Analysis quality audit (1.0)

Status: scaffolded (0.9.5) — registry rows generated; provisional judgements in analysis_quality_audit_judgements.md (0.9.7); owner sign-off open
Programme: pre_release_roadmap_1_0.md §11
Related: release_severity_triage_1_0.md, analysis_module_backlog_2026-07-17.md
Scaffold: analysis_quality_audit_scaffold.md (make docs-gen)
Judgements: analysis_quality_audit_judgements.md (human/agent overlay — not wiped by docs-gen)

Living sheet — one row per user-visible analysis. Prefer real corpora over fixtures alone.

0.9.x freeze: no new modules unless audit proves a release-critical repair.

Column template

Field	Meaning
Analysis / module	User-visible name + id
Intended question	What the user can answer
Output type	JSON / chart / text / …
Algorithm / model	Method identity
Meaningfulness	On real transcripts
Languages	Supported / known limits
Min data	Minimum segments / speakers
Confidence / abstention	How uncertainty is shown
Failure modes	Skip / empty / misleading
Overlap	Related modules
GUI presentation	Where shown; prominence
Group semantics	If applicable
Test quality	Coverage notes
Performance	Cost class
Recommendation	retain / improve / relabel / document as experimental / deprecate / remove
Severity	blocker / must-fix / known limitation / post-1.0

Mandatory scrutiny

Deterministic highlights, summaries, and action-item extraction vs LLM equivalents — improve, restrict claims, reduce prominence, or remove misleading fallbacks.

Rows

Machine-generated registry scaffold (module id, description, category, dependencies, determinism) lives in analysis_quality_audit_scaffold.md. Provisional Recommendation / Severity / Notes for 0.9.7 live in analysis_quality_audit_judgements.md (owner sign-off still required).