rang-core 0.1.0__tar.gz → 0.2.0__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 (407) hide show
  1. {rang_core-0.1.0 → rang_core-0.2.0}/.gitattributes +5 -0
  2. rang_core-0.2.0/CHANGELOG.md +610 -0
  3. {rang_core-0.1.0 → rang_core-0.2.0}/CITATION.cff +2 -2
  4. {rang_core-0.1.0 → rang_core-0.2.0}/Cargo.lock +2 -2
  5. {rang_core-0.1.0 → rang_core-0.2.0}/Cargo.toml +4 -4
  6. {rang_core-0.1.0 → rang_core-0.2.0}/PKG-INFO +45 -35
  7. {rang_core-0.1.0 → rang_core-0.2.0}/README.md +35 -32
  8. rang_core-0.2.0/THIRD-PARTY-NOTICES.txt +234 -0
  9. {rang_core-0.1.0 → rang_core-0.2.0}/core/Cargo.lock +4 -4
  10. {rang_core-0.1.0 → rang_core-0.2.0}/core/Cargo.toml +20 -2
  11. {rang_core-0.1.0 → rang_core-0.2.0}/core/py/Cargo.lock +5 -5
  12. {rang_core-0.1.0 → rang_core-0.2.0}/core/py/Cargo.toml +1 -1
  13. rang_core-0.2.0/core/py/src/cal_delay.rs +212 -0
  14. rang_core-0.2.0/core/py/src/cal_delay_offset.rs +255 -0
  15. rang_core-0.2.0/core/py/src/cal_fluxscale.rs +123 -0
  16. rang_core-0.2.0/core/py/src/cal_gain.rs +350 -0
  17. rang_core-0.2.0/core/py/src/cal_pol.rs +300 -0
  18. rang_core-0.2.0/core/py/src/cnnflag.rs +43 -0
  19. rang_core-0.2.0/core/py/src/f3flag.rs +87 -0
  20. rang_core-0.2.0/core/py/src/flag_extend.rs +61 -0
  21. rang_core-0.2.0/core/py/src/gemclean_cli.rs +23 -0
  22. rang_core-0.2.0/core/py/src/gemclean_predict.rs +89 -0
  23. {rang_core-0.1.0 → rang_core-0.2.0}/core/py/src/lib.rs +415 -58
  24. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/aipsio.rs +1 -0
  25. rang_core-0.2.0/core/src/bin/gemclean.rs +21 -0
  26. rang_core-0.2.0/core/src/bin/gram_solver_bench.rs +904 -0
  27. rang_core-0.2.0/core/src/bin/gridder_gram_bench.rs +580 -0
  28. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/blosc.rs +1 -0
  29. rang_core-0.2.0/core/src/cal/apply.rs +1604 -0
  30. rang_core-0.2.0/core/src/cal/bandpass.rs +163 -0
  31. rang_core-0.2.0/core/src/cal/crosshand.rs +454 -0
  32. rang_core-0.2.0/core/src/cal/delay.rs +1171 -0
  33. rang_core-0.2.0/core/src/cal/delay_offset.rs +1197 -0
  34. rang_core-0.2.0/core/src/cal/diag.rs +1154 -0
  35. rang_core-0.2.0/core/src/cal/ffi_apply.rs +250 -0
  36. rang_core-0.2.0/core/src/cal/fluxscale.rs +622 -0
  37. rang_core-0.2.0/core/src/cal/leakage.rs +626 -0
  38. rang_core-0.2.0/core/src/cal/mod.rs +25 -0
  39. rang_core-0.2.0/core/src/cal/parang.rs +114 -0
  40. rang_core-0.2.0/core/src/cal/reference.rs +368 -0
  41. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/caltable.rs +1 -0
  42. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/clean.rs +1 -0
  43. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/clean_certified.rs +1 -0
  44. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/clean_major.rs +1 -0
  45. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/closure.rs +1 -0
  46. rang_core-0.2.0/core/src/cnnflag.rs +331 -0
  47. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/correct.rs +1 -0
  48. rang_core-0.2.0/core/src/f3flag.rs +571 -0
  49. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ffi.rs +1 -0
  50. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ffi_closure.rs +1 -0
  51. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ffi_correct.rs +1 -0
  52. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ffi_flag.rs +43 -0
  53. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ffi_fluxstd.rs +1 -0
  54. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ffi_grid.rs +1 -0
  55. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ffi_ms.rs +1 -0
  56. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ffi_ms_write.rs +1 -0
  57. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ffi_weights.rs +1 -0
  58. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ffi_wgridder.rs +1 -0
  59. rang_core-0.2.0/core/src/flag_extend.rs +345 -0
  60. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/fluxstd.rs +1 -0
  61. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/fused.rs +5 -4
  62. rang_core-0.2.0/core/src/gemclean/cli.rs +3375 -0
  63. rang_core-0.2.0/core/src/gemclean/fits.rs +342 -0
  64. rang_core-0.2.0/core/src/gemclean/host.rs +327 -0
  65. rang_core-0.2.0/core/src/gemclean/input.rs +948 -0
  66. rang_core-0.2.0/core/src/gemclean/joined.rs +732 -0
  67. rang_core-0.2.0/core/src/gemclean/kernels.rs +415 -0
  68. rang_core-0.2.0/core/src/gemclean/mask.rs +554 -0
  69. rang_core-0.2.0/core/src/gemclean/measure.rs +727 -0
  70. rang_core-0.2.0/core/src/gemclean/memory.rs +123 -0
  71. rang_core-0.2.0/core/src/gemclean/multiscale.rs +1555 -0
  72. rang_core-0.2.0/core/src/gemclean/predict.rs +615 -0
  73. rang_core-0.2.0/core/src/gemclean/sim.rs +311 -0
  74. rang_core-0.2.0/core/src/gemclean/spectral.rs +362 -0
  75. rang_core-0.2.0/core/src/gemclean/stale_bound.rs +671 -0
  76. rang_core-0.2.0/core/src/gemclean/stokes.rs +464 -0
  77. rang_core-0.2.0/core/src/gemclean.rs +2954 -0
  78. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/gpu/context.rs +1 -0
  79. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/gpu/ffi.rs +1 -0
  80. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/gpu/fft.rs +1 -0
  81. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/gpu/invert.rs +1 -0
  82. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/gpu/mod.rs +1 -0
  83. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/gpu/shader.rs +1 -0
  84. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/gpu/tests.rs +1 -0
  85. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/grid.rs +11 -1
  86. rang_core-0.2.0/core/src/gridder_gram.rs +878 -0
  87. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ism.rs +1 -0
  88. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/jones.rs +1 -0
  89. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/lib.rs +12 -3
  90. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ms.rs +373 -45
  91. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ms_create.rs +1 -0
  92. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ms_desc.rs +1 -0
  93. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ms_schema.rs +1 -0
  94. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/ms_write.rs +287 -43
  95. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/multifile.rs +1 -0
  96. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/predict.rs +1 -0
  97. rang_core-0.2.0/core/src/prox_solver.rs +1509 -0
  98. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/robust.rs +1 -0
  99. rang_core-0.2.0/core/src/sara.rs +493 -0
  100. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/solve.rs +1 -0
  101. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/sumthreshold.rs +1 -0
  102. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/tiled.rs +1 -0
  103. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/weights.rs +221 -0
  104. {rang_core-0.1.0 → rang_core-0.2.0}/core/src/wgridder.rs +1021 -244
  105. rang_core-0.2.0/data/README.md +76 -0
  106. rang_core-0.2.0/data/meerkat.itrf.txt +64 -0
  107. {rang_core-0.1.0 → rang_core-0.2.0}/pyproject.toml +38 -9
  108. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/__init__.py +14 -13
  109. rang_core-0.2.0/python/rang/applycal.py +712 -0
  110. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/beam_holography.py +1 -0
  111. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/beam_identifiability.py +5 -4
  112. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/beams.py +1 -0
  113. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/blocked.py +1 -0
  114. rang_core-0.2.0/python/rang/cal_apply.py +337 -0
  115. rang_core-0.2.0/python/rang/cal_bandpass.py +307 -0
  116. rang_core-0.2.0/python/rang/cal_chain.py +640 -0
  117. rang_core-0.2.0/python/rang/cal_chunks.py +515 -0
  118. rang_core-0.2.0/python/rang/cal_delay.py +458 -0
  119. rang_core-0.2.0/python/rang/cal_fluxscale.py +441 -0
  120. rang_core-0.2.0/python/rang/cal_gain.py +1049 -0
  121. rang_core-0.2.0/python/rang/cal_pol.py +1411 -0
  122. rang_core-0.2.0/python/rang/cal_refant.py +264 -0
  123. rang_core-0.2.0/python/rang/cal_selfcal.py +443 -0
  124. rang_core-0.2.0/python/rang/calmodels.py +784 -0
  125. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/check.py +335 -33
  126. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core.py +25 -7
  127. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_actions.py +1 -0
  128. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_balanced.py +1 -0
  129. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_campaign.py +1 -0
  130. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_campaign_1830.py +1 -0
  131. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_campaign_array.py +1 -0
  132. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_campaign_closure.py +1 -0
  133. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_campaign_fieldmodel.py +1 -0
  134. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_cases.py +1 -0
  135. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_catalog.py +1 -0
  136. rang_core-0.2.0/python/rang/core_chunks.py +423 -0
  137. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_clean.py +1 -0
  138. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_cli.py +367 -38
  139. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_closure.py +1 -0
  140. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_compare.py +1 -0
  141. rang_core-0.2.0/python/rang/core_config.py +1958 -0
  142. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_controls.py +1 -0
  143. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_fetch.py +1 -0
  144. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_flag.py +17 -3
  145. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_forecast.py +1 -0
  146. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_grid.py +8 -0
  147. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_guard.py +1 -0
  148. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_image_fit.py +1 -0
  149. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_imaging.py +1 -0
  150. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_influence.py +1 -0
  151. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_injection_model.py +1 -0
  152. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_injections.py +1 -0
  153. rang_core-0.2.0/python/rang/core_model.py +222 -0
  154. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_ms.py +44 -3
  155. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_ms_write.py +21 -3
  156. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_patches.py +1 -0
  157. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_paths.py +1 -0
  158. rang_core-0.2.0/python/rang/core_pipeline.py +2716 -0
  159. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_policy.py +1 -0
  160. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_proposals.py +1 -0
  161. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_prospective.py +1 -0
  162. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_receipts.py +1 -0
  163. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_response.py +1 -0
  164. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_sensitivity.py +1 -0
  165. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_signal.py +1 -0
  166. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_solver_study.py +1 -0
  167. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_state.py +1 -0
  168. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_structured.py +1 -0
  169. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_templates.py +1 -0
  170. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_trend.py +1 -0
  171. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_trend_run.py +1 -0
  172. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_weights.py +1 -0
  173. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_wgridder.py +1 -0
  174. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_workflow.py +1 -0
  175. rang_core-0.2.0/python/rang/core_writeback.py +625 -0
  176. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/core_zarr.py +1 -0
  177. rang_core-0.2.0/python/rang/data/README.md +38 -0
  178. rang_core-0.2.0/python/rang/data/casacore_observatories.json +689 -0
  179. rang_core-0.2.0/python/rang/data/meerkat.itrf.txt +64 -0
  180. rang_core-0.2.0/python/rang/f2flag.py +504 -0
  181. rang_core-0.2.0/python/rang/f2flag_sim.py +545 -0
  182. rang_core-0.2.0/python/rang/f3flag.py +865 -0
  183. rang_core-0.2.0/python/rang/f3net.py +110 -0
  184. rang_core-0.2.0/python/rang/flag_extend.py +243 -0
  185. rang_core-0.2.0/python/rang/flagdata.py +1139 -0
  186. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/fluxstd.py +5 -0
  187. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/formats/__init__.py +1 -0
  188. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/formats/_aipsio.py +95 -11
  189. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/formats/caltable.py +21 -6
  190. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/formats/cli.py +1 -0
  191. rang_core-0.2.0/python/rang/formats/flagversions.py +284 -0
  192. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/formats/h5parm.py +1 -0
  193. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/formats/killms.py +1 -0
  194. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/formats/quartical_store.py +1 -0
  195. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/formats/sky.py +1 -0
  196. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/gauge.py +1 -0
  197. rang_core-0.2.0/python/rang/gemclean.py +33 -0
  198. rang_core-0.2.0/python/rang/listobs.py +546 -0
  199. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/lsm.py +1 -0
  200. rang_core-0.2.0/python/rang/mcp_server.py +202 -0
  201. rang_core-0.2.0/python/rang/mcp_tools.py +828 -0
  202. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/measures.py +32 -0
  203. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/meerkat.py +1 -0
  204. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/notebook/__init__.py +1 -0
  205. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/notebook/_plot.py +1 -0
  206. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/notebook/carta.py +1 -0
  207. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/notebook/data.py +1 -0
  208. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/notebook/experiment.py +1 -0
  209. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/notebook/flags.py +1 -0
  210. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/notebook/gains.py +1 -0
  211. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/notebook/image.py +1 -0
  212. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/notebook/science.py +25 -24
  213. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/notebook/steer.py +1 -0
  214. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/notebook/survey.py +1 -0
  215. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/observability.py +1 -0
  216. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/oracles/__init__.py +1 -0
  217. rang_core-0.2.0/python/rang/oracles/casa_listobs.py +301 -0
  218. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/oracles/casacure.py +1 -0
  219. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/pointing.py +1 -0
  220. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/pointing_selfcal.py +6 -5
  221. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/projector.py +3 -2
  222. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/rusage.py +1 -0
  223. rang_core-0.2.0/python/rang/simtruth.py +1127 -0
  224. rang_core-0.2.0/python/rang/simtruth_ms.py +600 -0
  225. rang_core-0.2.0/python/rang/tasks/__init__.py +313 -0
  226. rang_core-0.2.0/python/rang/tfcrop.py +205 -0
  227. rang_core-0.2.0/python/rang/tvflag_adjoint.py +585 -0
  228. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/uncertainty.py +1 -0
  229. {rang_core-0.1.0 → rang_core-0.2.0}/src/lib.rs +4 -3
  230. {rang_core-0.1.0 → rang_core-0.2.0}/src/linalg.rs +2 -1
  231. {rang_core-0.1.0 → rang_core-0.2.0}/src/main.rs +3 -2
  232. {rang_core-0.1.0 → rang_core-0.2.0}/src/projector.rs +1 -0
  233. {rang_core-0.1.0 → rang_core-0.2.0}/src/response.rs +2 -1
  234. {rang_core-0.1.0 → rang_core-0.2.0}/src/trend.rs +1 -0
  235. {rang_core-0.1.0 → rang_core-0.2.0}/src/tv.rs +1 -0
  236. {rang_core-0.1.0 → rang_core-0.2.0}/tests/conftest.py +2 -2
  237. rang_core-0.2.0/tests/data/strarray/oracle.json +1 -0
  238. rang_core-0.2.0/tests/data/strarray/table/table.dat +0 -0
  239. rang_core-0.2.0/tests/data/strarray/table/table.f0 +0 -0
  240. rang_core-0.2.0/tests/data/strarray/table/table.f1 +0 -0
  241. rang_core-0.2.0/tests/data/strarray/table/table.f1i +0 -0
  242. rang_core-0.2.0/tests/data/strarray/table/table.info +3 -0
  243. rang_core-0.2.0/tests/data/strarray/table/table.lock +0 -0
  244. rang_core-0.2.0/tests/data/w2/e2a.yaml +60 -0
  245. rang_core-0.2.0/tests/data/w2/e2b.yaml +74 -0
  246. rang_core-0.2.0/tests/test_a1_simtruth.py +233 -0
  247. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_b1_identifiability.py +6 -6
  248. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_b1_pointing_selfcal.py +25 -20
  249. rang_core-0.2.0/tests/test_c2e_calibrate.py +271 -0
  250. rang_core-0.2.0/tests/test_c4b_applycal.py +217 -0
  251. rang_core-0.2.0/tests/test_c5_selfcal.py +387 -0
  252. rang_core-0.2.0/tests/test_c5b_selfcal_masks.py +211 -0
  253. rang_core-0.2.0/tests/test_c6w_polwire.py +218 -0
  254. rang_core-0.2.0/tests/test_cal_apply.py +243 -0
  255. rang_core-0.2.0/tests/test_cal_bandpass.py +136 -0
  256. rang_core-0.2.0/tests/test_cal_delay.py +197 -0
  257. rang_core-0.2.0/tests/test_cal_fluxscale.py +217 -0
  258. rang_core-0.2.0/tests/test_cal_gain.py +173 -0
  259. rang_core-0.2.0/tests/test_cal_pol.py +276 -0
  260. rang_core-0.2.0/tests/test_calmodels.py +334 -0
  261. rang_core-0.2.0/tests/test_check_all.py +136 -0
  262. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_backends.py +2 -2
  263. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_cli.py +188 -16
  264. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_fetch.py +1 -1
  265. rang_core-0.2.0/tests/test_core_ms_strarray.py +97 -0
  266. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_ms_write.py +38 -2
  267. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_wgridder.py +4 -3
  268. rang_core-0.2.0/tests/test_correctness_audit.py +217 -0
  269. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_d1_docs.py +1 -1
  270. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_d3_notebooks.py +8 -8
  271. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_df2c_campaign_gridder.py +7 -3
  272. rang_core-0.2.0/tests/test_e0_refant_extend.py +290 -0
  273. rang_core-0.2.0/tests/test_e2_average_freq.py +176 -0
  274. rang_core-0.2.0/tests/test_e2_average_time.py +420 -0
  275. rang_core-0.2.0/tests/test_e5_decisions.py +42 -0
  276. rang_core-0.2.0/tests/test_e5_final.py +77 -0
  277. rang_core-0.2.0/tests/test_e5_later.py +169 -0
  278. rang_core-0.2.0/tests/test_e5_prereview.py +178 -0
  279. rang_core-0.2.0/tests/test_f1_adjoint.py +141 -0
  280. rang_core-0.2.0/tests/test_f2_cnnflag.py +55 -0
  281. rang_core-0.2.0/tests/test_f2_sim.py +34 -0
  282. rang_core-0.2.0/tests/test_f3_flag.py +104 -0
  283. rang_core-0.2.0/tests/test_g6_model.py +115 -0
  284. rang_core-0.2.0/tests/test_g7_defaults.py +16 -0
  285. rang_core-0.2.0/tests/test_g8_gemclean.py +236 -0
  286. rang_core-0.2.0/tests/test_g8_identity.py +46 -0
  287. rang_core-0.2.0/tests/test_heavy_lock_nested.py +207 -0
  288. rang_core-0.2.0/tests/test_heavy_lock_queue.py +223 -0
  289. rang_core-0.2.0/tests/test_mcp_server.py +378 -0
  290. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_native_support_diagnostics.py +1 -1
  291. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_notebook_execute.py +12 -1
  292. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_notebook_functions.py +1 -1
  293. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_p11_extract.py +6 -7
  294. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_p12_rename.py +5 -1
  295. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_p7_architecture.py +2 -2
  296. rang_core-0.1.0/tests/test_toy.py → rang_core-0.2.0/tests/test_pointing_sim.py +7 -7
  297. rang_core-0.2.0/tests/test_sliced_read.py +408 -0
  298. rang_core-0.2.0/tests/test_spdx_headers.py +58 -0
  299. rang_core-0.2.0/tests/test_v3_intro.py +135 -0
  300. rang_core-0.2.0/tests/test_v8a_listobs.py +207 -0
  301. rang_core-0.2.0/tests/test_v8b_flagdata.py +171 -0
  302. rang_core-0.2.0/tests/test_w1_writeback.py +234 -0
  303. rang_core-0.2.0/tests/test_w2_commands.py +1115 -0
  304. rang_core-0.2.0/tests/test_w3_chunking.py +411 -0
  305. rang_core-0.2.0/tests/test_w3b_chainchunks.py +355 -0
  306. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_w4_5_release.py +108 -3
  307. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_w4_6_portability.py +1 -1
  308. rang_core-0.2.0/tests/test_w4_unsupported.py +364 -0
  309. rang_core-0.2.0/tests/test_w5_recipes.py +97 -0
  310. rang_core-0.2.0/tests/test_xf_avg_fix.py +238 -0
  311. rang_core-0.2.0/tools/check.py +22 -0
  312. {rang_core-0.1.0 → rang_core-0.2.0}/tools/wheel/build-inside.sh +1 -0
  313. {rang_core-0.1.0 → rang_core-0.2.0}/tools/wheel/build.py +1 -0
  314. {rang_core-0.1.0 → rang_core-0.2.0}/tools/wheel/build.sh +1 -0
  315. {rang_core-0.1.0 → rang_core-0.2.0}/tools/wheel/classify_compare.py +1 -0
  316. {rang_core-0.1.0 → rang_core-0.2.0}/tools/wheel/pytest_wheel.py +2 -1
  317. {rang_core-0.1.0 → rang_core-0.2.0}/tools/wheel/run_installed.py +1 -0
  318. {rang_core-0.1.0 → rang_core-0.2.0}/tools/wheel/smoke.py +1 -0
  319. {rang_core-0.1.0 → rang_core-0.2.0}/tools/wheel/test.py +48 -6
  320. {rang_core-0.1.0 → rang_core-0.2.0}/tools/wheel/test.sh +4 -3
  321. rang_core-0.1.0/CHANGELOG.md +0 -80
  322. rang_core-0.1.0/python/rang/data/README.md +0 -13
  323. rang_core-0.1.0/python/rang/formats/flagversions.py +0 -106
  324. {rang_core-0.1.0 → rang_core-0.2.0}/LICENSE +0 -0
  325. {rang_core-0.1.0 → rang_core-0.2.0}/core/py/src/clean.rs +0 -0
  326. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/data/4k_lband_meerkat.staticmask +0 -0
  327. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/data/4k_uhfband_meerkat.staticmask +0 -0
  328. {rang_core-0.1.0 → rang_core-0.2.0}/python/rang/data/TRICOLOUR-LICENSE +0 -0
  329. {rang_core-0.1.0 → rang_core-0.2.0}/tests/data/core-backends-golden.json +0 -0
  330. {rang_core-0.1.0 → rang_core-0.2.0}/tests/data/core-backends-golden.npz +0 -0
  331. /rang_core-0.1.0/tests/data/toy-golden.json → /rang_core-0.2.0/tests/data/pointing-sim-golden.json +0 -0
  332. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_aflag.py +0 -0
  333. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_b1_baselines.py +0 -0
  334. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_beam_holography.py +0 -0
  335. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_beams.py +0 -0
  336. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_blocked.py +0 -0
  337. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_casacure_optional.py +0 -0
  338. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core.py +0 -0
  339. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_balanced.py +0 -0
  340. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_campaign.py +0 -0
  341. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_campaign_1830.py +0 -0
  342. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_campaign_array.py +0 -0
  343. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_campaign_closure.py +0 -0
  344. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_campaign_fieldmodel.py +0 -0
  345. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_cases.py +0 -0
  346. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_catalog.py +0 -0
  347. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_clean.py +0 -0
  348. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_closure.py +0 -0
  349. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_compare.py +0 -0
  350. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_controls.py +0 -0
  351. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_flag.py +0 -0
  352. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_grid.py +0 -0
  353. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_guard.py +0 -0
  354. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_image_fit.py +0 -0
  355. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_imaging.py +0 -0
  356. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_influence.py +0 -0
  357. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_injection_model.py +0 -0
  358. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_ms.py +0 -0
  359. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_patches.py +0 -0
  360. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_paths.py +0 -0
  361. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_policy.py +0 -0
  362. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_proposals.py +0 -0
  363. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_receipts.py +0 -0
  364. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_response.py +0 -0
  365. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_sensitivity.py +0 -0
  366. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_structured.py +0 -0
  367. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_templates.py +0 -0
  368. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_trend_run.py +0 -0
  369. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_trend_run_review.py +0 -0
  370. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_weights.py +0 -0
  371. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_workflow.py +0 -0
  372. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_core_zarr.py +0 -0
  373. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_df2b_wgrid.py +0 -0
  374. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_df5o.py +0 -0
  375. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_dfio_formats.py +0 -0
  376. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_e4_table_create.py +0 -0
  377. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_e5_table_writer_rest.py +0 -0
  378. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_e6_table_writer_final.py +0 -0
  379. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_flagging_followup_archives.py +0 -0
  380. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_fluxstd.py +0 -0
  381. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_information_budget.py +0 -0
  382. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_l5_solver.py +0 -0
  383. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_l5c.py +0 -0
  384. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_lsm.py +0 -0
  385. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_mk1_meerkat.py +0 -0
  386. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_ms_inspection.py +0 -0
  387. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_nb_carta.py +0 -0
  388. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_o_casa_optional.py +0 -0
  389. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_p1_solver.py +0 -0
  390. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_p2_gridder.py +0 -0
  391. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_p3_flags.py +0 -0
  392. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_p4_eigen.py +0 -0
  393. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_p5_memory.py +0 -0
  394. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_p6_signal.py +0 -0
  395. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_p6_trend.py +0 -0
  396. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_pc1_preconditions.py +0 -0
  397. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_pointing.py +0 -0
  398. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_pointing_example.py +0 -0
  399. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_projector.py +0 -0
  400. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_rv1_revalidate.py +0 -0
  401. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_t1_casacure_oracle.py +0 -0
  402. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_uncertainty.py +0 -0
  403. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_w4_1_receipts.py +0 -0
  404. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_w4_2_boundaries.py +0 -0
  405. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_w4_3_kernels.py +0 -0
  406. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_w4_8_docs.py +0 -0
  407. {rang_core-0.1.0 → rang_core-0.2.0}/tests/test_x2_gpu.py +0 -0
@@ -11,3 +11,8 @@
11
11
  # research/results/p11-extract/closure.py keeps its CRLF.
12
12
  research/results/** -text
13
13
  history/** -text
14
+ # CARACal's strategy file, byte for byte (SHA-256 in research/e2-acceptance.md 6.3):
15
+ research/results/e2-acceptance-scripts/mk_rfi_flagging_calibrator_fields_firstpass.yaml -whitespace
16
+
17
+ # casacore tables written by python-casacore, byte for byte (tools/oracles/strarray/fixture.py):
18
+ tests/data/strarray/table/** binary
@@ -0,0 +1,610 @@
1
+ # Changelog
2
+
3
+ All notable changes to rang-core (PyPI `rang-core`, package `rang`). The
4
+ version is set once, in `pyproject.toml`; the three `Cargo.toml`s, the
5
+ documentation and `rang.__version__` follow it, and
6
+ `tests/test_w4_5_release.py` checks that they agree. Pushing the tag `v<version>`
7
+ builds, tests and publishes that version (`.github/workflows/wheels.yml`);
8
+ the release notes are the section of this file for that version.
9
+
10
+ ## Unreleased
11
+
12
+ ## 0.2.0 (2026-11-14)
13
+
14
+ The owner released 0.2.0 on 2026-11-14. Its finish line
15
+ (`research/finish-plan.md` section 1) is met in part, and the parts not met
16
+ move to 0.3.0 (`research/roadmap.md`).
17
+
18
+ ### Known limitations
19
+
20
+ - **Not every stage is within its oracle tolerance.**
21
+ - E2-B is the acceptance run on processed LUNO MeerKAT L-band data against
22
+ CASA 6.7.6.14, WSClean and QuartiCal (`research/e2-acceptance.md` 7.3-7.5).
23
+ - **Met:** run and write-back, the readers (casacore, CASA listobs and
24
+ flagmanager, WSClean, dask-ms), FITS output, gridding (exact on all
25
+ 1.1e9 cells), K, KCROSS, Xf, Df, and 3C 286 polarisation parity
26
+ (|Δχ| ≤ 0.055°).
27
+ - **Not met, in the frozen and in the amended arm:**
28
+ - Gpol: median 0.124° in the amended arm, against a 0.1° limit.
29
+ - The fcal and gcal K G B composites: median 1.1e-3.
30
+ - CASA applycal on rang's tables: 40 cells over, max 25 eps; flags exact.
31
+ - Image gate 3c.
32
+ - Polarisation angle on 1830−211: |Δχ| up to 13°.
33
+ - Transfer χ (polarisation angle carried from calibrator to target).
34
+ - Rang's leakage: 0.30 %.
35
+ - The diagnosis (`research/e2b-misses.md`, from code and archived numbers;
36
+ no new runs) confirms no rang bug among the seven. Its likely causes:
37
+ - CASA's oracle solved on different flags and weights (composites);
38
+ - the estimators weight data differently (Gpol);
39
+ - float32 rounding in CASA's long apply chain (applycal);
40
+ - a scorer error and unmasked cleaning (3c);
41
+ - χ at p of 0.02-0.06 % (1830−211);
42
+ - LUNO's receptor angle and ionospheric RM (transfer χ).
43
+ - Pre-registered arms A1-A5 test these in v0.3.0.
44
+ - **The measurements still to come are now v0.3.0 work:**
45
+ - the E2-A run on ESO 137;
46
+ - E3, the comparison with CARACal's MeerKAT strategy (image quality, flux,
47
+ wall time and memory);
48
+ - the E5 review.
49
+ - **Native MeerKAT data is not tested.** The acceptance data is processed LUNO
50
+ data plus simulation (issue #7).
51
+ - **No absolute χ claim.** LUNO's χ offset is not explained; see
52
+ `research/chi-offset.md`.
53
+ - `flagdata` mode `tfcrop` is refused, and the trained flagger is not shipped.
54
+
55
+ ### Changes since 0.1.0
56
+
57
+
58
+ - Fixed (2026-11-14): a `tools/heavy_lock.py run` or `tools/check_all.py` started inside a heavy-lock
59
+ hold (`RANG_HEAVY_LOCK_HELD`, set by `heavy_lock.py run` for its child) runs inside that hold
60
+ instead of queueing again. A check wrapped in `heavy_lock.py run` held 4 slots idle for five hours
61
+ while its own inner ticket waited, and would deadlock on a full queue. A nested request for more
62
+ than the hold warns and still runs.
63
+ - New (2026-11-14): `tools/heavy_lock.py status` prints each hold's real CPU use (its process tree,
64
+ sampled for 2 s) next to its booked slots, marks a hold IDLE under 5 % of one core per booked slot,
65
+ and ends with `booked N slots, using ~M cores`. Report only; nothing is stopped.
66
+
67
+ - Fixed (correctness audit, `research/correctness-audit-2026-11-13.md`): bad input now fails clearly
68
+ instead of running wrong. A NaN visibility made every gain, bandpass, delay or Df solution NaN with
69
+ its flag clear; non-finite data, model and weight cells are now flagged, which gives exactly the
70
+ solution of the same cells flagged (gain, bandpass, delay, KCROSS, Xf, Df). `core_grid.invert(gridder="wgrid")` with a GPU device raises instead of
71
+ running on the CPU. gemclean's FITS reader (`-predict` models, `-fits-mask`) no longer panics on a
72
+ non-ASCII header byte. `-threshold`, `-abs-threshold` and `-auto-threshold` must be finite and not
73
+ negative. The gain and bandpass solvers refuse zero rows and per-row arrays of the wrong length; delay refuses zero rows.
74
+ Write-back refuses out-of-range DATA_DESCRIPTION ids (a negative one picked the last row). Flag-version
75
+ names with `\ : * ? " < > |` are refused on every platform. `fluxstd.manual` refuses non-positive
76
+ frequencies. No default's numbers change.
77
+
78
+ - Changed (task df-flag-rule, `research/p-flag-spread.md`): with `apply.parang`, P now flags as CASA's
79
+ `applycal(parang=True)` does on linear feeds: a cell with any correlation flagged after the tables
80
+ (input flags, a bad receptor, a Df-flagged cross hand) is flagged in all four, since P rotates XX and
81
+ YY out of all four correlations. New key `apply.parang_flags` (`spread`, the default; `none`, the
82
+ previous flags) and `parang_flags=` on `rang.applycal.applycal` and `rang.cal_pol.apply`. Values are
83
+ unchanged; without `parang` nothing changes.
84
+
85
+ - Fixed (E5 final E5-02): `gemclean -model-column DATA`, or a model column equal to the column being
86
+ imaged (`-data-column`, or CORRECTED_DATA when the MS has it), overwrote that column after imaging;
87
+ both are refused with exit 2 before anything is read. `-no-update-model-required` still writes nothing.
88
+ - Changed (E5 final F-05): the licence page and `THIRD-PARTY-NOTICES.txt` list CASA's C++ sources that
89
+ the calibration and flag-extension ports follow (GPL-2.0-or-later, LGPL-2.0/2.1-or-later), with
90
+ their copyright notices; each derived file names CASA.
91
+ - Changed (E5 final F-01): astropy is a required dependency. Every polarisation solve and `apply.parang`
92
+ take the parallactic angle from it, and a bare `pip install rang-core` stopped there with
93
+ `ModuleNotFoundError`; the `imaging` extra is kept and now adds nothing.
94
+ - New: a known-limits page (`docs/limits.md`, linked from the README): scope, flagging, calibration,
95
+ self-calibration, imaging and platform limits with the reports behind them; the E2 and E3 results
96
+ are added when they land (E5 final, `research/e5-final.md`).
97
+ - Changed (finish-plan item 27, f11fdd1): an unset `calibrate.refant` (it was refused) means CARACal's
98
+ `auto` choice, or antenna 0 when no antenna qualifies; the default selfcal reference antenna is the
99
+ same choice (it was antenna 0), and `selfcal.refant: auto` runs; the `image.niter` default is
100
+ 1,000,000 (it was 100,000).
101
+ - Changed: Measurement Set reads take a channel slice in the reader (`core/src/ms.rs`), 2.1-2.2x
102
+ faster on the channel-chunked polarisation solves and bit-identical on 19 frozen comparisons
103
+ (`research/sliced-read.md`).
104
+ - Changed (E5 pre-review "Later" findings, `research/e5-prereview.md` section 2.3): `image.fits_mask`
105
+ and a selfcal FITS `cleanmask_method` are read relative to the configuration file, as `flag.strategy`
106
+ and `calibrate.model.lsm`; **deprecated:** a mask found only relative to the working directory is still
107
+ used, with a warning, and a mask found in neither place is refused before anything runs (E5-41).
108
+ Repeated YAML keys are refused with their line numbers (PyYAML kept the last one). Unknown commands and
109
+ options exit 2 on stderr (was 1); `--version` works after other global options; `run --force STAGE` of a
110
+ stage outside `run.stages` is refused; the complex-term refusals and unknown term keys say what to change
111
+ (E5-40). Data whose OBSERVATION `TELESCOPE_NAME` is not MeerKAT run with a warning, and `state.json`
112
+ records `telescope` (E5-39). A flag version left unlisted by a kill is listed by the next save of the same
113
+ flags; a flag version the last attempt recorded but the MS no longer has is warned about
114
+ (`restore_missing`) (E5-42). Every source file carries an SPDX `GPL-3.0-or-later` line (E5-45).
115
+
116
+ - New: `calibrate.average_time` (default false; task E2, `research/e2-average-time.md`). With it on,
117
+ the K, G, B, F, Gpol, KCROSS, Xf and Df solves use CASA's estimator for long intervals: data and
118
+ model averaged in time within each solution interval (weighted means over unflagged cells after the
119
+ pre-applied tables and P, weight the sum, per baseline, channel and correlation), streamed in blocks
120
+ of rows, and the memory plan counts the averaged rows. E2's whole-scan solves on 4096-channel
121
+ MeerKAT data (176 GiB per interval unaveraged) then fit `data.memory_limit_gb: 8`. Off, every
122
+ solve and the stage hash are unchanged. Refused with a `complex` term.
123
+ - New: `selfcal.average_time` (default false; task E2): Sp and Sa solve on data and model averaged
124
+ in time within each solution interval, as `calibrate.average_time` does, so CARACal's solint 120 fits
125
+ 8 GiB at 4,096 channels (333 GiB per interval without). Not CubiCal's estimator, which solves every
126
+ sample (owner item 33(3)). Off, the stage hash is unchanged. Refused with Sd.
127
+ - New: `calibrate.average_freq` (default false; task E2, the pre-registered "E2-B amended" arm,
128
+ `research/e2-average-freq.md`): the G, F and Gpol solves average their data and model over the
129
+ window's channels before solving (weighted means per row and correlation after the pre-applies and
130
+ P), as CASA's G-type solves do; K, B, KCROSS, Xf and Df are unchanged. Off, every solve and the
131
+ stage hash are unchanged.
132
+
133
+ - Changed (owner decisions on the E5 pre-review, `research/e5-prereview.md` section 3): receipts are
134
+ `rang.receipt` version 5; a Measurement Set a command modified is listed by its `listing_sha256` (file
135
+ sizes and mtimes), not a full content hash; the stage summaries hash the flags, gain solutions and images
136
+ written and name the tables and settings behind a written column (E5-20). This holds for every directory
137
+ a command modified, not only the MS. The default `run.stages` is the whole chain, flag, calibrate, apply, image, selfcal (was flag,
138
+ calibrate, image, which imaged uncalibrated DATA) (E5-28). The simms antenna tables `data/*.itrf.txt` are
139
+ recorded in `docs/licence.md` as facts, not code; the sdist carries only `meerkat.itrf.txt`, which GeMClean
140
+ compiles in (E5-11). `docs/licence.md` lists codex-africanus and katdal's test helpers (BSD-3), and as
141
+ behaviour reimplemented with no code copied CARACal's `get_refant`, msutils, CASA's flag extension,
142
+ fluxscale and applycal semantics and casacore's table formats, plus casacore-data's observatory
143
+ positions (E5-33). `rang/data/README.md` records the source and SHA256 of `casacore_observatories.json`
144
+ (E5-34). The sdist leaves out `.claude/` and the checkout tools under `tools/` other than `tools/wheel/`,
145
+ and CONTRIBUTING lists what it carries (E5-37). The `recipes` extra installs stimela 2.1.4 on Python 3.13
146
+ and older only, so `pip install 'rang-core[recipes]'` resolves on 3.14 (E5-38).
147
+
148
+ - New (task E4, `research/e4-tutorial.md`): the tutorial `notebooks/00-ms-to-image.ipynb`, from a
149
+ Measurement Set to an image with `rang-core run` and `gemclean` on A1 small, with its configuration
150
+ and strategy in `examples/ms-to-image/`; the wheels workflow's job `e2e` runs the same chain on all
151
+ five wheels (`tools/e2e/run.py`); `docs/cli.md` gains `gemclean -h`. `examples/a1_simtruth.py`
152
+ finds the checkout's antenna table when rang-core is installed from a wheel.
153
+
154
+ - Fixed (E5 pre-review, `research/e5-prereview.md`): after `rang-core run ... --force flag`, a stale later stage
155
+ put back the flags saved before its earlier attempt (the old flag output); later stages now start from the
156
+ flags the rerun stage left, whenever that stage restored or wrote FLAG. Before a rerun restores a stage's
157
+ earlier flags, the flags it replaces are saved as the flag version `rang_rerun_<UTC time>`, so flags set
158
+ outside rang are never lost (one more version per rerun). DATA is never written: `apply.column: DATA` and
159
+ `selfcal.model_column: DATA` are configuration errors, selfcal refuses to copy apply's output into an
160
+ `input_column: DATA`, and `rang.core_writeback.WriteBack` refuses DATA. The run directory is locked on
161
+ Windows too, so a second `rang-core` command on the same MS is refused. Malformed configuration values give
162
+ the error report (exit 2), not a traceback. The changed-configuration refusal names `--force` forms both
163
+ commands accept. The wheels and sdist carry `THIRD-PARTY-NOTICES.txt` (QuartiCal's MIT, codex-africanus's
164
+ and Tricolour's BSD-3 notices). The release job publishes to PyPI only after its own checks pass. The MS
165
+ table lock holds on Linux only, as the docs now say.
166
+
167
+ - Research (task F3, `research/f3-flagger.md`): a second learned RFI flagger, not wired into
168
+ `rang-core flag`. New `rang.f3flag` (wrapper W3, features reference, injections, labels, measures),
169
+ `rang.f3net` (the JAX network for training), the Rust kernel `rang_core::f3flag` with
170
+ `rang._core.f3flag_features` and `f3flag_logits`, the AOFlagger driver
171
+ `tools/oracles/aoflagger/f3_aoflag.py` and `examples/f3_*.py`. `rang.f2flag_sim` gains the
172
+ held-out observation `f3sim-t` (seeds registered in `rang.simtruth.SEEDS`).
173
+
174
+ - New: `gemclean` is in the wheel (task G8, `research/g8-wheel.md`): the console script `gemclean`
175
+ (and `python -m rang.gemclean`) runs GeMClean's command line inside the extension
176
+ (`rang._core.gemclean_main`); no second binary ships. The command line moved from
177
+ `core/src/bin/gemclean.rs` into `rang_core::gemclean::cli`, which the source build's binary now
178
+ wraps; files, output and exit codes are the binary's (0, 2 on an error, 101 on a panic), and
179
+ Ctrl-C ends the run at once. New `-version` (also `--version`). On 64-bit Linux the extension
180
+ uses gemclean's huge-page allocator, so `-memory-report` works from the wheel.
181
+ - Changed: `rang-core`'s `image.gemclean: auto` runs the installed extension's gemclean
182
+ (`python -m rang.gemclean` of the running interpreter) and records the sha256 of `rang._core`;
183
+ the search of PATH and of the checkout's `core/target` is gone. A source build is used only when
184
+ `image.gemclean` names it.
185
+
186
+ - Changed: `gemclean` grids at `-wgridder-accuracy 1e-4` in single precision by default (task G7,
187
+ `research/g7-defaults.md`): on the 512-channel ESO 137 pilot 33 % less wall at `-j 16`, 27 % at
188
+ `-j 1` and 4.1 GB less memory, within the class B bounds written first. Single precision is the
189
+ default for an accuracy of 1e-5 or coarser; the new `-double` forces double precision, and
190
+ `-wgridder-accuracy 1e-6` alone gives the earlier form. `rang-core`'s `image.wgridder_accuracy`
191
+ defaults to 1e-4, the model column follows gemclean's precision, and the image and selfcal stage
192
+ hashes change once (`--force`).
193
+ - Fixed: `gemclean`'s restoring beam is fitted as WSClean 3.7's binary fits it (a unit-peak Gaussian
194
+ in linear space over a box of 10 theoretic beams, regrown up to five times, on the box's quadrant
195
+ that schaapcommon's unsigned offsets leave in; the start is the theoretic beam, raised to the
196
+ Gaussian taper). BMAJ and BMIN agree with WSClean's to 4.3e-7 on 19 cases; before, BMAJ was 9-30 %
197
+ narrower. Every restored image and every FITS header's `BMAJ`, `BMIN`, `BPA` change; dirty, PSF,
198
+ model and residual data do not. The report's `main_lobe_pixels` is now `fit_pixels`.
199
+
200
+ - New: per-image clean masks and cut-offs in `rang-core selfcal` (`research/c5b-selfcal-masks.md`),
201
+ CARACal's `clean_cutoff`, `cleanmask_method` and `cleanmask_thr` as `selfcal.clean_cutoff`,
202
+ `selfcal.cleanmask_method` (`wsclean`, `none` or a FITS mask path) and `selfcal.cleanmask_thr`:
203
+ lists over images 1 to `rounds + 1`, the last entry repeating, passed to gemclean as
204
+ `-auto-threshold`, `-auto-mask` and `-fits-mask`. Empty by default (images as before); `sofia`,
205
+ `breizorro` and local-rms masks are refused. The selfcal summary records each image's
206
+ components, major cycles and stop reason. The new keys change the selfcal stage hash once
207
+ (`--force`). Measured on C5's truths: CARACal's masks do not meet C5's T2 gate (held-out cases
208
+ included) and do not lift C5's model-absorption limit; archive
209
+ `research/results/c5b-selfcal-masks.json`.
210
+ - Research: E1, the whole chain on A1's simulated truth (`research/e1-chain.md`): `rang-core run`
211
+ (flag, CARACal's full-Stokes calibrate, apply, selfcal, image) on A1 full's first 32 integrations
212
+ per scan exits 0; no E1 gate of A0 is met as frozen and no rang bug was located (the K failures are
213
+ CASA's own on the same flags). Scripts in `research/results/e1-chain-scripts/`, archive
214
+ `research/results/e1-chain.json`. No package code changed.
215
+ - New: GeMClean masks, tapers and uv limits (`research/g4-masks.md`), WSClean 3.7's options:
216
+ `-fits-mask` (one plane, the image's size); radler's automatic mask `-auto-mask`,
217
+ `-abs-auto-mask`, `-auto-mask-nmiter` in the default `dual` mode (Högbom, joined channels and
218
+ multiscale with its per-scale masks; needs `-mgain` or `-minor-to-threshold`);
219
+ `-taper-gaussian` (arcseconds by default), `-taper-tukey`, `-taper-inner-tukey`, `-taper-edge`,
220
+ `-taper-edge-tukey`, `-minuv-l` and `-maxuv-l` on the weight grid's cells, and `-minuvw-m`,
221
+ `-maxuvw-m` on rows. `rang-core image` passes `auto_mask`, `abs_auto_mask`,
222
+ `auto_mask_nmiter`, `fits_mask`, `taper_gaussian`, `taper_tukey`, `taper_inner_tukey`,
223
+ `minuv_l`, `maxuv_l`, `minuvw_m`, `maxuvw_m`; `image.uvrange` stays refused (use those keys),
224
+ and `auto_mask` without `mgain` is a configuration refusal. `-local-rms`, `-casa-mask`,
225
+ `-horizon-mask` and `-clean-border` are refused.
226
+ - Fixed: `image.gemclean: auto` finds `core/target/<profile>/gemclean.exe` on Windows. The test
227
+ stand-in for gemclean is a Python script (a `.cmd` wrapper on Windows), and the selfcal refusal
228
+ test no longer needs a built binary. `examples/a1_simtruth.py` uses spawned workers where fork is
229
+ missing (Windows); fork elsewhere, unchanged. New `tools/heavy_lock.py`, one heavy job at a time across
230
+ Windows and WSL; CONTRIBUTING describes native Windows development.
231
+ - New: `gemclean -pol` and `-join-polarizations` (`research/g9-stokes.md`): Stokes I, Q, U, V and
232
+ XX, YY in any combination, parsed as WSClean 3.7 parses `-pol` (`IQUV`, `iquv`, `xx,yy`), formed
233
+ from linear-feed correlations with WSClean's arithmetic and weights (`min(4 w_a, 4 w_b)` for a
234
+ formed Stokes parameter); WSClean's file names (`<name>[-0000]-<pol>-<kind>.fits`, one PSF per
235
+ channel from the first polarisation), `CRVAL4`, imaging-weight sources and header weights; joined
236
+ polarisations pick peaks on the normalised linear sum, as WSClean 3.7's default loop does (its
237
+ help text says the sum of squares; recorded in the note), so with QU joined, components whose
238
+ Q and U cancel are not cleaned, in GeMClean as in WSClean 3.7 (IQUV joining, as E2-B's, is far
239
+ less exposed); both work with `-channels-out` and
240
+ `-join-channels`. XY, YX, circular correlations and `-link-polarizations` are refused.
241
+ `rang-core image` passes `image.pol` and `image.join_polarizations` (only when not the defaults),
242
+ so E2-B's image section is no longer refused. Single-polarisation I imaging is unchanged.
243
+ - New: `gemclean -multiscale` (`research/g5-multiscale.md`): WSClean 3.7's multiscale CLEAN,
244
+ radler's algorithm with its sub-minor loop (`rang_core::gemclean::multiscale`), with
245
+ `-multiscale-scales`, `-multiscale-max-scales`, `-multiscale-scale-bias`, `-multiscale-shape`
246
+ (tapered quadratic or Gaussian) and `-multiscale-gain`; scales from WSClean's theoretic beam;
247
+ the residual made exact for the image-domain beam after every sub-minor loop. Works with one
248
+ output channel, joined channels (with `-fit-spectral-pol`) and unjoined ones, and with
249
+ `-auto-threshold`. `rang-core image` passes `multiscale` and the new keys `multiscale_scales`,
250
+ `multiscale_max_scales`, `multiscale_scale_bias`, `multiscale_shape` and `multiscale_gain`.
251
+ Without `-multiscale` the images are byte-identical to before.
252
+ - New: `rang-core selfcal` (and `run` with `selfcal` in its stages) runs CARACal's self-calibration
253
+ rounds (`research/c5-selfcal.md`): `selfcal.rounds` rounds of a gemclean image (the input column
254
+ first, the output column after), its model images predicted into `selfcal.model_column`
255
+ (`rang.core_model.write_model_data`), one term solved on `selfcal.input_column` against it and
256
+ applied into `selfcal.output_column`, then a final image. One term per round, the last repeating
257
+ (CARACal's `gain_matrix_type`): `Sp` (phase, QuartiCal `phase`), `Sa` (amplitude and phase,
258
+ `diag_complex`) and `Sd` (CARACal's `Fslope`, QuartiCal `delay_and_offset`, started at zero
259
+ delay); default phase then amplitude. An integer `solint` is N integrations from each scan start
260
+ (CARACal's timeslots), `Ns` CASA's seconds bins; `freqint` N channels for Sp and Sa. Solves are
261
+ chunked within `data.memory_limit_gb`; `selfcal.gain_clip` flags solutions, `selfcal.minuvw_m`
262
+ drops short baselines, new key `selfcal.refant`. `CORRECTED_1GC` is copied from `apply.column`
263
+ after this run's apply. Solutions go to a QuartiCal gain store per round. New module
264
+ `rang.cal_selfcal`; `rang.cal_gain`'s chunked solve takes channel chunks. `complex`
265
+ self-calibration terms are refused (`refant: auto` in selfcal was refused too, until item 27).
266
+ - New: `rang-core calibrate` runs the K, G, B, A and F terms and the polarisation terms Gpol,
267
+ KCROSS, Xf and Df within `data.memory_limit_gb` (`research/w3b-chainchunks.md`): each solve reads
268
+ its field in chunks of whole solution intervals (B, Xf and Df also in channels of one interval),
269
+ pre-applied on whole rows, with the result of the one-call solve bit for bit. The stop test of the
270
+ G, B and K kernels is replayed over the chunks from new counted kernels (`rang._core.cal_gain_trace`,
271
+ `cal_delay_trace`, `cal_delay_offset_trace`; Rust `diag::solve_counted`, `delay::solve_counted`,
272
+ `delay_offset::solve_counted`), Df's gauge test is found exactly over the chunks, and the A step's
273
+ Tricolour pass flags in blocks of baselines on each group's time grid. `preflight` no longer refuses
274
+ these terms above the limit; it refuses only a solution interval (B, Xf, Df: one channel of one)
275
+ that does not fit, naming the GiB. The library functions take a `budget` (default: one chunk, as
276
+ before). P's angles come from a table per unique time and antenna (`rang.cal_pol.angle_table`,
277
+ `table_angles`), so `applycal(parang=True)` and the polarisation solves hold no per-row angle array.
278
+ New module `rang.cal_chunks`. The `extendflags` step (`rang.flag_extend.extend_ms`) splits a 60 s chunk
279
+ larger than its batch into blocks of whole baselines, with the same flags.
280
+ - New: `refant: auto` and `calibrate.polcal.extendflags: true` run in `rang-core calibrate`
281
+ (`research/e0-refant-extend.md`). `auto` is CARACal's `get_refant` (`rang.cal_refant`): the
282
+ first three antennas **by name** within `calibrate.refant_maxdist` (1000 m) of the array centre
283
+ whose every baseline is at least `calibrate.refant_minbase` (150 m); flags play no part, as in
284
+ CARACal's code (its comment says fewest flags). It is chosen once before anything is written, and
285
+ the stage stops when no antenna qualifies. `extendflags` is CASA's `flagdata(mode='extend')`
286
+ with CARACal's settings (`ntime='60s'`, `combinescans`, `growtime` and `growfreq` 80,
287
+ `growaround`, `flagneartime`, `flagnearfreq`, `extendpols`; `rang.flag_extend`, Rust kernel
288
+ `core/src/flag_extend.rs`), run before the polcal solves on the Gpol, KCROSS and Xf fields and then
289
+ the Df fields, within `data.scans`. The two new keys change the calibrate stage hash, so a
290
+ calibrate stage finished earlier reruns once with `--force`.
291
+ - New (research, task F2, `research/f2-flagger.md`): `rang.f2flag`, a 2,785-weight per-baseline
292
+ convolutional RFI flagger (features and network in NumPy/JAX, training in JAX on CPU), its Rust
293
+ inference kernel `rang_core::cnnflag` (`rang._core.cnnflag_logits`, rows in, logits out, parity with
294
+ JAX tested), and `rang.f2flag_sim`, simulated MeerKAT L-band scans on LUNO's window with RFI of known
295
+ mask from `rang.simtruth` (F2's seeds registered in `simtruth.SEEDS`). Not wired into `rang-core
296
+ flag`; weights are run outputs, not shipped.
297
+ - New: `gemclean` writes the model column after imaging, as WSClean 3.7 does (`-update-model-required`,
298
+ the default; `-no-update-model-required` leaves the MS unwritten), and has WSClean's `-predict` mode
299
+ and `-model-column` (`research/g6-predict.md`). The model images (`<name>-model.fits` or
300
+ `<name>-000k-model.fits`, gemclean's or WSClean's) are predicted with the gridder at the MS's own
301
+ channels into every selected row (field, data description 0, cross-correlations; flags play no part):
302
+ XX = YY = I, XY = YX = 0. An added column is TiledShapeStMan and starts at zero, as WSClean's does.
303
+ `rang-core image` with `image.write_model: true` writes MODEL_DATA through W1's write-back with the
304
+ same predictor (`rang.core_model`, `rang._core.gemclean_predict`), and always runs gemclean with
305
+ `-no-update-model-required`. The scripts in `examples/` that image shared data now pass
306
+ `-no-update-model-required`.
307
+ - New: `gemclean -scans <list>` and `-unit-weights` (rang's own options). gemclean now falls back to WEIGHT
308
+ as WSClean does: WEIGHT_SPECTRUM is read only when row 0 holds a cell of the data's shape
309
+ (`research/w6-imaging-inputs.md`). `rang-core image` passes `data.scans` and `data.weights: uniform`
310
+ to these options. It no longer refuses either of them, nor a declared WEIGHT_SPECTRUM with unwritten
311
+ cells, and it writes nothing into the MS for uniform weights. An imaged field with no rows in
312
+ `data.scans` is refused.
313
+
314
+ - New: the Stimela2 recipe `recipes/rang-chain.yaml` runs `rang-core flag`, `calibrate`, `apply`
315
+ and `image` in order on one configuration and MS, with the cabs `rang-flag`, `rang-calibrate`,
316
+ `rang-apply` and `rang-image` in `recipes/rang-cabs.yaml`; each cab runs the console script
317
+ (`research/w5-stimela.md`). The recipes `recipes/core-*.yaml` are unchanged in their steps and
318
+ now start with `# rang-recipe: research`: they replay frozen campaigns. `tests/test_w5_recipes.py`
319
+ checks the marker on every recipe and the cabs' arguments against the commands.
320
+ - New: `rang-core calibrate` runs the polcal section's Gpol, KCROSS, Xf and Df (CARACal's strategy 1,
321
+ `rang.cal_pol`; `research/c6w-polwire.md`) after the primary and secondary terms, and `rang-core
322
+ apply` corrects by the parallactic angle with `apply.parang: true` (after the tables, as CASA's
323
+ `parang=True`; `rang.applycal.applycal(parang=True)`, angles per field independent of the
324
+ chunking). New keys: `calibrate.polcal.model` (`nrao`: CARACal's `nrao_xcal.yml` 3C286 or 3C138
325
+ written to MODEL_DATA on the polcal fields; `column`), `calibrate.polcal.frame` (`casa` default),
326
+ KCROSS `estimator` (`casa` default, `refined`), Df `min_data_fraction` (off by default; CASA has
327
+ no such rule). Df's table is clipped at CARACal's 0.6 by default. `feed_angle_rotation` now sets
328
+ the receptor angle used for P, in memory (CARACal sets it in FEED); FEED is not rewritten. Applying
329
+ polarisation tables without `apply.parang`, `preavg`, Gpol `calmode: ap` and `extendflags: true`
330
+ (not ported, no owner) are configuration errors; nothing is refused for C6a, C6b or C6c any more.
331
+ The defaults of C6's open decisions are provisional.
332
+ - New: `rang.cal_pol`, the polarisation terms in CARACal's order (`research/c6-polterms.md`): `gpol`
333
+ (phase-only G with the parallactic angle), `kcross` (CASA's cross-hand delay FFT), `xf` (CASA's
334
+ cross-hand phase), `df` (the linearised leakage by least squares with CASA's subtractive reference),
335
+ `clip_df`, `parallactic_angles`/`row_angles` (CASA's frame by default, provisionally; `frame='rang'`
336
+ keeps the geodetic one), `rotate` and `apply` (tables, then P). Rust kernels
337
+ `core/src/cal/parang.rs`, `crosshand.rs`, `leakage.rs`; `cal/apply.rs` and `formats/caltable.py`
338
+ now apply `Kcross Jones` and `Xf Jones` tables. Wired into `rang-core calibrate` and `apply` by
339
+ task C6w (the entry above).
340
+ `rang.simtruth` gains the versions `poldx` and `polholdout` (CASA's parallactic frame, a leaking
341
+ reference antenna); the other versions are unchanged.
342
+ - New: `rang-core calibrate` runs CARACal's cross-calibration terms K, G, B, A and F in C1's order
343
+ (`research/c2e-calibrate.md`). The config's `calibrate.primary` and `calibrate.secondary` terms
344
+ go to `rang.cal_chain.run`: K (`delay_and_offset`), G and B with their listed tables pre-applied,
345
+ written as CASA tables in the run directory; F appends a G solved on the secondary against a
346
+ unit point model to a copy of the primary's last G, then `fluxscale`, and the fitted flux densities
347
+ go into the summary; A applies the section's last table of each type to its fields
348
+ (CORRECTED_DATA, `calflag`) and runs the Tricolour port on the corrected data, OR-ing its flags
349
+ into FLAG (CARACal runs tfcrop there). `rang-core apply` applies these tables through
350
+ `rang.applycal` with each target's `interp` and `gainfield`. calibrate now saves and restores a
351
+ flag version like the other writing stages. A K table applied without a G or F is reported:
352
+ its phase offsets are left for G. Refusals no longer name C2a to C2d; what the `complex` term
353
+ never takes is a configuration error pointing to the K, G and B terms. `refant: auto`,
354
+ `solnorm: true` and `calmode: a` are refused as not ported.
355
+ - Changed: `rang.cal_gain.gaincal`, `rang.cal_delay.gaincal`, `rang.cal_bandpass.bandpass` and
356
+ `rang.applycal.applycal` take `scan` (a list of SCAN_NUMBERs); the solves take
357
+ `weight_column=None` for unit weights and read a WEIGHT column per channel. Defaults are
358
+ unchanged. `rang.simtruth` has a held-out version `holdout128` (new seeds).
359
+ - Added: polarised calibrator models (task C7a, `research/c7a-polmodels.md`).
360
+ `rang.calmodels.Manual` takes CASA `setjy(standard='manual')`'s full form, `fluxdensity=[I, Q,
361
+ U, V]`, `polindex`, `polangle` and `rotmeas` (referenced to `reffreq`, as CASA), following
362
+ CASA's tabular and spectral-index paths, and writes all four correlations in the sky frame
363
+ (`XX = I+Q`, `XY = U+iV`, `YX = U-iV`, `YY = I-Q`; no parallactic angle, as `setjy`).
364
+ With one `spix` term it copies CASA's TOPO-to-LSRK conversion of channel 0
365
+ (`calmodels.lsrk_over_topo`, astropy; recorded in the receipt), within 1.5e-7 of CASA.
366
+ `calmodels.NRAO_XCAL` and `caracal_xcal()` hold CARACal's 3C286 and 3C138 models.
367
+ `rang-core calibrate` accepts `calibrate.model.flux_jy` with Q, U, V and the polarisation
368
+ keys for `source: manual`; `polindex` without `polangle` (or the reverse) and the polarisation
369
+ keys with another source are refused. Polarised sky-model components stay refused.
370
+ - New: `rang-core flag`, `calibrate` and `apply` run within `data.memory_limit_gb` (default 8 GiB of
371
+ resident memory) by chunking by time and row (`rang.core_chunks`, `research/w3-chunking.md`):
372
+ the flagger takes blocks of baselines of a field and scan on the group's whole time grid
373
+ (`core_flag.flag_rows(time_grid=...)`), the `complex` solve takes chunks of whole solution
374
+ intervals (time bins, or frequency bins of one time bin) and replays the solver's global stop
375
+ test over them, so gains equal the one-call solve bit for bit, and apply and the models use row
376
+ chunks of the limit. Results do not depend on the limit or on `data.chunk_rows` (now a cap on
377
+ the rows of every chunk), so neither enters a stage's configuration hash. A limit too small for
378
+ the process, one baseline of a scan or one solution interval is refused before anything is
379
+ written, naming the GiB needed. No visibility column is read whole by the three stages;
380
+ `cal_gain` and `cal_delay` read only the rows of the field they solve. The K, G, B, A and F
381
+ terms are not chunked yet: a field they would read beyond the limit is refused, naming the GiB
382
+ needed.
383
+ - New: A1's generator has a `w3` version (64 antennas, 4096 channels) and `--chunk-ints`.
384
+
385
+ - Changed: `rang-core flag`, `calibrate`, `apply`, `image` and `config check CONFIG MS` check the
386
+ Measurement Set per stage before anything is written and report every problem at once, each
387
+ naming its reason (`research/w4-unsupported.md`). Refused: several data descriptions in use
388
+ (split one spectral window out first), circular, single or reordered correlations, two
389
+ correlations for `calibrate` and `apply` (the complex term needs four), no FLAG or FLAG_ROW,
390
+ no weights for `calibrate` (unless `data.weights: uniform`) or `image`, a FIELD_ID without a
391
+ FIELD row, a named field or a `data.scans` selection with no rows, a FIELD name of several
392
+ fields in `image.fields`, and `image` when `data.scans` leaves out a scan of an imaged field
393
+ (gemclean has no scan selection), on a data description other than 0 or with a
394
+ WEIGHT_SPECTRUM column whose cells were never written. Two linear correlations (XX, YY)
395
+ are now flagged and imaged. Autocorrelations are flagged and corrected but left out of the
396
+ `complex` solve; `calibrate` reads WEIGHT when row 0 of WEIGHT_SPECTRUM has no cell of the
397
+ DATA shape, and `image` refuses such an MS because gemclean would stop on it.
398
+ `data.weights: uniform` for imaging is still refused, now naming W4 (an owner decision).
399
+ - New: `rang.cal_fluxscale`, the flux-scale term F (`research/c2d-fluxscale.md`). `fluxscale(caltable,
400
+ reference=..., transfer=...)` is CASA's `fluxscale` on a G table, with its statistics in the Rust
401
+ core (`core/src/cal/fluxscale.rs`): per-antenna mean amplitudes, the median ratio to the reference
402
+ (casacore's median rule), `fd = ratio²` and CASA's error, `gainthreshold`, several reference
403
+ fields, `refspwmap`, and the weighted spectral fit across windows (`fitorder`); it returns the
404
+ flux-scaled table and a record shaped as CASA's. `bootstrap(ms, primary, secondary)` runs
405
+ CARACal's F step: G on the secondaries against a unit point model (`smodel_array`), appended to a
406
+ copy of the primary's table (`append`), then `fluxscale`. Selections by antenna, time or scan and
407
+ `incremental=True` are refused. Measured against CASA 6.7.6 `fluxscale`: the same flux densities,
408
+ errors and F tables as CASA on CASA's own G tables (fit errors at `fitorder=2` within 2.6e-5), and
409
+ within 4e-4 end to end.
410
+ - Changed: `rang.cal_gain.gaincal` takes `spw` to solve one window of a multi-window MS (tables are
411
+ pre-applied on single-window data only).
412
+ - Changed: `examples/a1_simtruth.py generate` takes `--seeds` and `--gcal-flux`, `--gcal-spix`,
413
+ `--gcal-reffreq` (the default versions are unchanged).
414
+
415
+ - New: `gemclean -channels-out N`, `-join-channels`, `-fit-spectral-pol N`, `-channel-range S E`,
416
+ `-mf-weighting`/`-no-mf-weighting` and rang's own `-average-channels N`, as WSClean 3.7 has
417
+ them: output channels grouped as WSClean groups them, MF weighting by default with joined
418
+ channels, a joined minor loop that picks on the weighted sum of the channel residuals and puts a
419
+ component in every channel (GeMClean's exact value per channel, then WSClean's ordinary polynomial
420
+ fit in `nu/nu_ref - 1`), and per-channel `<name>-0000-<kind>.fits` and weighted
421
+ `<name>-MFS-<kind>.fits` images. `-fit-spectral-log-pol`, `-deconvolution-channels` and squared
422
+ joining are refused. One output channel is byte-identical to before. `rang-core image` passes
423
+ `channels_out`, `join_channels`, `fit_spectral_pol` and `average_channels` to gemclean instead
424
+ of refusing them (`research/g3-spectral.md`).
425
+ - New: `rang.applycal.applycal(ms, gaintable, interp, gainfield, calwt, field)`, CASA's
426
+ `applycal` on a Measurement Set: row chunks through `rang.core_writeback` (a flag version first),
427
+ the Rust kernel per field and data description, CORRECTED_DATA (added as `clearcal` does when
428
+ absent), FLAG OR'd, and with `calwt` on a G, T or B table WEIGHT and WEIGHT_SPECTRUM reset from
429
+ SIGMA and scaled as CASA does (K, Df and general Jones terms leave weights alone). Field maps
430
+ `''`, a field name and `nearest` as CASA's. Measured against CASA 6.7.6 `applycal`
431
+ (`research/c4b-apply-ms.md`).
432
+ - Changed: the apply kernel (`rang.cal_apply`) writes cells under a one-sided diagonal solution flag
433
+ as CASA does (the good side still applied), tracks phases over the two slots it interpolates
434
+ between, and uses casacore's linear-interpolation formula; new `cal_apply.apply_jones` applies
435
+ a general 2×2 term on its own solution bins.
436
+ - New: `rang-core apply` applies the `complex` terms of `rang-core calibrate` to the fields they were
437
+ solved on, into CORRECTED_DATA; `rang-core calibrate` writes `setjy`, `manual` and `lsm` models to
438
+ MODEL_DATA through `rang.calmodels` and solves against them. Applying a complex term to another
439
+ field and polarised models are refused, naming C2a and C7a.
440
+
441
+ - Changed: `rang.cal_delay` solves K as QuartiCal 0.2.7's `delay_and_offset` by default (a delay
442
+ and a free phase per antenna and hand, `core/src/cal/delay_offset.rs`), with rang's offset start
443
+ (the FFT peak's phase, `offset_start=True`; `False` is QuartiCal's estimator exactly) and a delay
444
+ error with the offset marginalised; `kind="delay"` keeps C2b's term. It meets CASA's
445
+ `gaincal(gaintype='K')` with nothing pre-applied, as CARACal solves K. The `K Jones` table holds
446
+ the delays only; the offsets are in `DelaySolution.offsets` and G absorbs them
447
+ (`research/c2c-bandpass.md`).
448
+ - New: `rang.cal_bandpass`, the bandpass B: `diag_complex` per channel in Rust with CASA's rules per
449
+ channel (`minblperant`, `minsnr`, the reference antenna) and CASA's `fillgaps`
450
+ (`core/src/cal/bandpass.rs`), `bandpass(ms, ...)` as CASA's `bandpass(bandtype='B')` (defaults
451
+ `combine='scan'`, `fillgaps=0`), writing CASA `B Jones` tables. `rang.cal_chain.primary` solves
452
+ the primary calibrator's K, G, B in CARACal's order, each with the earlier tables applied.
453
+ `rang.simtruth` has a `holdout256` version (new seeds, 256 channels) (`research/c2c-bandpass.md`).
454
+ - New: `rang.cal_delay`, the delay term K in Rust (`core/src/cal/delay.rs`): QuartiCal 0.2.7's
455
+ `delay` kernel with its FFT initial estimate (FFT, then Gauss-Newton over all baselines).
456
+ `solve(...)` on arrays; `gaincal(ms, ...)` as CASA's `gaincal(gaintype='K')`, with CASA's
457
+ solution intervals, pre-applied tables through `rang.cal_apply` (SIGMA weights then, as
458
+ `rang.cal_gain`) and CASA's K referencing (to the first refant, no flex or strict, no `minsnr`
459
+ or `minblperant`, which CASA's K does not apply); `to_caltable` writes a CASA `K Jones` table
460
+ (float32 delays in ns, the one-channel window at the band centre), which `rang.cal_apply`
461
+ applies. It matches CASA only with a G pre-applied; without one, antenna phases bias it
462
+ (`research/c2b-delay.md` section 6; the K model is an owner decision). `rang.simtruth` has a
463
+ `delay50` version (delays of 1-50 ns, 256 channels) (`research/c2b-delay.md`).
464
+ - Changed: `gemclean` needs far less memory, with byte-identical images (task G2,
465
+ `research/g2-memory.md`). It reads the Measurement Set at the exact size with Stokes I in single
466
+ precision, compacts in place, keeps no copy of the gridding weights, holds single-precision
467
+ visibilities as read, forms each residual in the gridder's sorted values instead of a
468
+ visibility-sized buffer, forms the 2n PSF before the plan of the image grid and frees it after
469
+ the beam is built. Peak RSS on the ESO 137-006 pilot at 512 channels went from more than 24 GB
470
+ (about 32 GB estimated) to 19.7 GB; at 256 channels from 18.4 to 12.2 GB; on 2048² simulated
471
+ skies from 3.46 to 2.99 GB.
472
+ - New: `gemclean -memory-report` prints the heap held per phase of a run (read, weights, PSF,
473
+ beam, each major cycle) and the main buffers' sizes; `-report.json` carries the same under
474
+ `memory` (`research/g2-memory.md`). Images are unchanged.
475
+ - New: `rang.calmodels`, calibrator models written to `MODEL_DATA` as CASA's `setjy` writes them
476
+ (C1 term T0). `set_model(ms, field, spec)` takes a flux-density standard of `rang.fluxstd`, a
477
+ `manual` model (flux density, reference frequency, log10 spectral terms), per-channel Stokes rows
478
+ (`PointStokes`), or a component list such as CARACal's `pks1934-638.lsm`, predicted through
479
+ `core.predict_components`. It writes through `rang.core_writeback` and leaves other fields'
480
+ rows as they are. Polarised `manual` models are refused until C7a (`research/c3a-models.md`).
481
+ - New: commands for your own Measurement Set, `rang-core flag`, `calibrate`, `apply`, `image`,
482
+ `selfcal` and `run CONFIG MS`, driven by one YAML configuration (schema `rang-config`, version
483
+ 1; `rang-core config defaults` prints every key with its default, `rang-core config check`
484
+ validates). `flag` runs the Tricolour port per field and scan and writes FLAG; `calibrate`
485
+ solves the `complex` term and writes a QuartiCal gain store; `image` runs `gemclean`. The rest
486
+ of CARACal's chain (K, G, B, F, polarisation terms, apply, self-calibration, multi-channel and
487
+ polarised imaging) can be written in the configuration and is refused with the task that adds
488
+ it. A run directory (`<ms>.rang-run`) keeps the state, so `run` skips finished stages and
489
+ resumes at a failed one. Flags are saved as the CASA flag version `rang_original` before the
490
+ first write, and as `rang_<stage>_<time>` before each stage writes (`research/w2-commands.md`).
491
+ - Changed: the research stages moved under `rang-core research` (for example `rang-core research
492
+ flags compare`); the old top-level forms now say where they went. PyYAML is a dependency.
493
+ - New: `rang.cal_gain`, the gain term G in Rust (`core/src/cal/diag.rs`, `reference.rs`):
494
+ diagonal complex (`calmode='ap'`) and phase-only (`'p'`) solves ported from QuartiCal 0.2.7's
495
+ `diag_complex` and `phase` kernels, with CASA's solution intervals (`solint`, `combine`),
496
+ `uvrange` (metres), `minblperant`, `minsnr` and reference antenna (`flex`, `strict`); earlier
497
+ tables are pre-applied through `rang.cal_apply`, and the result is a CASA `G Jones` table
498
+ (`research/c2a-gain.md`).
499
+ - `tools/check-all.sh` has a new last step, `gain-hashes` (`tools/gain_hash_gate.py`): it recomputes
500
+ the archived `complex` gains (`4ed2aa89…`, `7b370c13…`) and core phase 2's LUNO convergence, and
501
+ fails on any change. The archived inputs are found in `$RANG_ARCHIVE_OUTPUTS`, `outputs/` or
502
+ `~/Github/rang/outputs`; `RANG_GAIN_HASHES=optional` skips the step on a host without them.
503
+
504
+ - New: `rang.core_writeback`, write-back to an existing Measurement Set. `WriteBack` adds
505
+ `MODEL_DATA`, `CORRECTED_DATA` (or any data column) and `WEIGHT_SPECTRUM` as CASA's `clearcal`
506
+ adds them, writes any column, `FLAG` and `FLAG_ROW` in row chunks, saves the original flags as a
507
+ CASA flag version (`rang_original`) before the first write, and refuses to start after an
508
+ interrupted session until the flags are restored (`restore_flags`). `rang.formats.flagversions`
509
+ now saves in chunks, atomically, and gains `restore`. CASA, WSClean, python-casacore and dask-ms
510
+ read the written columns (`research/w1-writeback.md`).
511
+ - The core table writer now writes a tiled column's header through a temporary file and a rename,
512
+ so a process killed during a flush cannot leave it truncated. The bytes written are unchanged.
513
+
514
+ - New: `rang.simtruth` and `examples/a1_simtruth.py`, a simulated MeerKAT L-band observation in
515
+ full polarisation with known truth. It has K, B, G, KCROSS, Xf, linear Df and parallactic
516
+ rotation, a self-calibration residual, RFI with bit masks, and thermal noise of known σ, and is
517
+ written as a Measurement Set through rang's table writer. There are two versions: small (16
518
+ antennas, 64 channels, 41 MB) and full (64 antennas, 256 channels, 35.8 GB). CASA
519
+ `applycal` with tables made from the truth inverts it within class C
520
+ (`research/a1-simtruth.md`).
521
+ - New: `rang.cal_apply.apply`, a Rust kernel (`rang._core.cal_apply`) that corrects or corrupts
522
+ visibilities in memory by CASA K, G, T, B and Df tables, with `applycal`'s `linear` and `nearest`
523
+ interpolation in time and frequency and its `calflag` flags, including how Df flags and
524
+ substitutes bad leakages (`research/c4a-apply.md`). KCROSS, Xf and the parallactic angle are
525
+ refused until their steps. Weights are not changed and nothing is written to a Measurement Set yet.
526
+ - New: `gemclean -beam-band`, off by default. Resyncs use the beam's spectrum only as far as the
527
+ longest baseline and the w-term reach. Images differ slightly from flag off. Over 12 interleaved
528
+ pairs with coarse gridding (`-single -wgridder-accuracy 1e-4`) it saved 11-16 % of the wall on
529
+ two simulated skies and 4-5 % on a third (`research/gemclean-speed.md`).
530
+ - The root crate and binary are now `rang-sim`, `rang.toy_output` is now `rang.sim_output`,
531
+ `RANG_TOY_BIN` and `RANG_TOY_LIB` are now `RANG_SIM_BIN` and `RANG_SIM_LIB`, and
532
+ `examples/toy.py` is now `examples/pointing_sim.py`. No aliases are kept. Output bytes are
533
+ unchanged. Receipts are `rang.receipt` version 4.
534
+ - `gemclean` takes w-planes in groups of 4, within 2 GiB of grids. Results are unchanged; peak
535
+ memory rises (2.84 to 3.38 GB on one simulated job). Other callers of the exact gridder stay at
536
+ one plane, because larger groups were measured only on gemclean. `RANG_WGRIDDER_PLANE_GROUP`
537
+ overrides either default.
538
+ - New: the reference imager `gemclean` and its binary, and the flagged Gram operator. Their
539
+ speed figures are provisional.
540
+ - New: `gemclean -stop-rule certified`, a minor-cycle stop rule with Lean-proved consequences
541
+ (`research/stop-rule-theorem.md`). It costs about three times the default's major cycles and a
542
+ table of about 100 gridder calls, so it is not the default, and the default's images are unchanged.
543
+
544
+ ## 0.1.0 (2026-11-02)
545
+
546
+ First release. A research demonstrator for MeerKAT calibration and
547
+ flagging, not an operational calibrator; no advantage over the state of
548
+ the art is claimed.
549
+
550
+ **The Rust core.** The calibration chain of the gain-TV flagging
551
+ experiment runs in one Rust core, reached from Python through the PyO3
552
+ extension `rang._core`: fused visibility statistics, a QuartiCal-style
553
+ gain solver, the component predictor, blosc decoding of the public zarr
554
+ stores with HTTP range reads, the Tricolour strategies and SumThreshold,
555
+ WSClean imaging weights, a w-stacking gridder and the exact wgridder, a
556
+ read-only Measurement Set reader and a casacore table writer that creates
557
+ Measurement Sets and CASA calibration tables byte-identical to casacore
558
+ 3.8.1 and CASA 6.7.6 (bar uninitialised memory). Every port is checked against its upstream oracle
559
+ (QuartiCal, Tricolour, AOFlagger, WSClean, python-casacore, casacure) or
560
+ against the archived results of the frozen campaigns.
561
+
562
+ **Simplification and speed.** The solver, the fused pass, SumThreshold and
563
+ the gridder run in parallel; the w-stacking gridder and FFT also run on the
564
+ GPU through `wgpu` behind the `gpu` feature. The ctypes backend is gone:
565
+ importing `rang` or calling a kernel never runs Cargo, and the extension
566
+ releases the GIL around every kernel. One command line, `rang-core STAGE
567
+ VERB`, writes a `rang.receipt` per call and a `rang.archive` per run. The
568
+ frozen QuartiCal and Stimela campaigns, their containers and modules moved
569
+ to `history/`; the package keeps verbatim extracts of what the core uses.
570
+ The package was renamed from `rangtoy` to `rang` (distribution
571
+ `rang-core`). `rang-core --version` and `rang.__version__` report the
572
+ version.
573
+
574
+ **Wheels and CI.** One abi3 wheel per platform for CPython 3.10 and newer:
575
+ Linux x86_64 and aarch64 (manylinux_2_28), macOS arm64 and x86_64, and
576
+ Windows x86_64, each tested on CPython 3.10 and 3.14 on its own OS with
577
+ Cargo absent, with casacure as a second oracle of the table reader and
578
+ writer. A `v*` tag that equals the `pyproject.toml` version attaches the
579
+ wheels and the sdist to a GitHub release and publishes them to PyPI by
580
+ trusted publishing. The sdist holds only the crates, the Python package
581
+ and the tests the wheel runs.
582
+
583
+ **Documentation.** A Sphinx site (<https://pythonic.nl/rang-core/>)
584
+ with installation, concepts, the command line, the API reference and the
585
+ research story; five executed tutorial notebooks (data and rank, flagging
586
+ step by step, gains and TV, imaging, the flagging experiment) and two
587
+ science notebooks (the gain-TV verdict; beam and pointing).
588
+
589
+ **Science verdicts.** The gain-TV-guided flagging experiment is closed as
590
+ null: on LUNO 1934−638 at four stations and across the whole array, on
591
+ PKS 1830−211 under a stored and a scalar model, under CARACal's field model
592
+ and with a closure-invariant arm, 90 pre-registered calls gave 64 null and
593
+ 26 undecidable, none better or worse, and the method costs 4.6–5.9× the
594
+ baseline. The solver study changed no default (the QuartiCal issue 452
595
+ weight fix is opt-in through `RANG_SOLVE_OPTIONS`). For beam and pointing,
596
+ the SARAO holography beam is the reference and the SARAO/SKA Gaussian
597
+ pointing fit is ported bit-exact; the identifiability survey finds
598
+ per-antenna pointing and squint constrainable on the calibrator and ESO137
599
+ data, array-common pointing and ellipticity on the ESO137 track, and beam
600
+ width nowhere. Pointing self-calibration against those baselines is the
601
+ next step. Negative results are kept in `research/`.
602
+ This covers a gain-TV-guided threshold policy; a flagger trained on gain TV
603
+ is a separate question, not yet tested.
604
+
605
+ **Provenance and the wheel (wave 4 fixes).** Receipts (`rang.receipt`
606
+ version 3) record every result-affecting `RANG_*` variable from one registry
607
+ and the package source; the solver takes its options as an argument; `commit`
608
+ is null outside a checkout (W4-1). The closure and flux-standard kernels run
609
+ in the extension, so no kernel call builds with Cargo and the wheel is
610
+ self-contained (W4-3).
@@ -8,5 +8,5 @@ authors:
8
8
  repository-code: "https://github.com/gijzelaerr/rang-core"
9
9
  url: "https://pythonic.nl/rang-core/"
10
10
  license: GPL-3.0-or-later
11
- version: 0.1.0
12
- abstract: "MeerKAT calibration, flagging and imaging research on one Rust core, driven from Python."
11
+ version: 0.2.0
12
+ abstract: "ra-ng: agentic radio astronomy, MeerKAT first. rang-core is its Rust core and Python command line for flagging, calibration and imaging."