Skip to content

Editorial QA report — published sample pieces

Corresponds to issue PAR-12 (Editorial QA pass on published sample pieces). Reviewed against the editorial standards in How PartlyGood works and the playbook template contract in samples.

Scope

Two published playbooks reviewed:

  • docs/management/drafting-meeting-minutes-with-ai.md (management)
  • docs/operations/vendor-spend-review.md (operations)

What was checked

  1. AI-usage claim metering — every playbook must meter AI work against human work: an explicit AI usage statement, at least one human-verification gate, and zero overreaching "AI decides/owns" claims.
  2. Sources — every factual claim about a tool/model carries a source URL; every source resolves; sources are annotated in the colocated sources.json.
  3. Report metrics — report.json regenerated with per-piece ai_meter and sources_verified records.
  4. Quality bar — structural validation, the 9-point quality gate, and a strict build must all pass.

Result per piece

Piece AI meter Sources Gate Verdict
Meeting minutes statement ✓ · 1 human gate ✓ · 0 overstated ✓ 1/1 verified 9/9 PASS
Vendor spend review statement ✓ · 1 human gate ✓ · 0 overstated ✓ 4/4 verified 9/9 PASS

Meeting minutes (drafting-meeting-minutes-with-ai.md)

  • AI usage metered: the blockquote at the top names what AI does (transcription + draft) and what the human owns (verify, correct, decide). Includes an explicit human-verification step (step 2 review + step 3 action item pass). No overstated "AI decides/owns" language.
  • Sources: 1 URL declared; now https://support.microsoft.com/en-us/word/transcribe-your-recordings (the prior .../office/transcribe-your-meetings-... URL was a 404). Resolves 200.
  • Sentence-level fixes: none required beyond the source URL.

Vendor spend review (vendor-spend-review.md)

  • AI_usage metered: Top AI-usage statement names what AI drafts and what the human cross-checks; Step 4 is an explicit human verification gate ("You are the final reviewer. This step is not optional."). No overstated claims.
  • Sources: 4 declared. One (Copilot privacy, previously learn.microsoft.com/en-us/copilot/privacy-and-protected-material) was a 404; replaced with the canonical https://learn.microsoft.com/en-us/microsoft-365-copilot/microsoft-365-copilot-privacy (200). APQC resolves (200 via GET), OpenAI help center resolves. All 4 verified reachable.
  • Sentence-level fixes: the dead source URL was the only content change.

Fixes published

  • docs/management/drafting-meeting-minutes-with-ai.md — source URL fix.
  • docs/operations/vendor-spend-review.md — source URL fix.
  • playbooks/ops/vendor-spend-review/sources.json — corrected src4 URL/title.
  • playbooks/mgmt/meeting-agenda-outlook/sources.json — annotated sources now present for the meeting-minutes piece.
  • scripts/quality_checks.py — added the ai_usage_meter check (the metering gate).
  • scripts/quality_report.py — report now emits per-piece ai_meter and sources_verified records; adds --verify-sources network verification.
  • report.json — regenerated with the new metrics.

Re-run

python scripts/quality_checks.py --all
python scripts/quality_report.py --verify-sources
python scripts/validate_content.py --strict
python -m mkdocs build --strict