell (viability) auto-cleaning.** The
3808
+ auto-clean gate's **Debris** method now defaults to an *absolute-size* cut:
3809
+ when a size-calibration bead sample is loaded (name contains bead / rainbow /
3810
+ calibration), its median FSC-A anchors a µm ruler and events below
3811
+ `min_um` (default 4 µm, bead diameter default 8 µm) are dropped
3812
+ (`FSC-A ≥ min_um · bead_FSC / bead_um`) — a reproducible absolute-size ruler
3813
+ that, with a sub-cell `min_um` (≈4 µm), removes only genuine sub-cellular
3814
+ fragments and keeps small real cells (lymphocytes). With no bead file it
3815
+ falls back to a **2-D FSC-A × SSC-A scatter gate** matching the standard
3816
+ manual debris polygon (debris = low FSC AND low SSC; granular low-FSC/high-SSC
3817
+ cells are rescued when they form a separate lobe) — never a 1-D cut that would
3818
+ bisect a real population. A new **Dead cells (viability dye)** method
3819
+ finds the live/dead stain by name (`find_viability_channel`: Live/Dead,
3820
+ Zombie, Ghost, FVS/FVD, 7-AAD, PI, DAPI, …) and drops the high-signal dead
3821
+ population at a genuine bimodal valley (`_bimodal_valley`; no-op on an
3822
+ all-live, unimodal sample). Right-clicking the Debris or Dead-cells method
3823
+ rows switches mode (Beads ↔ Auto valley), sets bead / min size, re-detects
3824
+ the bead reference, or pins the viability channel. (FSC-A stays linear and
3825
+ the dye is logicle-transformed in the editor, so both cuts are
3826
+ scale-correct.) `openflo.pipeline.find_viability_channel`. The synthetic
3827
+ dataset gains a `beads/` size-calibration file (single tight 8 µm population
3828
+ matched to a real instrument's FSC scale; `openflo.synthetic.size_bead_sample`
3829
+ / `make_size_beads`) so bead-mode debris is testable headlessly, plus locked
3830
+ continuity reference drops (seeded ≈7 % debris / 8 % dead / 5 % doublets) that
3831
+ flag any future change to the cleaning maths. A method that removes **nothing**
3832
+ now explains why on its tree row (`autoclean_method_diagnostic`: "no viability
3833
+ dye detected", "FSC-A is unimodal — no low-debris mode; load size beads",
3834
+ "the high-signal population is the majority — not treated as dead") instead of
3835
+ a silent 0-drop. Both the debris valley and the viability split use the strict
3836
+ bimodal-valley detector, so a **unimodal** channel is never bisected (a clean
3837
+ single population — beads, a comp control, a pre-gated sample — correctly
3838
+ yields 0 drops rather than a spurious half-cut).
3839
+ - **Compliance / sign-off layer (tamper-evident, 21 CFR Part 11-style).** On
3840
+ top of the audit trail, the **History** window gains **Sign & export record…**
3841
+ and **Verify record…**. Signing builds an integrity *manifest* — SHA-256 of
3842
+ every loaded data file plus a hash of the audit trail and the software
3843
+ version — and attaches an **electronic signature** (signer, meaning, time)
3844
+ bound to that manifest's hash; it writes a signed JSON record + a Markdown
3845
+ copy. Verifying re-hashes everything and flags any signature whose content
3846
+ changed after signing (data edited, audit altered) — so the record is
3847
+ tamper-evident. New pure `openflo.compliance` (`build_manifest`,
3848
+ `sign_manifest`, `verify_record`, `record_to_markdown`), exported. (Scope:
3849
+ tamper-evidence + attributable sign-off, not access control — it complements,
3850
+ not replaces, a controlled-access environment.)
3851
+ - **Fluorescence calibration to standardized units (MESF / ABC).** A
3852
+ **Calibration…** dialog detects the bead-population peaks in a channel
3853
+ (k-means on log intensity), takes each peak's assigned MESF/ABC value from
3854
+ the bead datasheet, fits `value = slope·MFI + intercept` (with R²), and
3855
+ applies it across all samples as a plottable `MESF:<marker>` column — the
3856
+ fluorescence sibling of the existing FSC→µm bead-size calibration. New pure
3857
+ `openflo.calibration` (`detect_bead_peaks`, `fit_mesf_calibration`,
3858
+ `apply_calibration`), exported. The synthetic dataset now includes a
3859
+ `calibration/` rainbow-bead FCS + `mesf_peaks.csv`.
3860
+ - **t-SNE and PHATE embeddings.** The Cluster dialog's UMAP checkbox is now an
3861
+ **Embedding** picker — UMAP / t-SNE / TriMap / PaCMAP / PHATE / none. t-SNE
3862
+ ships in core deps (scikit-learn; perplexity auto-clamped, subsampled); PHATE
3863
+ (diffusion-based, great for continuous / trajectory structure) is an optional
3864
+ `embed` extra. New `FlowSample.run_tsne` / `run_phate` write `TSNE1/2` /
3865
+ `PHATE1/2`; the view switches to the chosen embedding's axes after clustering
3866
+ (only if it produced columns, so an uninstalled backend degrades gracefully).
3867
+ - **Sample QC (EMD + MDS) and AnnData interop.** A **Sample QC…** window
3868
+ computes a pairwise **Earth-Mover's-distance** matrix between the enabled
3869
+ samples (mean over markers of the 1-D Wasserstein distance, pooled-SD scaled)
3870
+ and an **MDS** embedding — batch effects and outlier samples show up as
3871
+ separated points (coloured by trial). Exports the distance matrix, the
3872
+ figure, and an **AnnData `.h5ad`** (events × markers, with `sample` + any
3873
+ `leiden`/`cluster`/`flowsom_meta`/`pseudotime` columns in `obs`) for the
3874
+ scanpy / single-cell Python ecosystem. New pure `openflo.interop`
3875
+ (`sample_distance_matrix`, `mds_embed`, `to_anndata`, `write_h5ad`); AnnData
3876
+ is an optional `interop` extra (`pip install 'openflo[interop]'`).
3877
+ - **FlowSOM star-tree visualization.** A **SOM tree…** button draws the iconic
3878
+ FlowSOM plot: the SOM nodes laid out on their **minimal spanning tree**, each
3879
+ rendered as a **star glyph** of its per-marker prototype profile, coloured by
3880
+ metacluster, with node size ∝ event count — plus a reference star (marker →
3881
+ spoke) and a metacluster legend. PNG/PDF/SVG export. New pure
3882
+ `openflo.pipeline.flowsom_mst` / `flowsom_layout` (scipy MST + igraph layout,
3883
+ exported).
3884
+ - **Rigorous differential abundance (negative-binomial GLM).** A **Diff.
3885
+ abundance…** button in the Frequencies window runs a diffcyt-DA-edgeR-style
3886
+ test: each population's per-sample counts are modelled with a negative-
3887
+ binomial GLM, `log(library size)` as offset (so it accounts for sequencing-
3888
+ depth / composition), a shared method-of-moments dispersion (edgeR
3889
+ common-dispersion-style, stable with few samples), a Wald test on the group
3890
+ coefficient and BH correction — replacing Mann-Whitney-on-fractions for the
3891
+ abundance question. Results table (log2FC, per-group %, p, adjusted p, stars)
3892
+ with CSV export. New pure `openflo.diffexp.differential_abundance` (scipy
3893
+ only — no statsmodels), exported.
3894
+ - **Automated population annotation (MEM + reference table).** An
3895
+ **Annotate…** window turns numeric clusters into biological labels.
3896
+ **MEM** (Marker Enrichment Modeling, Diggins 2017) computes a quantitative
3897
+ per-marker enrichment score for each cluster vs the rest (capturing both the
3898
+ median shift and the IQR change), yielding labels like `CD3+5 CD4+3 CD8-6`.
3899
+ A **reference cell-type table** (`CD4 T: CD3+ CD4+ CD8-`, ACDC/Scyan style)
3900
+ then assigns each cluster a best-matching name (weighting the defining
3901
+ positive markers so a shared negative can't win), written back onto the
3902
+ populations and the cluster-label store. Exports the MEM table. New pure
3903
+ `openflo.annotate` (`mem_scores`, `mem_label`, `population_states`,
3904
+ `parse_signature_table`, `annotate_by_reference`), exported.
3905
+ - **Synthetic example dataset generator** (`openflo.synthetic` +
3906
+ `scripts/make_synthetic_dataset.py`). A generic, regenerable dataset — not
3907
+ tied to any one study — that between its sub-datasets exercises every feature:
3908
+ a **PBMC immunophenotyping** set (CD3/CD4/CD8/CD19/CD56/CD14 lineages, the
3909
+ marquee generic example) for gating / clustering / Leiden / UMAP / frequencies
3910
+ / expression / heatmap / report; a **3-batch variant** with a technical gain
3911
+ shift for CytoNorm batch correction; **FMO controls**; a **cell-cycle**
3912
+ (DNA-content G1/S/G2-M) set; **conventional-compensation** single-stain
3913
+ controls with a known spillover matrix + a sibling `compensation.csv`; the
3914
+ **differentiation** time-course for trajectory; and **spectral** controls for
3915
+ unmixing/QC. Pure (numpy/pandas; FlowIO to write FCS), tested, and gitignored
3916
+ output.
3917
+ - **One-click analysis report (HTML).** An **Analysis report (HTML)…** button
3918
+ bundles the whole session into a single, portable, self-contained `.html`
3919
+ file (images embedded as base64 data URIs — no sidecar files): metadata
3920
+ header, sample & gate summary, the current plot, the population-statistics
3921
+ table, a cluster × marker median-expression heatmap (column z-scored, when a
3922
+ `leiden` / `cluster` / `flowsom_meta` column exists), and the full provenance
3923
+ / audit trail. Opens in the browser on save. New pure `openflo.report`
3924
+ (`build_html_report`, `df_to_html_table`, `figure_to_data_uri`), exported.
3925
+ - **Leiden clustering.** The current field-standard for high-dimensional
3926
+ spectral cytometry, alongside the existing Phenograph and FlowSOM. The
3927
+ Cluster dialog gains a **Leiden** method with a **resolution** control
3928
+ (higher → more, finer clusters); it builds a shared-nearest-neighbour
3929
+ (Jaccard) graph — the Phenograph/Seurat construction, so communities track
3930
+ real populations — and partitions it with `leidenalg` (RBConfiguration).
3931
+ Writes a ``leiden`` column imported as populations; large samples are graph-
3932
+ partitioned on a subsample and the rest assigned by nearest neighbour.
3933
+ `FlowSample.run_leiden`. (`igraph` + `leidenalg` were already declared
3934
+ dependencies.)
3935
+ - **Export gated population as FCS.** Right-click any gated population →
3936
+ *Export population as FCS…* writes that population's events to a standalone,
3937
+ re-importable `.fcs` (FlowJo / FCS Express). Exports the sample's **raw**
3938
+ detector values when they're row-aligned with the gated events (so the file
3939
+ isn't in transformed coordinates), else the processed data, and carries the
3940
+ antibody labels through as `$PnS`. New pure `openflo.pipeline.write_fcs`
3941
+ (FlowIO-backed, exported) zeroes non-finite cells and supports
3942
+ channel subset/reorder.
3943
+ - **Marker-expression distributions (violin / ridgeline) by group.** An
3944
+ **Expression…** window pools each enabled sample's per-cell values for a
3945
+ chosen marker (resolved across fluors by antibody label), groups samples by a
3946
+ factor (trial/day, comp-vs-samples, or a name token), and draws a **violin**
3947
+ or **ridgeline** plot per group. Significance comes from a per-SAMPLE-median
3948
+ comparison (each sample a replicate, not each cell) — also what the GraphPad
3949
+ Prism Column export contains. New pure `openflo.stats.group_kde` (KDE per
3950
+ group over a shared grid) backs the ridgeline and is exported/tested.
3951
+ - **Trajectory / pseudotime (GUI + backend).** A **Trajectory…** tool orders
3952
+ cells along a differentiation trajectory: a symmetric kNN graph over the
3953
+ enabled samples' shared fluor channels (concatenated, so a day-series becomes
3954
+ one continuous trajectory), with pseudotime = geodesic distance from a root
3955
+ cell chosen at the extreme of a marker (e.g. CD902-high progenitors as t=0).
3956
+ It writes a ``pseudotime`` column to every sample (selectable as a plot
3957
+ colour) and draws each marker's mean expression along pseudotime — the
3958
+ CD902-down / CD901b-up maturation curve — with CSV / Prism XY / figure export.
3959
+ Backend (`openflo.trajectory`: `compute_pseudotime`, `robust_root`,
3960
+ `pseudotime_trends`) is pure (numpy/scipy/sklearn), subsamples large data for
3961
+ the graph and propagates by nearest neighbour, and is exported.
3962
+ - **Population frequencies & group comparison (GUI + backend) with GraphPad
3963
+ Prism export.** A new **Frequencies…** window collects each sample's
3964
+ per-population frequency, groups samples by a factor (trial/day, comp-vs-
3965
+ samples, or a name token like `Stim`/`Ctrl`), and for a chosen population +
3966
+ metric (%Parent / %Total / Count) draws a box+strip comparison with
3967
+ significance annotations plus an all-population overview. Statistics pick the
3968
+ right test automatically — Mann-Whitney U / Welch t for two groups,
3969
+ Kruskal-Wallis / one-way ANOVA + BH-adjusted pairwise post-hoc for more
3970
+ (`openflo.stats.compare_groups`). Exports: tidy CSV, **Prism Column** and
3971
+ **Prism Grouped** tables (columns = groups, rows = replicates; ragged groups
3972
+ padded — paste straight into GraphPad Prism), a stats summary, and the figure
3973
+ (White/Transparent/Translucent). Backend (`compare_groups`, `to_prism_column`,
3974
+ `to_prism_grouped`, `p_to_stars`) is pure and exported.
3975
+ - **Spectral unmixing QC + CLI batch-unmix.** New diagnostics for how
3976
+ trustworthy an unmix is: a spectral **similarity matrix** (cosine between
3977
+ reference spectra — flags fluorophore pairs too collinear to resolve), the
3978
+ **condition number** of the spectra matrix, and the **Spillover Spread
3979
+ Matrix** (SSM, Nguyen 2013 / Cytek — the spreading error each single-stain
3980
+ injects into every other fluor). After an Unmix the GUI opens a **Spectral
3981
+ QC** window with similarity + SSM heatmaps, the flagged similar / high-spread
3982
+ pairs, and Markdown / PNG export; the condition number and similar-pair
3983
+ count are recorded in the audit trail. New CLI mode `--unmix` builds
3984
+ reference spectra from `--unmix-controls` (a fluor→FCS JSON map, optional
3985
+ `unstained`), unmixes `--unmix-input` FCS into per-fluor CSVs, and writes
3986
+ `spectral_qc.{md,json}` + `reference_spectra.png` — so unmixing is no longer
3987
+ GUI-only. Backend: `spectral_similarity_matrix`, `spectral_condition_number`,
3988
+ `spillover_spread_matrix`, `unmixing_qc` (pure numpy, exported).
3989
+ - **Provenance / audit trail (GUI + backend).** A new append-only
3990
+ `AuditLog` records the meaningful operations of an analysis session in
3991
+ order — sample load (with path, event count, compensation source),
3992
+ transforms, cleaning, gate add/remove, auto-gate proposals (with their
3993
+ quality scores), clustering, batch normalization (with before/after QC
3994
+ distance), spectral unmixing, figure export and session reload. A
3995
+ **History…** button opens a live viewer that exports the trail to
3996
+ **Markdown** (a methods-section-ready table with an OpenFlo-version
3997
+ header), **CSV**, or **JSON**. The trail is embedded in the saved session
3998
+ and restored on load, so the record of *how* a result was produced travels
3999
+ with it. Pure/stdlib backend (`openflo.audit`), fully unit-tested.
4000
+ - **Trustworthy automated gating (GUI + backend).** The **Auto-gate** button
4001
+ (previously disabled — its single-contour heuristic mis-placed gates) now
4002
+ opens a dialog offering three well-posed, reviewable methods, each reported
4003
+ with a quality score in the status bar:
4004
+ - *Singlet gate* — a robust FSC-A/FSC-H ratio band (median ± k·MAD) emitted
4005
+ as a polygon; reports the fraction kept and ratio CV (`auto_singlet_gate`).
4006
+ - *Find populations (GMM ellipses)* — fits a Gaussian mixture on the current
4007
+ X/Y plot, picks the component count by BIC, and emits one **ellipsoid gate**
4008
+ per population at a chi-square coverage radius, each tagged with its weight
4009
+ and a separation score so overlapping (untrustworthy) splits are flagged
4010
+ (`gmm_ellipse_gates`).
4011
+ - *1-D threshold* — the existing valley/Otsu split.
4012
+ Every proposal is added as an ordinary undoable gate to accept / tweak /
4013
+ delete — review, not auto-apply. `describe_gate` now names polygon/rect gates
4014
+ and renders ellipsoid gates (previously shown as `? ellipsoid`).
4015
+ - **Multi-panel figure layout / export (GUI).** A **Figure…** button assembles
4016
+ the current plot into a publication-style small-multiples figure: one panel
4017
+ per sample (current channels), one panel per channel pair (samples overlaid),
4018
+ a samples × pairs grid, or a single panel. Channel pairs accept marker labels
4019
+ or channel names (e.g. `CD902/CD901b, CD901b/CD903`). Each panel reuses the live
4020
+ rendering pipeline (mode, density/colour, axis scales, gates) via an
4021
+ axes-swap (`_render_into`), so panels match the on-screen plot exactly. A
4022
+ preview window saves to PNG / PDF / SVG / TIFF at 300 dpi, with a
4023
+ **background** option — White (default), Transparent, or Translucent
4024
+ (50%) — for placing publication figures on a coloured page / poster.
4025
+ - **Spectral unmixing workflow (GUI).** An **Unmix** button designates loaded
4026
+ single-stain controls (→ fluorophore) + an unstained control, builds the
4027
+ reference spectra (with an autofluorescence endmember) and unmixes every
4028
+ other loaded sample into per-fluor `U:` abundance channels (OLS, optional
4029
+ non-negative) that become plottable/gateable, plus a spectrum-signature
4030
+ plot. Wraps the `spectral.py` backend.
4031
+ - **Batch correction (CytoNorm).** The flow-cytometry standard for removing
4032
+ technical batch/acquisition variation: FlowSOM-metacluster the pooled data,
4033
+ then per metacluster + channel quantile-normalize each batch onto a shared
4034
+ goal distribution. One engine, two modes — `goal` (CytoNorm 2.0, control-
4035
+ free, default) and `controls` (classic, fit on per-batch controls; CLI-only).
4036
+ The fitted model serializes and applies to new samples; a QC report gives
4037
+ per-channel Wasserstein before/after. GUI: a **Batch-norm** button (2.0,
4038
+ groups by trial/day). CLI: `--batch-correct` with `--cytonorm-mode` /
4039
+ `--cytonorm-control` / `--cytonorm-metaclusters`.
4040
+ - **Backgating.** Right-click a gate/population → *Backgate (show on plot)*
4041
+ projects its events, coloured, on top of the current plot — so you can see
4042
+ where a downstream population/cluster sits on any axes. Multi-select gives
4043
+ several colours + a legend; *Clear backgating* removes them.
4044
+ - **Auto-clean drop-count readout.** Each auto-clean gate row in the tree now
4045
+ shows how many events the recipe removes — `autocleaned sample — drops N
4046
+ (X%)` — with a per-method breakdown under it (each method's standalone
4047
+ contribution, shown even when toggled off so you can preview it). Computed on
4048
+ the full sample and cached by data identity + recipe signature.
4049
+ - **Staining-panel `.xlsx` → channel labels (CLI).** `--panel <file>` (or
4050
+ `--panel auto`, which searches the trial folders and a few ancestor levels)
4051
+ reads a CD↔fluorophore sheet and maps each fluorophore to its detector
4052
+ channel, merged with `--labels`. New `read_staining_panel` / `find_panel_xlsx`.
4053
+ - **Per-group marker-pair scatters (CLI).** Every group now emits, for each
4054
+ pair in `--pairs` (default CD902/CD901b, CD901b/CD903, CD902/CD903), an *overlay*
4055
+ (all samples on one axes, coloured by sample) and a *grid* (one density panel
4056
+ per sample, shared limits). New `save_group_pair_scatters`.
4057
+ - **Adjustable plot point cap.** A **Max points** control (presets + free
4058
+ entry, `250k`/`All` accepted) replaces the fixed 60 k scatter cap; drives
4059
+ scatter / pseudocolor / contour, updates the tree's shown/total counts, and
4060
+ persists in the session.
4061
+ - **“Show cleaned-out events” overlay.** A plot-control toggle that draws the
4062
+ events the auto-clean recipe removes *in red, on top* of whatever's plotted —
4063
+ computed on the full sample and **bypassing the display cap**, so a small
4064
+ error rate stays visible against the full population instead of being
4065
+ subsampled away. Scatter modes overlay red dots (with a count); histogram
4066
+ mode overlays the removed events' channel distribution scaled to be visible.
4067
+ Reflects the current recipe and persists in the session.
4068
+
4069
+ ### Changed
4070
+ - **Config-driven batch runner.** `scripts/run_analyses.py` reads a JSON
4071
+ config (default the git-ignored `private/analysis_config.json`; see
4072
+ `scripts/analysis_config.example.json`) describing analyses via reusable
4073
+ `group_by` strategies — no data paths baked into tracked code. `--dry-run`
4074
+ resolves + verifies groups without clustering. Keeps the tracked tree
4075
+ generic so real experiment paths stay in `private/` (git-ignored).
4076
+ - **Smoother density rendering.** Pseudocolor samples each event's colour by
4077
+ **cubic** interpolation of the smoothed density field (C2-continuous, so no
4078
+ per-cell colour blocks *or* residual bin-grid box facets), with an adaptive
4079
+ smoothing floor, and colours via `PowerNorm` so large samples no longer wash
4080
+ out to one flat hue. Histograms render as kernel-smoothed filled curves
4081
+ instead of chunky step bars. Contour density is zero-padded so every level
4082
+ closes.
4083
+ - **GUI caps BLAS threads at startup** (mirroring the CLI) so OpenBLAS can't
4084
+ exhaust memory and abort the console-less launch under pressure.
4085
+ - **Auto-clean gate.** A new **Auto-clean** button adds an *“autocleaned
4086
+ sample”* recipe gate (a collapsible group of toggleable cleaning methods —
4087
+ debris, doublets, margin/saturation, flow-rate bubbles/clogs, signal drift).
4088
+ It stores the *calculation*, not coordinates: its mask is the AND of the
4089
+ enabled methods, recomputed from each sample's own data, so copying it to
4090
+ other samples re-runs the cleaning per sample rather than reusing one
4091
+ sample's geometry. Build downstream gates under it to gate on cleaned events.
4092
+ Not FlowJo-representable — WSP export drops it and re-roots any children.
4093
+ - **Folder drag-and-drop import.** Dropping a folder recurses into its
4094
+ `.fcs`/`.wsp` files; dropping a parent of several trial folders imports each
4095
+ independently. A bounded background load queue (fixed worker pool) replaces
4096
+ one-thread-per-file so large folder drops can't exhaust memory, with a
4097
+ determinate progress bar showing *N/M loaded*.
4098
+ - **Histogram Y-axis selector** — Fraction (default) / Count / % of Max. Raw
4099
+ Count honours the auto-downsample toggle (and bypasses the scatter-only 60k
4100
+ cap) so counts are truthful.
4101
+ - **Event counts in the Samples & Gates tree.** Each sample row shows its
4102
+ event count, displayed as `shown/total` when auto-downsampling scales it to
4103
+ the smallest sample (and updating when the toggle changes).
4104
+ - **Auto-clean parameter dialog.** Double-click an auto-clean gate (or method
4105
+ row), or right-click → *Edit auto-clean parameters…*, to tune each method's
4106
+ enabled flag and parameters (bin counts, MAD thresholds, doublet tolerance,
4107
+ an optional manual debris FSC cutoff).
4108
+ - **Auto-clean masks are cached** per (sample data, recipe) and reused across
4109
+ replots — recomputed only when the data or recipe changes — so gating on
4110
+ cleaned events stays responsive on large samples. The mask is computed on the
4111
+ full sample data (a per-acquisition property), so filter and highlight views
4112
+ flag the same events even when a plotted axis is sparse (e.g. an embedding).
4113
+
4114
+ ### Changed
4115
+ - **Imported day groups split into Comps + Samples subgroups.** When a day
4116
+ group contains compensation controls (names matching comp / control /
4117
+ (un)stained), the tree shows a *Samples* sub-header (expanded) and a *Comps*
4118
+ sub-header (collapsed by default); each subgroup's ✓ toggles its members'
4119
+ display. Days without comps list samples directly as before.
4120
+ - **Imported gates load disabled.** Gates brought in with no explicit enabled
4121
+ flag (e.g. from a `.wsp`) start unchecked, so a freshly-loaded sample isn't a
4122
+ wall of active toggles. A restored session's gates keep their saved state.
4123
+ - **Drag samples between groups.** A sample row can now be dragged to another
4124
+ day, or between the Comps and Samples subgroups, to fix a mis-import or for
4125
+ convenience (a manual Comps/Samples choice overrides the name-based guess).
4126
+ Multi-selection is honoured, and the regrouping persists in saved sessions.
4127
+ - **“Clear all” keeps auto-clean gates by default**, with a checkbox in the
4128
+ confirm dialog to also clear them — so a bulk gate wipe doesn't discard the
4129
+ cleaning foundation.
4130
+ - **Folder grouping is now by collection “Day N”.** `derive_trial_name` scans
4131
+ ancestor folders for a `Day N` token (at whatever depth it sits) and groups
4132
+ by it, falling back to the grandparent folder when absent; day groups sort
4133
+ numerically. Samples whose filenames repeat across days are disambiguated
4134
+ (e.g. `… [Day 9]`) so identical names no longer silently overwrite one
4135
+ another.
4136
+ - **“Clear all” now clears all gates but keeps the samples** (undoable),
4137
+ reversing the 1.0.0 behaviour where it removed every sample. **Clear** now
4138
+ acts on the selection: a gate (cascade), a sample's gates, or a whole
4139
+ trial's gates — never removing samples (use **Remove** for that).
4140
+
4141
+ ## [1.0.0] — 2026-05-29
4142
+
4143
+ First public release.
4144
+
4145
+ ### Fixed
4146
+ - **UMAP/TriMap runs no longer flash a console window** on Windows — the
4147
+ per-unit worker subprocess (and the Cancel `taskkill`) launch with
4148
+ `CREATE_NO_WINDOW`.
4149
+ - **"Clear all" now actually clears the panel.** It previously only emptied
4150
+ the active sample's gates; it now removes every loaded sample and all gates
4151
+ (confirmed, since sample removal isn't undoable).
4152
+ - The **Auto-gate** button is greyed out for now — its density heuristic
4153
+ mis-placed gates often enough to be untrustworthy.
4154
+
4155
+ ### Changed
4156
+ - **Repo layout consolidated.** Loose root scripts moved into folders —
4157
+ `smoke_test.py` → `scripts/`, `HANDOFF.md` → `docs/`; the top level now
4158
+ keeps only standard docs, config, and launchers.
4159
+ - **Statistics is strictly population-based.** The window accepts only gate /
4160
+ population rows — dragged from the Samples & Gates panel or from a *gated*
4161
+ Pipeline Workspace item — never whole samples or trials. The two Import
4162
+ buttons (**Import S&G gates** / **Import workspace**) REPLACE the current
4163
+ set; dragging a gate APPENDS. A **Source** column tags each row
4164
+ `editor` / `workspace` / `editor+workspace`.
4165
+ - **Editor bottom-left buttons unified.** Clear / Clear all / Copy / Pops are
4166
+ now equal-width and compact (↶/↷ stay as small icon buttons), making room
4167
+ for the new log pane.
4168
+ - Selecting a **trial** row and pressing **Delete** (or **Remove**, or
4169
+ right-click → *Remove trial*) now clears that trial's samples and gates
4170
+ (confirmed).
4171
+ - **Pipeline default grouping is now by day, not a fixed two-group split.**
4172
+ With no `--groups`/`--samples`, OpenFlo discovers every folder that
4173
+ directly holds FCS files — point it at a single PARENT and each
4174
+ sub-folder becomes its own day/group, sampled independently and
4175
+ compared across days in one analysis. Folder names are tidied to
4176
+ `Day N` when a day token is present; duplicate day names are
4177
+ disambiguated by parent. Explicit `--groups` and the legacy
4178
+ `--samples` split still work; `DEFAULT_GROUPS` remains the final
4179
+ fallback.
4180
+ - **Per-sample FMO assignment.** A group's `samples` entry may be a
4181
+ plain string (inherits the group's `fmo_set`) or
4182
+ `{'name', 'fmo_set'}` to point one sample at a different FMO control
4183
+ set. FMO thresholds + both run modes resolve per sample. Compensation
4184
+ and antibody labels were already per-sample-automatic (each FCS's
4185
+ `$SPILL` / `$PnS`).
4186
+ - **The gate editor is now the entire GUI.** `openflo-gui` opens straight
4187
+ into the editor (it owns a hidden Tk root); closing it exits. Pipelines
4188
+ run from the editor's docked **Pipeline Workspace** — drag samples /
4189
+ gated populations in and Run. The separate pipeline-config window was
4190
+ removed (see *Removed*).
4191
+ - **Pipeline Workspace runs Phenograph + UMAP + TriMap per RUN UNIT, each
4192
+ in its own subprocess.** A *unit* is a **group's samples co-embedded into
4193
+ one UMAP** (events tagged by source sample); a **Concatenate** toggle
4194
+ merges all groups into a single UMAP so groups compare in one embedding
4195
+ (FlowJo-style); loose items run on their own. Each run writes a
4196
+ cluster-frequency CSV, a **cluster × group/sample composition CSV**, and
4197
+ embedding PNGs coloured by cluster *and* by source. Embeddings use the
4198
+ proper per-marker channels (height/width detector duplicates dropped) on
4199
+ an up-front subsample. A native crash / hang / OOM is isolated to that
4200
+ child (the GUI survives); **Cancel** terminates the job's whole process
4201
+ tree; a crashed/OOM unit is requeued once at a lower event cap, then
4202
+ skipped. The editor's Undo button also reverts workspace edits;
4203
+ workspaces save/load to JSON; a Results viewer shows the outputs.
4204
+
4205
+ ### Removed
4206
+ - **The legacy pipeline run-plan / staging window (the `App` class) and
4207
+ its in-process + subprocess run engine.** It was discontinued — most of
4208
+ its features were unreliable (crashes / restart loops). ~3,900 lines
4209
+ removed; its role is taken by the Pipeline Workspace. Also dropped the
4210
+ now-unused Windows Job-object / memory-watchdog / GPU-probe
4211
+ infrastructure and `tests/test_run_plan.py`.
4212
+
4213
+ ### Added
4214
+ - **Collapsible in-app log pane.** A "Show log" toggle at the bottom of the
4215
+ editor's left column reveals a small terminal that mirrors stdout/stderr
4216
+ (diagnostics, tracebacks) without needing a console; "Clear log" empties it.
4217
+ - **Pipeline Workspace item drag.** Drag an item between groups — or onto
4218
+ empty space to pop it back to the top level — to fix a mis-drop. Dragging a
4219
+ *gated* item onto an open Statistics window adds its population.
4220
+ - **In-editor clustering + the full cluster→name→use loop.** A "Cluster…"
4221
+ button runs Phenograph or FlowSOM (+ optional UMAP) on loaded samples in
4222
+ a worker thread, then auto-imports the result as populations and switches
4223
+ the plot to the UMAP coloured by the label. Population import is now
4224
+ generic — the "Populations…" menu detects any present label column
4225
+ (`cluster`, `flowsom_meta`, cell-cycle phases) and offers import + rename
4226
+ for each. Clustered/UMAP'd data can also be brought in from outside via
4227
+ **"Load CSV…"** (`FlowSample.from_dataframe` ingests a pipeline
4228
+ `*_processed.csv`, preserving cluster/UMAP/flowsom columns); derived
4229
+ columns are auto-excluded from marker lists.
4230
+ - **Spectral unmixing.** New `openflo.spectral`: `build_reference_spectra`
4231
+ turns single-stain (+ unstained autofluorescence) controls into a
4232
+ reference spectra matrix; `unmix` solves per-event fluorophore abundances
4233
+ by least squares (OLS, optional non-negativity); `apply_unmixing` adds
4234
+ one abundance column per fluor to a sample. For full-spectrum cytometers
4235
+ (Cytek Aurora, BD S8) where compensation alone doesn't apply.
4236
+ - **Differential abundance / expression.** New `openflo.diffexp`:
4237
+ `differential_test` (Mann-Whitney U + log2 fold-change + Benjamini-
4238
+ Hochberg FDR) over per-sample feature values, with `cluster_abundance`
4239
+ and `marker_expression` builders that turn two groups of samples into the
4240
+ per-sample feature dicts. The diffcyt/OMIQ-style comparison OpenFlo
4241
+ lacked.
4242
+ - **FlowSOM clustering + metaclustering.** `FlowSample.run_flowsom()` trains
4243
+ a self-organizing map over the marker space, assigns each event to a node,
4244
+ and agglomerates nodes into metaclusters — writing `flowsom` (node) and
4245
+ `flowsom_meta` (metacluster) columns. Compact, dependency-free
4246
+ (numpy + sklearn), fast on large files.
4247
+ - **More transforms + per-channel transform editor.** `transform_values` /
4248
+ `inverse_transform_values` add **arcsinh** and **hyperlog** (and a linear
4249
+ pass-through) alongside logicle/log, with FlowJo's t/m/w/a knobs (arcsinh
4250
+ uses an intuitive `cofactor`). A "Transforms…" editor in the GUI re-maps
4251
+ each channel's transform across all loaded samples by inverting the
4252
+ current one and applying the new — no re-compensation needed.
4253
+ - **Boolean gates (AND / OR / NOT).** New `boolean` gate kind combining
4254
+ other gates' cumulative masks (cycle-guarded). Build one from the gate
4255
+ tree's right-click menu ("Create boolean gate…"); it toggles, highlights,
4256
+ filters, and feeds the stats table like any population. Dropped from
4257
+ `.wsp` export with a lossy-export note.
4258
+ - **Automated density-based gating (auto-gate).** `auto_threshold` (valley
4259
+ between the two density modes, else Otsu) and `auto_polygon_gate` (a
4260
+ contour around the dominant 2-D density mode). An "Auto-gate" button
4261
+ proposes a threshold (histogram) or polygon (2-D) for the active sample
4262
+ to accept or tweak.
4263
+ - **Undo / redo in the gate editor.** Ctrl+Z / Ctrl+Y (and ↶/↷ buttons)
4264
+ over a snapshot history of the gate state. Every structural change —
4265
+ add, delete, drag, reparent, paste, cluster/cell-cycle import,
4266
+ annotate — is one undoable step (mutations in a single gesture coalesce);
4267
+ bulk session/template loads don't pollute the history.
4268
+ - **Cell-cycle recognition (DNA content).** `FlowSample.cell_cycle()`
4269
+ auto-detects a DNA-stain channel (PI / DAPI / FxCycle / 7-AAD / Hoechst /
4270
+ DRAQ5 / …; `find_dna_channel`), optionally pre-gates singlets on the
4271
+ DNA-A vs `-W`/`-H` ratio (doublet exclusion), then models the histogram
4272
+ (`analyze_dna`): locates the G1 peak and the G2/M peak at ~2× DNA,
4273
+ estimates each peak's robust spread, and assigns every event a phase
4274
+ (G1 / S / G2M / sub-G1 / >G2M) → %G1/%S/%G2M. Writes a categorical
4275
+ `cell_cycle` column. In the editor, a "Cell cycle…" button runs it on
4276
+ the active (or all) sample(s), surfaces each phase as a selectable
4277
+ population (new `category` gate kind), and shows a DNA histogram +
4278
+ phase-percentage window.
4279
+ - **Acquisition QC now detects clogs, bubbles, and saturation.**
4280
+ `AcquisitionQC` gained two detectors beyond the existing signal-drift
4281
+ one: **flow-rate anomalies** (time bins whose event count is a MAD
4282
+ outlier, plus empty interior bins — clog collapses and bubble gaps/
4283
+ bursts) and **margin/saturation events** (per-event removal of pile-ups
4284
+ at a channel's ceiling). All three combine into one clean-event index;
4285
+ `qc.report` breaks down removals by category. A clean acquisition trips
4286
+ none of them.
4287
+ - **TriMap and PaCMAP dimensionality reduction.** `FlowSample.run_trimap()`
4288
+ and `run_pacmap()` mirror `run_umap` (shared `_embedding_input` /
4289
+ `_store_embedding` helpers), writing `TRIMAP1/2` and `PACMAP1/2`. Both
4290
+ are optional (`pip install openflo[embed]`) and degrade gracefully when
4291
+ not installed. They preserve global structure better than UMAP on some
4292
+ panels.
4293
+ - **Voltage titration / Stain Index tool.** New `openflo.voltage` module +
4294
+ `openflo-voltage` CLI: point it at a titration series (one FCS per PMT
4295
+ voltage) and a channel, and it reads `$PnV` per detector, auto-splits the
4296
+ negative/positive populations (2-component GMM), computes per-voltage
4297
+ Stain Index = (med⁺−med⁻)/(2·rSD⁻) and the robust CV of the negative,
4298
+ and recommends the lowest voltage on the SI plateau. Generalized — any
4299
+ channel, any file set; pure metric layer is independently importable as
4300
+ `VoltageTitration`.
4301
+ - **Plot axes resolve by antibody label per sample.** When overlaying
4302
+ samples whose marker sits on different fluorophores, picking an axis
4303
+ (a detector from the global panel) now resolves to *each* sample's own
4304
+ detector by antibody label (`_axis_alias_for_sample`), so the samples
4305
+ overlay on a common label axis instead of being dropped. The chosen
4306
+ name is aliased onto the sample's own column (the original detector
4307
+ column stays, so per-sample gate masks are unaffected); a sample that
4308
+ lacks the marker entirely is simply skipped. Completes the label-first
4309
+ follow-up to gate-by-label retargeting.
4310
+ - **Clusters as selectable, annotatable populations in the editor.** A
4311
+ "Clusters…" button imports each clustering label (the pipeline's
4312
+ `cluster` column) as a root population — a new `cluster` gate kind
4313
+ whose mask is `cluster == id` (`gate_to_mask`; a missing column selects
4314
+ nothing rather than no-op all-True). Imported populations toggle,
4315
+ highlight, filter, and feed the statistics table like any gate. An
4316
+ "Annotate clusters…" dialog names them with phenotypes, persisted in
4317
+ the session's `cluster_labels` slot and shown as the population name.
4318
+ Cluster populations have no FlowJo geometry, so the `.wsp` export
4319
+ drops them with a clear lossy-export warning.
4320
+ - **Gate templates retarget by antibody label.** Saving a template now
4321
+ stamps each gate's channel with its antibody label; applying that
4322
+ template to a sample where the marker sits on a *different* detector
4323
+ retargets the gate to that sample's detector (`relabel_gate_for_sample`).
4324
+ So a CD901b gate applies wherever CD901b lives in each sample, across
4325
+ panels — compensation is unaffected (only which column the gate
4326
+ reads changes).
4327
+ - **Per-sample FMO override in the config GUI.** A group's samples
4328
+ field accepts `name:FMOset` to point one sample at a different FMO
4329
+ control set (e.g. `m1, m2:Late, m3`); `_get_groups` emits the
4330
+ per-sample dict form the pipeline resolves. A hint documents the
4331
+ syntax.
4332
+ - **Cross-sample label-first tying + common-fluor warning.** The same
4333
+ antibody can sit on a different fluorophore across samples/days, so
4334
+ cross-sample analysis now aligns by antibody **label**, not detector
4335
+ (compensation stays keyed on detectors — each sample compensates its
4336
+ own `$SPILL`). New `openflo.pipeline` utilities: `align_fluor_labels`
4337
+ (common labels + per-sample label→detector + missing map),
4338
+ `common_fluor_warning`, and `concatenate_by_label` (merge samples on
4339
+ the common label set, renaming each sample's fluors to labels). The
4340
+ statistics table now names per-channel columns by **each sample's own**
4341
+ label, so a marker on different fluors merges into one column
4342
+ (`Median CD901b`) across samples. The editor flags a non-common fluor
4343
+ panel on sample load and in the Statistics window; non-common labels
4344
+ are simply blank where absent.
4345
+ - **Population statistics table (FlowJo-style).** A "Statistics…" window
4346
+ in the editor tabulates, per sample × population (gate node, evaluated
4347
+ as the cumulative gate chain): Count, %Parent, %Total, and per-channel
4348
+ Median / Mean / CV. Columns are modular (checkbox toggles); the table
4349
+ exports to analysis-ready CSV. Populations show a FlowJo-style path
4350
+ (`Cells/Singlets/CD901b+`). Computed on full sample data, not the plot
4351
+ downsample.
4352
+ - **Full FlowJo gate parity — ellipsoid + quadrant.** `WspReader`
4353
+ parses `EllipsoidGate` (mean + covariance + distanceSquare) and
4354
+ `QuadrantGate` (two dividers → 4 linked rects); `WspWriter` emits
4355
+ both (collapsing a `quad_set` rect group back into one QuadrantGate);
4356
+ `gate_to_mask` evaluates ellipsoids via squared Mahalanobis distance.
4357
+ Round-trip is self-consistent (our writer ↔ reader); FlowJo v10's
4358
+ exact serialization still needs validation against a real file.
4359
+ - **Editor: ellipsoid rendering + interactive Ellipse tool.** Ellipses
4360
+ render (rotated too, via covariance eigendecomposition); a new
4361
+ Ellipse tool draws them, and the Edit tool moves / resizes (drag rim)
4362
+ / rotates (drag grip) them.
4363
+ - **`.flowsession` save/load.** Captures the full editor state —
4364
+ samples (by path + colour + plot-enabled), per-sample gates at full
4365
+ fidelity (incl. ellipsoid / quadrant / colour / enabled), per-channel
4366
+ scale + range, plot mode, channel labels, downsample toggles, and a
4367
+ reserved `cluster_labels` slot. Autosaves to
4368
+ `~/.openflo/last_session.flowsession` on editor close and offers to
4369
+ resume it on next open. Save/Load Session buttons in the editor.
4370
+ - **Batch template application.** "Load Template…" now pops a dialog to
4371
+ choose which loaded samples to apply to (multiselect + select all /
4372
+ none) and whether to **overwrite** each target's gates or **add to**
4373
+ them. Previously a template loaded into the active sample only. Gates
4374
+ referencing channels a target sample lacks are reported in a
4375
+ post-apply warning (they install but sit inert).
4376
+ - **Lossy-export warning.** Exporting to `.wsp` now checks for
4377
+ OpenFlo-only state the FlowJo schema can't hold (custom per-channel
4378
+ axis scales / ranges, disabled gates, cluster labels) and warns
4379
+ before writing, offering to save a full `.flowsession` instead.
4380
+ Gates + compensation always survive, so a plain gating export
4381
+ doesn't nag.
4382
+ - **End-to-end CLI tests** (`tests/test_cli_e2e.py`). Two tiers:
4383
+ fast `--help`-based wiring checks (always run — catch console-
4384
+ script breakage + flag-parsing regressions); and a full-pipeline
4385
+ subprocess run against the synthetic FCS, opt-in via
4386
+ `OPENFLO_RUN_SLOW_TESTS=1` (it runs Phenograph + UMAP, ~35 s warm
4387
+ but timing-sensitive under load, so it's gated like the real-data
4388
+ fixtures rather than making the default suite flaky).
4389
+ - **WSP per-sample extract tests** (`tests/test_wsp_writer.py`) —
4390
+ exercise the `extract_gates(sample_node=...)` kwarg added
4391
+ during the gate-editor WSP-ingest work. Multi-sample synthetic
4392
+ workspace, per-sample subsetting, parent_id chain preservation,
4393
+ default-walk regression.
4394
+ - **OSS infrastructure** — `.github/ISSUE_TEMPLATE/` (bug + feature
4395
+ + config routing questions to Discussions), `PULL_REQUEST_TEMPLATE`
4396
+ with a "Scientific impact" section, `.pre-commit-config.yaml`
4397
+ (trailing-whitespace, EOF, large-files, ruff check + format),
4398
+ `environment.yml` (conda mirror with optional RAPIDS).
4399
+ - **`docs/algorithms.md`** — ~250 lines covering compensation
4400
+ sources + optimizer heuristics, logicle T/M/W/A defaults with
4401
+ FlowJo parity notes, FMO threshold rationale, Phenograph k
4402
+ rule-of-thumb table, subsample + KD-tree-assign trick, GPU
4403
+ determinism caveats, UMAP defaults, what the pipeline is NOT
4404
+ good at. Cites Parks 2006, Levine 2015, McInnes 2018, Roederer
4405
+ 2011. README links via a new `## Algorithms` section.
4406
+ - **README "Common workflows" section** — three concrete examples
4407
+ (single-sample GUI exploration, multi-trial batch run with
4408
+ `--groups` + `--fmo-sets` + `--export-wsp`, `openflo-compare`
4409
+ against a FlowJo workspace).
4410
+ - **Vulture dead-code config** in `pyproject.toml` `[tool.vulture]`
4411
+ with documented false-positive exclusions for PEP-562 hooks,
4412
+ public API surface, and ctypes Structure fields.
4413
+
4414
+ ### Changed
4415
+ - **`pipeline.py` lazy-imports `matplotlib.pyplot`.** Moved from
4416
+ module-top to local imports inside the 5 plot methods. Saves
4417
+ ~300 ms on `import openflo.pipeline` (1050 ms → 750 ms) —
4418
+ matters for the gate editor / compare tool / any WSP-only
4419
+ caller. PEP-562 hook now exposes `pipeline.plt` for external
4420
+ callers that still want the bare attribute.
4421
+ - **Gate editor write paths surface failures visibly.**
4422
+ `_save_template`, `_export_flowjo_wsp`, and `_apply_save_gates`
4423
+ now `messagebox.showerror` on failure in addition to the
4424
+ status-bar message. Silent data loss after a Save dialog is
4425
+ worse than the alert pop-up.
4426
+ - **Removed the `_LazyFlowio` proxy** from `gui.py`. Replaced
4427
+ with a function-local `import flowio` at the single call site
4428
+ in `_inspect_channels_for_labels`. Same lazy effect; 16 fewer
4429
+ lines; no `# type: ignore[assignment]` workaround.
4430
+
4431
+ ## [0.2.0] — 2026-05-27
4432
+
4433
+ ### Added
4434
+ - **Gate editor: Edit tool** with modifier-key gestures — left-drag to
4435
+ move a vertex/line, shift+drag to translate the whole gate,
4436
+ right-click on a polygon vertex to delete (refuses below 3 verts),
4437
+ right-click on an edge to insert, alt+left-click anywhere to drop
4438
+ a vertex into the polygon under the cursor. Per-tool gesture hint
4439
+ shown below the tool selector.
4440
+ - **Per-channel axis scale + range.** ⚙ buttons next to the X/Y combos
4441
+ open a dialog: Linear / Symlog / Log scale, plus optional custom
4442
+ (min, max) range. State is keyed by channel name so swapping the
4443
+ X combo to a different channel picks up that channel's saved
4444
+ preference. Symlog `linthresh` is data-driven (5th percentile of
4445
+ |nonzero|, floor 1e-6).
4446
+ - **Auto-downsample toggles.** "Auto-downsample display to smallest
4447
+ sample" (default ON) caps every plotted sample at the smallest
4448
+ loaded sample's size for honest overlay comparisons; underlying
4449
+ data is untouched. "…and propagate to data" (default OFF) actually
4450
+ trims `FlowSample.data` so clustering / stats see the trimmed set.
4451
+ Seeded so the same subsample renders across replots.
4452
+ - **WSP ingest in the gate editor's Add-FCS button.** Picking a `.wsp`
4453
+ walks each `<Sample>`, resolves its `<DataSet uri="...">` to a
4454
+ local FCS path (tries as-is, then the WSP's own directory, then
4455
+ the editor's `fcs_dir`), queues the FCS for load, and stages the
4456
+ sample's gate subtree to attach as the FCS finishes parsing.
4457
+ - `WspReader.extract_gates(*, sample_node=...)` — opt-in per-sample
4458
+ walk that reuses the existing parsers. Default behaviour
4459
+ unchanged.
4460
+ - **Log-spaced histogram bins** when a channel's axis scale is `log`
4461
+ (linear / symlog continue to use linear-spaced bins). New
4462
+ `_hist_bin_edges` helper clamps non-positive lower bounds to a
4463
+ small positive floor and falls back to linear when the clamped
4464
+ range degenerates.
4465
+ - Comprehensive unit tests for the new gate-editor helpers
4466
+ (`tests/test_gate_editor_helpers.py`, 47 tests) — covers
4467
+ `_gid_from_hit`, polygon vertex add/delete/find, downsample floor,
4468
+ axis scale apply path, and log-spaced bin edges.
4469
+
4470
+ ### Changed
4471
+ - Compensation matrix actually round-trips through the workspace
4472
+ export. Both `gui._export_flowjo_wsp` and
4473
+ `cli._export_pipeline_workspace` now call `WspWriter.set_compensation`;
4474
+ `FlowSample._apply_comp` persists the matrix on
4475
+ `self.comp_matrix` / `self.comp_channels` so callers can read it
4476
+ back. Two new regression tests in `tests/test_wsp_writer.py`.
4477
+ - `OptimizeCompensationDialog._autofill_from_dir` — the auto-detect
4478
+ for single-stain control files now uses an ordered candidate
4479
+ tokenizer (joined form first, then each dash-separated part
4480
+ longest-first) instead of the naïve "first dash-separated token"
4481
+ heuristic. `FL3-A` now produces tokens `['fl3', 'cy7']`
4482
+ instead of just `['pe']`. Ambiguous and unmatched channels surface
4483
+ in the status bar.
4484
+
4485
+ ### Fixed
4486
+ - **Sample-name collision across groups.** Per-sample tasks were keyed
4487
+ by bare sample name in the dispatcher, so two groups (e.g. two day
4488
+ folders) containing identically-named FCS (`sample_1.fcs`)
4489
+ silently bucketed both results into one group and dropped the other.
4490
+ Tasks are now keyed by group+name. Surfaced + guarded by the by-day
4491
+ e2e test.
4492
+ - **Histogram blank rendering on wide-range fluor data.** Non-finite
4493
+ values (NaN / ±inf) silently made matplotlib's hist skip entries;
4494
+ auto-ranging across samples with vastly different scales (one
4495
+ logicle ~0–1, one raw 0–262144) collapsed the narrow-range sample
4496
+ into a single bin at zero. Now filters non-finite up-front and
4497
+ pins all samples to a shared bin grid built from the union of
4498
+ robust per-sample percentile ranges.
4499
+
4500
+ ## [0.1.0] — 2026-05-27
4501
+ Baseline version captured for the first OSS-ready release. See git log for
4502
+ the full pre-OSS feature set (compensation editor, WSP round-trip, GUI
4503
+ gate editor, comparison tool, GPU clustering, seeded reproducibility).