report-workflow 4.41.0__tar.gz → 4.41.2__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 (226) hide show
  1. {report_workflow-4.41.0 → report_workflow-4.41.2}/CHANGELOG.md +139 -0
  2. {report_workflow-4.41.0 → report_workflow-4.41.2}/PKG-INFO +40 -20
  3. {report_workflow-4.41.0 → report_workflow-4.41.2}/README.md +39 -19
  4. {report_workflow-4.41.0 → report_workflow-4.41.2}/pyproject.toml +1 -1
  5. {report_workflow-4.41.0 → report_workflow-4.41.2}/skills/report-workflow/SKILL.md +14 -6
  6. {report_workflow-4.41.0 → report_workflow-4.41.2}/skills/report-workflow/reference/authoring.md +42 -10
  7. {report_workflow-4.41.0 → report_workflow-4.41.2}/skills/report-workflow/reference/setup-and-preflight.md +22 -11
  8. {report_workflow-4.41.0 → report_workflow-4.41.2}/skills/report-workflow/reference/tools.md +8 -6
  9. {report_workflow-4.41.0 → report_workflow-4.41.2}/skills/report-workflow/skill.yaml +30 -13
  10. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/__init__.py +1 -1
  11. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/agent_wrapper.py +168 -27
  12. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/artifact_contract.py +96 -2
  13. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/artifact_lint.py +78 -7
  14. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/artifact_packaging/final_qa.py +7 -0
  15. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/automation_harness.py +146 -11
  16. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/blueprints/proposal.yaml +12 -2
  17. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/cli.py +7 -1
  18. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/derived_evidence.py +158 -11
  19. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/draft_check.py +13 -5
  20. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/engineering_audit.py +7 -1
  21. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/mcp_server.py +3 -2
  22. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/abstract_check.py +15 -3
  23. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/agent_tasks.py +176 -109
  24. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/base_document_diff.py +16 -1
  25. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/blueprint_plan.py +18 -0
  26. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/citation_bind.py +49 -2
  27. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/cited_sources.py +71 -2
  28. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/claim_plan.py +75 -0
  29. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/docx_render.py +214 -11
  30. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/factuality_check.py +384 -33
  31. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/figure_build.py +60 -1
  32. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/figure_plan_audit.py +21 -0
  33. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/figure_quality.py +69 -7
  34. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/figure_recommend.py +408 -14
  35. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/front_matter_build.py +26 -16
  36. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/heading_contract_check.py +28 -4
  37. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/outline_plan.py +68 -2
  38. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/paper_scope_freeze.py +14 -1
  39. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/post_render_validate.py +6 -3
  40. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/qa_gate.py +4 -2
  41. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/section_contract.py +4 -3
  42. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/section_draft.py +27 -2
  43. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/source_tables.py +110 -7
  44. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/preflight_decisions.py +51 -3
  45. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/prompt_questions.py +7 -2
  46. report_workflow-4.41.2/src/report_workflow/quote_gate.py +90 -0
  47. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/reference_match.py +111 -8
  48. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/source_citation.py +18 -2
  49. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/state.py +8 -4
  50. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow.egg-info/PKG-INFO +40 -20
  51. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow.egg-info/SOURCES.txt +7 -0
  52. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_automation_harness.py +81 -6
  53. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_calculation_record.py +2 -2
  54. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_check_draft.py +1 -0
  55. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_chinese_quantity_expressions.py +23 -0
  56. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_declared_source_citations.py +35 -4
  57. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_delivered_attributions.py +9 -5
  58. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_delivered_layout.py +12 -0
  59. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_document_language.py +9 -9
  60. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_evidence_traceability.py +89 -44
  61. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_figure_recommend.py +128 -0
  62. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_format_standards.py +111 -1
  63. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_formula_derivations.py +19 -0
  64. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_hallucination_gate.py +47 -0
  65. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_intake_summary_measures.py +15 -0
  66. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_mvp_workflow.py +50 -1
  67. report_workflow-4.41.2/tests/test_quote_gate.py +45 -0
  68. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_raw_data_reports.py +125 -0
  69. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_reference_match.py +92 -0
  70. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_roadmap_contracts.py +16 -8
  71. report_workflow-4.41.2/tests/test_small_issues_2026_11_17.py +164 -0
  72. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_structured_draft_paragraphs.py +45 -0
  73. report_workflow-4.41.2/tests/test_usability_audit_2026_10_19.py +393 -0
  74. report_workflow-4.41.2/tests/test_usability_consolidated_2026_10_21.py +172 -0
  75. report_workflow-4.41.2/tests/test_usability_review_2026_10_20.py +783 -0
  76. report_workflow-4.41.2/tests/test_usability_round3_2026_10_20.py +459 -0
  77. {report_workflow-4.41.0 → report_workflow-4.41.2}/.env.example +0 -0
  78. {report_workflow-4.41.0 → report_workflow-4.41.2}/LICENSE +0 -0
  79. {report_workflow-4.41.0 → report_workflow-4.41.2}/MANIFEST.in +0 -0
  80. {report_workflow-4.41.0 → report_workflow-4.41.2}/requirements.txt +0 -0
  81. {report_workflow-4.41.0 → report_workflow-4.41.2}/setup.cfg +0 -0
  82. {report_workflow-4.41.0 → report_workflow-4.41.2}/skills/report-workflow/reference/benchmarking.md +0 -0
  83. {report_workflow-4.41.0 → report_workflow-4.41.2}/skills/report-workflow/reference/engineering-lab.md +0 -0
  84. {report_workflow-4.41.0 → report_workflow-4.41.2}/skills/report-workflow/reference/figures.md +0 -0
  85. {report_workflow-4.41.0 → report_workflow-4.41.2}/skills/report-workflow/reference/profiles.md +0 -0
  86. {report_workflow-4.41.0 → report_workflow-4.41.2}/skills/report-workflow/reference/revision.md +0 -0
  87. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/__main__.py +0 -0
  88. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/artifact_packaging/__init__.py +0 -0
  89. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/artifact_packaging/common.py +0 -0
  90. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/artifact_packaging/template_reports.py +0 -0
  91. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/blueprints/__init__.py +0 -0
  92. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/blueprints/academic_paper.yaml +0 -0
  93. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/blueprints/admissions_project_report.yaml +0 -0
  94. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/blueprints/admissions_report.yaml +0 -0
  95. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/blueprints/business_report.yaml +0 -0
  96. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/blueprints/custom.yaml +0 -0
  97. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/blueprints/engineering_lab_report.yaml +0 -0
  98. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/blueprints/quant_research_memo.yaml +0 -0
  99. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/config.py +0 -0
  100. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/configs/__init__.py +0 -0
  101. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/configs/banned_phrases.json +0 -0
  102. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/configs/governance_policy.json +0 -0
  103. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/configs/guideline_rules.json +0 -0
  104. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/configs/style_profiles/academic.json +0 -0
  105. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/configs/style_profiles/hybrid.json +0 -0
  106. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/configs/style_profiles/work.json +0 -0
  107. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/configs/style_vocab.json +0 -0
  108. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/connectors/__init__.py +0 -0
  109. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/connectors/notebooklm_connector.py +0 -0
  110. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/connectors/research_backends.py +0 -0
  111. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/errors.py +0 -0
  112. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/guidelines/CHINESE_ENGINEERING.json +0 -0
  113. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/guidelines/CONSORT.json +0 -0
  114. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/guidelines/PRISMA.json +0 -0
  115. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/guidelines/SRQR.json +0 -0
  116. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/guidelines/STARD.json +0 -0
  117. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/guidelines/STROBE.json +0 -0
  118. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/guidelines/TRIPOD.json +0 -0
  119. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/language.py +0 -0
  120. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/__init__.py +0 -0
  121. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/admissions_monograph_polish.py +0 -0
  122. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/admissions_tone_gate.py +0 -0
  123. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/artifacts.py +0 -0
  124. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/base_document_parse.py +0 -0
  125. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/claim_verify_execute.py +0 -0
  126. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/corpus_build.py +0 -0
  127. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/derived_evidence.py +0 -0
  128. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/evidence_normalize.py +0 -0
  129. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/evidence_store.py +0 -0
  130. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/figure_types.py +0 -0
  131. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/figure_utils.py +0 -0
  132. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/final_publish.py +0 -0
  133. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/guideline_select.py +0 -0
  134. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/heading_dedup.py +0 -0
  135. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/intake.py +0 -0
  136. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/merge_draft.py +0 -0
  137. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/methods_protocol_build.py +0 -0
  138. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/notebook_sync.py +0 -0
  139. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/post_render_repair.py +0 -0
  140. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/project_identity_gate.py +0 -0
  141. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/publication_naturalness_pass.py +0 -0
  142. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/reference_reality_check.py +0 -0
  143. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/reference_relevance_gate.py +0 -0
  144. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/reference_verify.py +0 -0
  145. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/register_gate.py +0 -0
  146. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/remediation_router.py +0 -0
  147. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/research_execute.py +0 -0
  148. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/revision_apply.py +0 -0
  149. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/section_plan_freeze.py +0 -0
  150. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/section_role_check.py +0 -0
  151. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/source_appendix_render.py +0 -0
  152. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/source_parse.py +0 -0
  153. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/style_pass.py +0 -0
  154. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/supplementary_package_build.py +0 -0
  155. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/table_transform.py +0 -0
  156. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/nodes/visual_render_check.py +0 -0
  157. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/parsers/__init__.py +0 -0
  158. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/parsers/agent_fallback.py +0 -0
  159. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/parsers/code_parser.py +0 -0
  160. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/parsers/office_math.py +0 -0
  161. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/parsers/semi_structured_parser.py +0 -0
  162. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/parsers/source_text.py +0 -0
  163. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/parsers/structured_parser.py +0 -0
  164. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/policies/__init__.py +0 -0
  165. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/policies/policy_pack.py +0 -0
  166. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/preflight.py +0 -0
  167. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/profiles.py +0 -0
  168. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/prompts/__init__.py +0 -0
  169. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/prompts/intake_prompt.py +0 -0
  170. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/reference_list.py +0 -0
  171. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/run_workflow.py +0 -0
  172. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/runtime_support.py +0 -0
  173. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/source_selection.py +0 -0
  174. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/templates/reference.docx +0 -0
  175. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/validators/__init__.py +0 -0
  176. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/validators/parse_validator.py +0 -0
  177. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow/verify.py +0 -0
  178. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow.egg-info/dependency_links.txt +0 -0
  179. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow.egg-info/entry_points.txt +0 -0
  180. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow.egg-info/requires.txt +0 -0
  181. {report_workflow-4.41.0 → report_workflow-4.41.2}/src/report_workflow.egg-info/top_level.txt +0 -0
  182. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_adversarial_benchmark.py +0 -0
  183. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_append_only_ledger_contract.py +0 -0
  184. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_benchmark_contracts.py +0 -0
  185. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_cited_sources.py +0 -0
  186. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_column_roles.py +0 -0
  187. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_decimal_comma_evidence.py +0 -0
  188. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_declared_citation_propagation.py +0 -0
  189. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_delivery_contracts.py +0 -0
  190. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_docx_render_references.py +0 -0
  191. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_emission_intensity_units.py +0 -0
  192. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_external_benchmark.py +0 -0
  193. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_fe_names_the_span_it_read.py +0 -0
  194. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_figure_recommendation_loss.py +0 -0
  195. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_figure_value_support.py +0 -0
  196. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_form_feed_line_numbers.py +0 -0
  197. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_grouped_derived_evidence.py +0 -0
  198. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_improvements.py +0 -0
  199. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_line_chart_markers.py +0 -0
  200. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_long_block_evidence.py +0 -0
  201. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_mcp_server.py +0 -0
  202. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_measurement_and_union_rule.py +0 -0
  203. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_mermaid.py +0 -0
  204. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_naturalness_machine_tells.py +0 -0
  205. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_outline_brief_lists_required_sections.py +0 -0
  206. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_outline_obligations.py +0 -0
  207. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_pdf_inline_scripts.py +0 -0
  208. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_pdf_table_cells.py +0 -0
  209. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_quant_research_memo_profile.py +0 -0
  210. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_reference_list.py +0 -0
  211. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_register_gate.py +0 -0
  212. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_report_axes.py +0 -0
  213. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_report_quality_benchmark.py +0 -0
  214. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_revision_change_types.py +0 -0
  215. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_revision_diff.py +0 -0
  216. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_scientific_notation_evidence.py +0 -0
  217. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_source_selection.py +0 -0
  218. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_state_checkpoint.py +0 -0
  219. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_structured_source_tables.py +0 -0
  220. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_table_display_precision.py +0 -0
  221. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_trend_claims.py +0 -0
  222. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_verify_adapter.py +0 -0
  223. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_version_sync.py +0 -0
  224. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_visual_review_status.py +0 -0
  225. {report_workflow-4.41.0 → report_workflow-4.41.2}/tests/test_wide_table_landscape.py +0 -0
  226. {report_workflow-4.41.0 → report_workflow-4.41.2}/workflow_config.yaml +0 -0
@@ -1,5 +1,144 @@
1
1
  # Changelog
2
2
 
3
+ ## 4.41.2 - 2026-11-18
4
+
5
+ Six small fixes found by writing reports end to end; no gate, profile or public
6
+ option changes.
7
+
8
+ - FE no longer reads figure/table labels (圖 2, 表 1, Figure 2) or 筆電 as quantities.
9
+ - A complete two-parameter grid (one row per cell) is recommended as a heatmap, and
10
+ the build-time density check uses the recommender's own 16 x 12 ceiling.
11
+ - A Chinese document numbers first-level headings 一、二、三、 instead of 1. 2. 3.
12
+ - A pipe table the author typed gets a numbered caption, 表 N. 標題.
13
+ - Test count 1490.
14
+
15
+ ## 4.41.1 - 2026-10-22
16
+
17
+ Three usability passes on the same question — what stops an author between
18
+ `start_report` and a document they can hand in — consolidated into one change,
19
+ then checked by writing three new reports end to end on the result: a
20
+ quantitative research memo (Chinese), an engineering lab report (Chinese) and a
21
+ business report (English). On 4.41.0, on the head of the first two passes and
22
+ on the head of the third, none of the three publishes from the same authored
23
+ files; here all three publish on the first pass with no refusal. Nothing here
24
+ adds a gate, a profile or a public option, and the adversarial benchmark's
25
+ verdicts are unchanged. Run record:
26
+ `benchmarks/evidence/usability_consolidated_2026-10-21/`.
27
+
28
+ ### Fixed — what stopped a report
29
+
30
+ - A numeric column named `units` marked a file mixed-unit and dropped every
31
+ numeric summary. 「每個擺長重複 5 次」 read as the rate 5 次/個, 「每月底依過去
32
+ 6 個月」 as 6 個月/月, and `7x24` lost its 24; English function words ("and")
33
+ were read as units. Folding `7x24` now keeps every character offset, so
34
+ `check-draft` finds the same figures it did before.
35
+ - FE misread a backtest's and a lab sheet's own vocabulary. 「樣本內」/「樣本外」
36
+ read as dataset scope; 「12 個月」 against `lookback_months` as a different
37
+ unit; 「回看 12 個月」 as not naming 回看月數, so the 12 had to come from the
38
+ Sharpe column; 「負重 50 g」 as not naming 砝碼質量_g; 「前四組」 as a count of
39
+ 4; "resolved within 24 hours" as a 24 the row does not state, although the
40
+ column is `resolved_within_24h_pct`; a claim reporting the research question
41
+ as answered by nothing. Each still blocks its false twin: 年化報酬 10.8%
42
+ against a volatility of 10.8, 平均伸長 9.9 cm against the second trial's
43
+ 9.9 cm, 48 hours, 24%, 實驗分三組.
44
+ - 「每月…20%」 against evidence that itself says 每月 read as 20%/月, and a
45
+ paraphrase of a statistic whose figures all checked failed on vocabulary.
46
+ - A failure routed back to an earlier stage kept the old baseline, so work
47
+ already written became a `scope_violation`; publish re-stamping `_contract`
48
+ was taken for the author editing a file. The repair scope spans the rewound
49
+ stages, editing an accepted artifact re-opens its stage, and the harness
50
+ hashes artifacts without the stamp.
51
+ - `lint_artifacts` passed claim matrices that submission and FA refuse, and
52
+ the claim stage now refuses them too; revise_existing's pre-check refused
53
+ the documented `retitle`/`remove_section`, and FA ignored revision-plan
54
+ claim links.
55
+ - The structured-drafts brief lists the sections to write without `cover`,
56
+ and SECTION_DRAFT and lint refused the draft for a missing `cover`.
57
+ - academic_paper: the thesis is checked at the outline stage, missing author
58
+ metadata is named at start and can be supplied in `outline.json`, the
59
+ abstract labels the brief shows pass, keywords never come from another
60
+ project, a sentence may open with a spelled-out number.
61
+ - A derivation that failed to register stayed in `derived_evidence.json` and
62
+ later stopped publish; a band printed `7.50–9` refused the author's `7.5-9`;
63
+ ratio and formula can read the cross tables intake built; a claim citing an
64
+ unregistered `E_D_` id is told so.
65
+ - `remap_agent_artifacts` finds an earlier run in another workspace and copies
66
+ the files the current stage may write.
67
+ - `check_environment`'s `required_preflight_decisions` lists only what needs
68
+ an answer; optional tools left out are skipped instead of refusing the start.
69
+
70
+ ### Fixed — what made the document worse than its sources
71
+
72
+ - Every sentence restating the author's own files ended 「(本研究整理,計算附件
73
+ C4)」/"(compiled in this study, …)", pointing at a file the reader was not
74
+ sent and, after an image, costing the figure its caption. Only computed
75
+ figures carry a label now.
76
+ - Placed tables printed raw headers (「年化報酬_pct 平均」, `review_count`);
77
+ chart titles did too. Both use the reader's words; the evidence keeps the
78
+ column names, so no id moves. A formula label naming its unit printed it
79
+ twice (「拉力 (N) (N)」).
80
+ - An empty cover section after the front matter printed a 「封面」 chapter.
81
+ - Starter figures: unplaced ones are neither demanded nor built; a computed
82
+ table's counts, shares and total row are not stacked; a chart with no values
83
+ is stopped; two or three crossed label columns (lookback × sample ×
84
+ portfolio) are a grid with one series per value, not an axis reading 6, 6, 6,
85
+ 6, 12, …; a lab sheet plots measured against theory, or the mean of its
86
+ trials against the setting; Chinese renders on Linux.
87
+ - Intake statistics that restated rows are gone: quartiles of two values,
88
+ one-row-per-group tables the outline then had to waive, the mean of a row
89
+ counter or of a design setting — including a setting crossed with two label
90
+ columns, which took the Sharpe ratio's slot in both memo tables — and a
91
+ "joinable" pair meeting on one median. An empty open band above the last
92
+ bucket edge is not a row.
93
+ - Briefs carry their own profile's guidance only (a business report's
94
+ drafting brief 29 → 20 KB); `paper_spine`/`lab_spine` are no longer asked
95
+ for (a `paper_spine.contribution` already written still counts as the
96
+ thesis), and `results_mode` is shown only where it is read.
97
+ The questions a task asks start after the request's colon and lose a leading
98
+ "and". Heading numbers count the sections the document has.
99
+ - Run directories use hyphens and no commas, so unquoted paths work.
100
+ - `examples/source_to_report.py`, the README's first run, works again and CI
101
+ runs it.
102
+
103
+ ### Fixed — `check-draft` names the unit when only the unit is wrong
104
+
105
+ A figure that is not found names the nearest figure printed beside the
106
+ sentence's words. When the page prints the same number in another unit, that is
107
+ now the one it names: "3.3GB" gets `p. 2 states 3.3TB there`, where it used to
108
+ get the 2.45 in the next column. On the planted drafts in
109
+ `benchmarks/check_draft/`, the hint names the original figure for 19 of the 34
110
+ planted figures, up from 12; all seven wrong units are among them. Verdicts do
111
+ not change (`scripts/run_draft_benchmark.py --check` passes).
112
+
113
+ ### Docs — what you need to install, said once and correctly
114
+
115
+ - README: it says up front that it runs inside an AI agent; install uv first;
116
+ then two lines in Claude Code, one `codex mcp add` line in Codex, and the
117
+ same `uvx` command for Cursor or any other MCP client (JSON in
118
+ `docs/mcp.md`). The first start downloads about 1 GB, longer than an agent
119
+ waits for a new server, so the README and `docs/mcp.md` have you download it
120
+ once in a terminal first.
121
+ - `docs/mcp.md`: the server runs the whole pipeline, not only the gates; all
122
+ fifteen tools listed in the order a run calls them; register commands for
123
+ Claude Code, Codex and Cursor.
124
+ - `SKILL.md` and `reference/setup-and-preflight.md`: through the plugin or the
125
+ MCP server there is nothing else to install; from a clone, Python 3.11+ and
126
+ `pip install -e ".[render]"`. The quickstart notebook's last cell installs the
127
+ plugin instead of copying a folder that no longer exists.
128
+ - README figures checked against their archives: 46 attempts to sneak a false
129
+ claim through and 27 true claims (not 69 attempts), the three blind judges
130
+ are one model in three fresh contexts, and a 0–0 axis reads "tied, not won".
131
+
132
+ ### Not taken from the three passes
133
+
134
+ - `pearson_r` and `linear_fit` (#2): a new computation, not a repair; AGENTS.md
135
+ "Scope" puts that to the owner.
136
+ - #3's preflight change: #1 had already made optional installs need no answer,
137
+ and moved them to `optional_preflight_decisions`.
138
+ - The recorded runs under `benchmarks/evidence/usability_*` (about a hundred
139
+ logs and documents, several printing the recording machine's paths); the
140
+ pull requests keep their findings.
141
+
3
142
  ## 4.41.0 - 2026-10-21
4
143
 
5
144
  `check-draft`: the second check for a report that is already written. Where
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: report-workflow
3
- Version: 4.41.0
3
+ Version: 4.41.2
4
4
  Summary: Turn sources into a submission-ready report, with deterministic gates that block any claim not traceable to the evidence.
5
5
  Author-email: 0Smallcat0 <noiroao@gmail.com>
6
6
  License-Expression: MIT
@@ -53,7 +53,7 @@ mcp-name: io.github.0Smallcat0/report-workflow
53
53
 
54
54
  [![CI](https://github.com/0Smallcat0/report-workflow/actions/workflows/ci.yml/badge.svg)](https://github.com/0Smallcat0/report-workflow/actions/workflows/ci.yml)
55
55
  ![Python](https://img.shields.io/badge/python-3.11%2B-blue)
56
- ![Tests](https://img.shields.io/badge/tests-1305%20passing-brightgreen)
56
+ ![Tests](https://img.shields.io/badge/tests-1490%20passing-brightgreen)
57
57
  [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
58
58
 
59
59
  **繁體中文說明 → [README.zh-TW.md](README.zh-TW.md)**
@@ -72,7 +72,8 @@ whatever you already have. Plus one sentence, like "write a lab report on this".
72
72
  and charts drawn from your own numbers. It can follow your department's or your
73
73
  company's template. Chinese or English.
74
74
 
75
- Two lines to install. No API key. Nothing to configure.
75
+ It runs inside an AI agent you already use — Claude Code, Codex, Cursor. No API
76
+ key of its own, nothing to configure.
76
77
 
77
78
  ## Look before you install
78
79
 
@@ -93,19 +94,31 @@ repository, so you can see the output before you run anything:
93
94
 
94
95
  ## Use it
95
96
 
96
- In Claude Code:
97
+ It starts with [uv](https://docs.astral.sh/uv/getting-started/installation/),
98
+ so install that first. Then download it once in a terminal: about 1 GB of
99
+ Python packages and a pandoc binary, more than an agent waits for on a first
100
+ start (Codex gives a new server 10 seconds). Later starts take a few seconds.
101
+
102
+ ```bash
103
+ uvx --from "report-workflow[mcp,render]" report-workflow --help
104
+ ```
105
+
106
+ Then, in Claude Code:
97
107
 
98
108
  ```text
99
109
  /plugin marketplace add 0Smallcat0/report-workflow
110
+ /plugin install report-workflow@report-workflow
100
111
  ```
101
112
 
102
- Then `/plugin install report-workflow@report-workflow`. Using something else
103
- that speaks MCP (Codex, Cursor, your own setup)? One command instead:
113
+ In Codex:
104
114
 
105
115
  ```bash
106
- claude mcp add report-workflow -- uvx --from "report-workflow[mcp,render]" report-workflow-mcp
116
+ codex mcp add report-workflow -- uvx --from "report-workflow[mcp,render]" report-workflow-mcp
107
117
  ```
108
118
 
119
+ In Cursor or any other MCP client, register that same `uvx` command; the JSON
120
+ is in [docs/mcp.md](docs/mcp.md#register).
121
+
109
122
  Then just ask, in your own words:
110
123
 
111
124
  > Use report-workflow to turn the files in ./data into a business report for the
@@ -136,7 +149,9 @@ installed: `uvx report-workflow check-references draft.docx`.
136
149
 
137
150
  It reads `.docx`, `.md`, `.txt` and `.bib`. Each DOI is looked up at doi.org
138
151
  (which also carries retraction notices), each arXiv id at arXiv, and a reference
139
- with neither is searched for in Crossref and OpenAlex. No API key, no account,
152
+ with neither is searched for in Crossref and OpenAlex; when that settles
153
+ nothing and the reference links a web page (a statute, a course, a news
154
+ story), the page is opened and its own title looked for in the reference. No API key, no account,
140
155
  and no AI deciding anything: the same file gets the same verdicts.
141
156
 
142
157
  A five-reference draft with four errors planted in it:
@@ -170,7 +185,7 @@ references here:
170
185
 
171
186
  | Checker | Problem references caught (of 33) | Correct references flagged (of 71) |
172
187
  | --- | ---: | ---: |
173
- | **report-workflow** | **30** | **15** |
188
+ | **report-workflow** | **30** | **10** |
174
189
  | RefChecker | 32 | 36 |
175
190
  | CheckIfExist | 31 | 34 |
176
191
  | Hallucinator | 29 | 28 |
@@ -178,9 +193,10 @@ references here:
178
193
  | HalRef | 24 | 53 |
179
194
 
180
195
  "Flagged" counts anything short of accepted — for the five tools that includes
181
- "requires verification", here it is every verdict except `verified`. All 15 of
182
- those here are `needs_review`, and seven of them are standards, software, datasets,
183
- model cards and workshop pages that no bibliographic registry holds. Of the 13
196
+ "requires verification", here it is every verdict except `verified`. All 10 of
197
+ those here are `needs_review`; three are a standard, a software repository and a
198
+ model card that no bibliographic registry holds and whose pages title them
199
+ differently from the reference. Of the 13
184
200
  references it called `mismatch`, `not_found` or `retracted`, every one is wrong
185
201
  in the hand-checked labels. The three it passes cite the wrong venue or the
186
202
  wrong first name, which it does not compare.
@@ -198,7 +214,8 @@ set exists yet to test it on.
198
214
 
199
215
  The references can all be real and the sentence citing them still wrong: 76%
200
216
  where the paper says 67.0%, "not the cheapest to train" where it says "not the
201
- fastest". Give the command the files the draft cites:
217
+ fastest". Give the command the files the draft cites (`.pdf`, `.docx`, `.md`
218
+ or `.txt`):
202
219
 
203
220
  ```bash
204
221
  uvx report-workflow check-draft draft.docx --source llama.pdf --source qlora.pdf --appendix appendix.md
@@ -207,7 +224,9 @@ uvx report-workflow check-draft draft.docx --source llama.pdf --source qlora.pdf
207
224
  Only sentences that state a figure or quote are checked. Each figure must be
208
225
  printed on a page at the precision the draft gives, a few lines from the
209
226
  sentence's own words; each quotation must be on a page word for word. Every
210
- answer names the file, page and the source's own sentence:
227
+ answer names the file, page and the source's own sentence. The draft and PDFs
228
+ below are in [`benchmarks/check_draft/`](benchmarks/check_draft/), so you can
229
+ run it yourself:
211
230
 
212
231
  ```text
213
232
  [4] NOT_FOUND: In their words, "the preferred model is not the cheapest to train but the fastest at inference".
@@ -238,9 +257,9 @@ in the text really appear in your files. It cannot tell whether the AI
238
257
  understood your data. If the AI writes a smooth sentence that gets your result
239
258
  backwards, that sentence goes through.
240
259
 
241
- We measured that limit instead of talking around it: 69 hand-checked attempts to
242
- sneak something false past it, plus 10,000 test pairs from a public dataset
243
- nobody here wrote. **The cases it still misses are kept in the test set on
260
+ We measured that limit instead of talking around it: 46 hand-checked attempts to
261
+ sneak something false past it and 27 true claims it must let through, plus
262
+ 10,000 test pairs from a public dataset nobody here wrote. **The cases it still misses are kept in the test set on
244
263
  purpose** — [docs/EVIDENCE.md](docs/EVIDENCE.md).
245
264
 
246
265
  **It does not write.** Your AI writes; this decides what stays. So you need an
@@ -298,8 +317,9 @@ python scripts/run_drone_market_benchmark.py --check
298
317
  Three CSVs and one market question. Three recorded documents — a hand-written
299
318
  control, this pipeline's delivered output, and the strongest of four drafts by
300
319
  an AI given the same files. Three axes: counted (numeric), rule-scored
301
- (layout), and judged against a fixed rubric by three independent blind judges
302
- who are told nothing about how any arm was produced (argument). Every number
320
+ (layout), and judged against a fixed rubric by three blind judges — one model
321
+ in three fresh contexts — who are told nothing about how any arm was produced
322
+ (argument). Every number
303
323
  below is from
304
324
  [`benchmarks/evidence/drone_market_2026-09-16/`](benchmarks/evidence/drone_market_2026-09-16/summary.md).
305
325
  Check the date: the next re-recording moves these.
@@ -343,7 +363,7 @@ this benchmark. That is the largest known limitation of these numbers.
343
363
  | --- | --- | --- |
344
364
  | numeric — counted properties of the document | 3–2 won | 4–1 won |
345
365
  | layout — rule-scored structure and table furniture | 4–2 won | 4–2 won |
346
- | argument — three blind judges against a fixed rubric | **0–0 lost** | **0–0 lost** |
366
+ | argument — three blind judges against a fixed rubric | **0–0 tied, not won** | **0–0 tied, not won** |
347
367
 
348
368
  Argument axis, median of three votes per dimension
349
369
  (`claim_strength`/`evidence_depth`/`counter_specificity`): hand **4/4/4**,
@@ -11,7 +11,7 @@ mcp-name: io.github.0Smallcat0/report-workflow
11
11
 
12
12
  [![CI](https://github.com/0Smallcat0/report-workflow/actions/workflows/ci.yml/badge.svg)](https://github.com/0Smallcat0/report-workflow/actions/workflows/ci.yml)
13
13
  ![Python](https://img.shields.io/badge/python-3.11%2B-blue)
14
- ![Tests](https://img.shields.io/badge/tests-1305%20passing-brightgreen)
14
+ ![Tests](https://img.shields.io/badge/tests-1490%20passing-brightgreen)
15
15
  [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
16
16
 
17
17
  **繁體中文說明 → [README.zh-TW.md](README.zh-TW.md)**
@@ -30,7 +30,8 @@ whatever you already have. Plus one sentence, like "write a lab report on this".
30
30
  and charts drawn from your own numbers. It can follow your department's or your
31
31
  company's template. Chinese or English.
32
32
 
33
- Two lines to install. No API key. Nothing to configure.
33
+ It runs inside an AI agent you already use — Claude Code, Codex, Cursor. No API
34
+ key of its own, nothing to configure.
34
35
 
35
36
  ## Look before you install
36
37
 
@@ -51,19 +52,31 @@ repository, so you can see the output before you run anything:
51
52
 
52
53
  ## Use it
53
54
 
54
- In Claude Code:
55
+ It starts with [uv](https://docs.astral.sh/uv/getting-started/installation/),
56
+ so install that first. Then download it once in a terminal: about 1 GB of
57
+ Python packages and a pandoc binary, more than an agent waits for on a first
58
+ start (Codex gives a new server 10 seconds). Later starts take a few seconds.
59
+
60
+ ```bash
61
+ uvx --from "report-workflow[mcp,render]" report-workflow --help
62
+ ```
63
+
64
+ Then, in Claude Code:
55
65
 
56
66
  ```text
57
67
  /plugin marketplace add 0Smallcat0/report-workflow
68
+ /plugin install report-workflow@report-workflow
58
69
  ```
59
70
 
60
- Then `/plugin install report-workflow@report-workflow`. Using something else
61
- that speaks MCP (Codex, Cursor, your own setup)? One command instead:
71
+ In Codex:
62
72
 
63
73
  ```bash
64
- claude mcp add report-workflow -- uvx --from "report-workflow[mcp,render]" report-workflow-mcp
74
+ codex mcp add report-workflow -- uvx --from "report-workflow[mcp,render]" report-workflow-mcp
65
75
  ```
66
76
 
77
+ In Cursor or any other MCP client, register that same `uvx` command; the JSON
78
+ is in [docs/mcp.md](docs/mcp.md#register).
79
+
67
80
  Then just ask, in your own words:
68
81
 
69
82
  > Use report-workflow to turn the files in ./data into a business report for the
@@ -94,7 +107,9 @@ installed: `uvx report-workflow check-references draft.docx`.
94
107
 
95
108
  It reads `.docx`, `.md`, `.txt` and `.bib`. Each DOI is looked up at doi.org
96
109
  (which also carries retraction notices), each arXiv id at arXiv, and a reference
97
- with neither is searched for in Crossref and OpenAlex. No API key, no account,
110
+ with neither is searched for in Crossref and OpenAlex; when that settles
111
+ nothing and the reference links a web page (a statute, a course, a news
112
+ story), the page is opened and its own title looked for in the reference. No API key, no account,
98
113
  and no AI deciding anything: the same file gets the same verdicts.
99
114
 
100
115
  A five-reference draft with four errors planted in it:
@@ -128,7 +143,7 @@ references here:
128
143
 
129
144
  | Checker | Problem references caught (of 33) | Correct references flagged (of 71) |
130
145
  | --- | ---: | ---: |
131
- | **report-workflow** | **30** | **15** |
146
+ | **report-workflow** | **30** | **10** |
132
147
  | RefChecker | 32 | 36 |
133
148
  | CheckIfExist | 31 | 34 |
134
149
  | Hallucinator | 29 | 28 |
@@ -136,9 +151,10 @@ references here:
136
151
  | HalRef | 24 | 53 |
137
152
 
138
153
  "Flagged" counts anything short of accepted — for the five tools that includes
139
- "requires verification", here it is every verdict except `verified`. All 15 of
140
- those here are `needs_review`, and seven of them are standards, software, datasets,
141
- model cards and workshop pages that no bibliographic registry holds. Of the 13
154
+ "requires verification", here it is every verdict except `verified`. All 10 of
155
+ those here are `needs_review`; three are a standard, a software repository and a
156
+ model card that no bibliographic registry holds and whose pages title them
157
+ differently from the reference. Of the 13
142
158
  references it called `mismatch`, `not_found` or `retracted`, every one is wrong
143
159
  in the hand-checked labels. The three it passes cite the wrong venue or the
144
160
  wrong first name, which it does not compare.
@@ -156,7 +172,8 @@ set exists yet to test it on.
156
172
 
157
173
  The references can all be real and the sentence citing them still wrong: 76%
158
174
  where the paper says 67.0%, "not the cheapest to train" where it says "not the
159
- fastest". Give the command the files the draft cites:
175
+ fastest". Give the command the files the draft cites (`.pdf`, `.docx`, `.md`
176
+ or `.txt`):
160
177
 
161
178
  ```bash
162
179
  uvx report-workflow check-draft draft.docx --source llama.pdf --source qlora.pdf --appendix appendix.md
@@ -165,7 +182,9 @@ uvx report-workflow check-draft draft.docx --source llama.pdf --source qlora.pdf
165
182
  Only sentences that state a figure or quote are checked. Each figure must be
166
183
  printed on a page at the precision the draft gives, a few lines from the
167
184
  sentence's own words; each quotation must be on a page word for word. Every
168
- answer names the file, page and the source's own sentence:
185
+ answer names the file, page and the source's own sentence. The draft and PDFs
186
+ below are in [`benchmarks/check_draft/`](benchmarks/check_draft/), so you can
187
+ run it yourself:
169
188
 
170
189
  ```text
171
190
  [4] NOT_FOUND: In their words, "the preferred model is not the cheapest to train but the fastest at inference".
@@ -196,9 +215,9 @@ in the text really appear in your files. It cannot tell whether the AI
196
215
  understood your data. If the AI writes a smooth sentence that gets your result
197
216
  backwards, that sentence goes through.
198
217
 
199
- We measured that limit instead of talking around it: 69 hand-checked attempts to
200
- sneak something false past it, plus 10,000 test pairs from a public dataset
201
- nobody here wrote. **The cases it still misses are kept in the test set on
218
+ We measured that limit instead of talking around it: 46 hand-checked attempts to
219
+ sneak something false past it and 27 true claims it must let through, plus
220
+ 10,000 test pairs from a public dataset nobody here wrote. **The cases it still misses are kept in the test set on
202
221
  purpose** — [docs/EVIDENCE.md](docs/EVIDENCE.md).
203
222
 
204
223
  **It does not write.** Your AI writes; this decides what stays. So you need an
@@ -256,8 +275,9 @@ python scripts/run_drone_market_benchmark.py --check
256
275
  Three CSVs and one market question. Three recorded documents — a hand-written
257
276
  control, this pipeline's delivered output, and the strongest of four drafts by
258
277
  an AI given the same files. Three axes: counted (numeric), rule-scored
259
- (layout), and judged against a fixed rubric by three independent blind judges
260
- who are told nothing about how any arm was produced (argument). Every number
278
+ (layout), and judged against a fixed rubric by three blind judges — one model
279
+ in three fresh contexts — who are told nothing about how any arm was produced
280
+ (argument). Every number
261
281
  below is from
262
282
  [`benchmarks/evidence/drone_market_2026-09-16/`](benchmarks/evidence/drone_market_2026-09-16/summary.md).
263
283
  Check the date: the next re-recording moves these.
@@ -301,7 +321,7 @@ this benchmark. That is the largest known limitation of these numbers.
301
321
  | --- | --- | --- |
302
322
  | numeric — counted properties of the document | 3–2 won | 4–1 won |
303
323
  | layout — rule-scored structure and table furniture | 4–2 won | 4–2 won |
304
- | argument — three blind judges against a fixed rubric | **0–0 lost** | **0–0 lost** |
324
+ | argument — three blind judges against a fixed rubric | **0–0 tied, not won** | **0–0 tied, not won** |
305
325
 
306
326
  Argument axis, median of three votes per dimension
307
327
  (`claim_strength`/`evidence_depth`/`counter_specificity`): hand **4/4/4**,
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
4
4
 
5
5
  [project]
6
6
  name = "report-workflow"
7
- version = "4.41.0"
7
+ version = "4.41.2"
8
8
  description = "Turn sources into a submission-ready report, with deterministic gates that block any claim not traceable to the evidence."
9
9
  readme = "README.md"
10
10
  requires-python = ">=3.11"
@@ -70,16 +70,23 @@ supports:
70
70
  ```
71
71
 
72
72
  CLI exit codes: `0` success, `1` crash, `2` hard-block validation failure,
73
- `3` waiting for user decisions or agent-authored artifacts. Full signatures and
73
+ `3` waiting for user decisions or agent-authored artifacts. A `prepare` that
74
+ ends at `awaiting_agent_artifacts` exits `0`: writing the briefs is its job; `3`
75
+ comes from `validate`/`run` called before those artifacts exist. Full signatures and
74
76
  parameters: [reference/tools.md](reference/tools.md).
75
77
 
76
78
  ## Required Setup
77
79
 
78
- Python 3.11+, `pip install -e .`, and Pandoc 3.x for high-quality DOCX rendering.
79
- Call `check_environment` before every `start_report`, ask the user about every
80
- pending install and optional integration, then pass `preflight_confirmed=True`
81
- with a complete `preflight_decisions` record. Required dependencies must actually
82
- pass preflight; a decision string does not override a still-missing dependency.
80
+ Through the plugin or MCP server, nothing to install: the server brings its own
81
+ packages and pandoc. From a clone, Python 3.11+ and `pip install -e ".[render]"`.
82
+ Call `check_environment` before `start_report`. Ask the user only about what it
83
+ reports as `required` or `critical` (pandoc), or about an optional integration
84
+ they want to turn on; optional tools and integrations nobody asked for are
85
+ skipped without an answer, so a long task is not interrupted by the same
86
+ questions every report. Pass `preflight_confirmed=True` with a
87
+ `preflight_decisions` record — the one in "Start a Run" below is complete when
88
+ nothing required is missing. Required dependencies must actually pass
89
+ preflight; a decision string does not override a still-missing dependency.
83
90
  Details, UTF-8 setup, and decision examples: [reference/setup-and-preflight.md](reference/setup-and-preflight.md).
84
91
 
85
92
  ## Tool Surface
@@ -134,6 +141,7 @@ start_report(
134
141
  output_dir="/path/to/output",
135
142
  report_profile="engineering_lab_report",
136
143
  task_intent="new_draft",
144
+ title="Step response of a second-order system", # profiles without a cover section print no title without it
137
145
  template_fields={"course_name": "Control Systems", "student_id": "S12345"},
138
146
  preflight_confirmed=True,
139
147
  preflight_decisions={
@@ -38,10 +38,11 @@ paths are accepted for new drafts.
38
38
  ### Declaring what a source is a copy of
39
39
 
40
40
  A source file is not a publication, so by default a `.md`, `.txt` or `.csv`
41
- source is cited as the report's own material — `(本研究整理,計算附件 C3)` /
42
- `(compiled in this study, calculation record C3)` — and nothing is guessed from
43
- its file name. A figure the pipeline computed is labelled `本研究計算` /
44
- `calculated in this study`. Both point at an entry of the calculation record, a
41
+ source is the report's own material: a sentence restating it carries no label in
42
+ the text, and nothing is guessed from its file name. Which row or paragraph
43
+ backs each sentence is in `published/traceability/client_readable_qa_note.md`.
44
+ A figure the pipeline computed is labelled `本研究計算` /
45
+ `calculated in this study` and points at an entry of the calculation record, a
45
46
  document published beside the report (`published/calculation_record.docx`) that
46
47
  gives the result, the formula and every input with its unit, the row it was read
47
48
  from and its source — for a formula table, each row's value of a per-row input.
@@ -93,6 +94,11 @@ checks against.
93
94
 
94
95
  For academic-style profiles, pass structured front matter when available:
95
96
  `title`, `author_block`, `affiliation_block`, `correspondence`, and `keywords`.
97
+ `academic_paper` does not render without any of them, and keywords are never
98
+ inferred for it. A field `start_report`
99
+ was not given can still be supplied during the run as a top-level
100
+ `"front_matter": {"correspondence": "..."}` block in `outline.json` — values
101
+ the user gives you, never invented; `start_report` arguments take precedence.
96
102
  For school/company fixed templates, pass `template_fields` such as `course_name`,
97
103
  `student_id`, `instructor`, `lab_section`, `date`, or `department`. Use
98
104
  `project_identity` when a report must preserve specific project terms, domain
@@ -142,6 +148,11 @@ authoring for new drafts:
142
148
  }
143
149
  ```
144
150
 
151
+ The keys under `sections` are this run's blueprint section ids, the ones the
152
+ outline brief lists — `results` above is only an example. Profiles differ:
153
+ `engineering_lab_report` writes its results under `results_discussion`, and
154
+ `business_report` under `findings`.
155
+
145
156
  When validation sees `structured_drafts.json` and the canonical draft files are
146
157
  missing, the pipeline writes `section_drafts/*.md`, inserts `[CITE:]` markers from
147
158
  `evidence_ids`, and writes `sentence_map.jsonl`.
@@ -263,6 +274,10 @@ contains the value, put it in the sentence ("the error rate fell from 9.0% to
263
274
  pass. A quantitative section with no numbers reads as evasive even when every
264
275
  claim is verified.
265
276
 
277
+ Optional: a `facts_freeze.json` in the run directory (`{"total_files": "388"}`)
278
+ lists confirmed facts; render hard-blocks if any frozen value is missing from
279
+ the document.
280
+
266
281
  ## Document Language
267
282
 
268
283
  Write body prose in the language of the source evidence. When the evidence is
@@ -271,7 +286,10 @@ canonical Chinese section headings; the pipeline detects the document language
271
286
  deterministically and renders section headings from the blueprint's localized
272
287
  titles (`title_zh`), so do not hand-translate headings or mix English
273
288
  boilerplate sentences into a Chinese document. English documents keep the
274
- blueprint's English titles unchanged.
289
+ blueprint's English titles unchanged. When a competition or grant form fixes
290
+ the chapter names, set `"title"` on that section in `outline.json`
291
+ (`"problem_statement": {"title": "原創性說明", ...}`); it replaces the blueprint
292
+ heading in either language.
275
293
 
276
294
  For `academic_paper` and `engineering_lab_report`, make Methods/Procedure
277
295
  reproducible: name the source or sample basis, procedure, parameters or
@@ -293,9 +311,18 @@ Failure returns:
293
311
 
294
312
  - `validation_failed`: repair only the returned `allowed_repair_paths` using
295
313
  `repair_context`.
296
- - `scope_violation`: undo or isolate the out-of-scope artifact changes before
297
- retrying the same stage. Do not manually advance stages by editing
298
- `harness_manifest.json`.
314
+ - `scope_violation`: a file owned by a *later* stage was written early; remove
315
+ it (or restore it) and retry the same stage. Do not manually advance stages by
316
+ editing `harness_manifest.json`. Changing a file an *earlier* stage already
317
+ accepted — fixing the claim matrix during the outline stage because lint
318
+ named it — is not a violation: the run goes back to that stage, validates it
319
+ again, and says so in `note`; submit again for each later stage.
320
+ - A failure routed back to an earlier stage (a publish failure goes to the
321
+ earliest stage that owns part of the fix) opens every artifact from that stage
322
+ onward for repair, and keeps the later artifacts you already wrote: fix the
323
+ claim and the sentence together, then submit stage by stage. After a
324
+ `publish_report` failure you may also edit first and call `submit_action`
325
+ afterwards; the run re-opens at the stage that owns what you changed.
299
326
  - `blocked_non_author_repair`: the failure is outside the current controlled
300
327
  authoring surface. Inspect the returned evidence paths; it needs a workflow,
301
328
  source, environment, or deterministic-code fix, not edits to read-only workflow
@@ -309,8 +336,13 @@ stage or all required artifacts are already present.
309
336
 
310
337
  If you reuse artifacts from an older run, call
311
338
  `remap_agent_artifacts(job_id=..., previous_job_id=...)` first in dry-run mode,
312
- then rerun with `write=True` only after the mapping is reasonable. Do not manually
313
- edit evidence IDs.
339
+ then rerun with `write=True` only after the mapping is reasonable. The write
340
+ copies the files this run does not have yet from the earlier run; under the
341
+ controlled workflow only the open stage's files are copied, and
342
+ `files_deferred` names the rest with the stage that accepts them - call it again
343
+ with `write=True` when that stage opens. Pass `previous_workspace_root` when the
344
+ earlier run was started with a different `output_dir`. Do not manually edit
345
+ evidence IDs.
314
346
 
315
347
  ## Validation Failure Repair
316
348
 
@@ -13,11 +13,13 @@ Load this when preparing a run, resolving a missing dependency, or building the
13
13
 
14
14
  ## Required Runtime
15
15
 
16
- - Python 3.11+
17
- - `pip install -e .` from the repository root
18
- - Pandoc 3.x for high-quality DOCX rendering. Without pandoc, the pipeline falls
19
- back to a limited `python-docx` renderer with degraded table, list, and layout
20
- fidelity.
16
+ - Through the plugin or the MCP server: nothing. `uvx --from
17
+ "report-workflow[mcp,render]" report-workflow-mcp` brings Python packages and
18
+ a pandoc binary of its own.
19
+ - From a clone: Python 3.11+ and `pip install -e ".[render]"` from the
20
+ repository root. Without `[render]` (or a system Pandoc 3.x), the pipeline
21
+ falls back to a limited `python-docx` renderer with degraded table, list,
22
+ and layout fidelity.
21
23
 
22
24
  Optional integrations:
23
25
 
@@ -28,12 +30,21 @@ Optional integrations:
28
30
 
29
31
  ## Run `check_environment` First
30
32
 
31
- Call `check_environment` before every `start_report`. It returns the pending
32
- installs, the features to ask the user about, a human-readable `message`, and the
33
- `required_preflight_decisions` template you should fill in.
34
-
35
- Then ask the user about every pending install and optional integration. After
36
- installing a dependency, rerun `check_environment` to verify. Do not treat an
33
+ Call `check_environment` before `start_report`. It returns the pending
34
+ installs, the features that could be turned on, a human-readable `message`, and
35
+ the `required_preflight_decisions` template. That template holds only what
36
+ `start_report` needs answered — a missing required or critical dependency, a
37
+ feature already switched on — so when nothing is missing it is complete as
38
+ returned. Optional items are listed separately under
39
+ `optional_preflight_decisions`.
40
+
41
+ Ask the user about a pending install only when its severity is `required` or
42
+ `critical`, and about an optional integration only when it is not yet decided
43
+ and the user might want it (web research, NotebookLM). Optional installs such as
44
+ `mmdc` and optional features left out of the record are treated as skipped, so
45
+ `"install_decisions": {}` is a complete answer when nothing required is
46
+ missing — do not stop a long task to re-ask the same optional questions before
47
+ every report. After installing a dependency, rerun `check_environment` to verify. Do not treat an
37
48
  `install` or `installed` decision as proof when `check_environment` still reports the
38
49
  dependency missing: required dependencies must actually pass preflight before
39
50
  start.