v1.9.2

PATCH release. After 8 minor releases (v1.4.0 → v1.9.1) shipping tools,
protocols, audits, refactors, vendored assets, and deprecations, the
system needed an end-to-end audit before further feature work
(v1.10.0 slides+poster, v2.0.0 deprecation cleanup). v1.9.2 ships the
findings doc + only the most trivial fixes; substance is deferred to
v1.9.3.

Added

• docs/AUDIT_v1.9.2.md — comprehensive audit synthesis
  (~12,000 words) aggregating 75 findings across 10 lenses: 5 CRITICAL,
  27 HIGH, 29 MEDIUM, 22 LOW. Includes per-finding IDs (AUDIT-v1.9.2-001
  through 075), reproduction steps, suggested fixes, target-version
  routing, and a 35-item v1.9.3 work-list (~28 agent-hours).
• docs/audit_v1.9.2/ — per-lens raw findings:
  • lens_01_biology_stress.md — full RNA-seq workflow trace
  • lens_02_humanities_stress.md — DH/literary-analysis composition
  • lens_03_qualitative_stress.md — 12-participant interview study
  • lens_04_tool_protocol_consistency.md — tool↔protocol↔server wiring
  • lens_05_protocol_graph.md — recommendation graph topology
  • lens_06_docs_accuracy.md — doc currency + cross-ref integrity
  • lens_07_config_schema.md — researcher_config field health
  • lens_08_test_coverage.md — coverage + brittleness audit
  • lens_09_dead_code.md — vulture + orphan-module sweep
  • lens_10_audit_gates.md — audit-of-audits consistency
• docs/PROTOCOL_GRAPH.mermaid — recommendation graph (Mermaid)
  rendered from next_protocol + see_also edges across all 114
  protocols.

Fixed (trivial only — substantive bugs deferred to v1.9.3)

• Stale protocol/tool counts corrected across 8 user-facing docs:
  README.md, START.md, AI_GUIDE.md, PROTOCOLS.md, TOOLS.md,
  FAQ.md, RESEARCHER_GUIDE.md, ROADMAP.md, CONTRIBUTING.md
  (113 → 114 protocols, 146 → 212 tools, 438 → 872 tests).
• sys_help topic descriptions: methodology count (29 → 42),
  synthesis count (14 → 18), sys_active_tools description
  (143 → 212 tools).
• Stale tool_audit_master references (renamed to
  tool_audit_quality_full earlier): fixed in docs/ROADMAP.md and
  src/research_os/tools/actions/audit/step_literature.py docstring.
• src/research_os/protocols/_router_index.yaml: removed accidental
  "re-running by re-running" duplication in provenance_completeness
  decomposition purpose.
• src/research_os/tools/actions/state/path.py: removed unreachable
  return out after the real return (dead code, no behavior change).
• src/research_os_humanities/detector.py: docstring extension list
  corrected (.pdf → .tex to match actual filter).
• .gitignore: added .claude/ and .coverage (dev artefacts).

Deferred to v1.9.3 (next release — bug fix + coherence sweep)

35 work-items totalling ~28 agent-hours. Highlights:

• CRITICAL: gate_strictness / project_tier / reliability
  model_profile are read from the wrong path — v1.5.1 features
  silently inert in real projects (lens 07).
• CRITICAL: override_discussion_coverage documented but handler
  discards args and function takes no override (lens 10).
• CRITICAL: Step-completeness gate hard-requires a PNG/SVG focal
  figure per step — structurally blocks every humanities project from
  reaching synthesis (lens 02).
• CRITICAL: Seven tools referenced by digital_humanities_workflow +
  scholarly_edition do not exist; humanities pack unrunnable as
  written (lens 02).
• CRITICAL: project_tier never propagates to
  resolve_gate_strictness; throwaway/sketch/production tags do
  nothing (lens 10).
• HIGH: tool_audit_quality_full runs 6 gates but description
  advertises 5; grounding_verify is invisible (lens 10).
• HIGH: override_no_pdfs bypass does not write to
  override_log.md; audit trail invisible (lens 10).
• HIGH: RESEARCHER_GUIDE.md config schema is 5 fields behind
  template (lens 06).
• HIGH: PROTOCOLS.md silently omits 34 of 114 protocols including
  8 of 14 visualization protocols and 4 newest synthesis protocols
  (lens 06).
• MEDIUM: Router miscarries "fit a DESeq2 differential expression
  model" to methodology/bayesian_analysis (lens 01).
• MEDIUM: synthesis_paper decomposition omits
  tool_paper_compile_typst — small-model agents never produce a PDF
  (lens 01).

Subsystem health snapshot: protocol_graph, tool_wiring, tests,
dead_code, biology_core_path → green; documentation,
qualitative_pack → yellow; researcher_config_schema,
audit_gate_machinery, humanities_pack → red.

Surface

No protocol or tool surface changes. No new dependencies (dev tools
pytest-cov and vulture used in-session but not added to runtime
requirements). 114 protocols, 212 tools, 872 tests — identical to
v1.9.1.

---