Repository navigation
Add a completeness-critic review lens (closes #55) - #56
Merged
rmsharp merged 2 commits intoJul 9, 2026
Merged
Conversation
Promotes Learning KJ5HST#7 (cross-reference completeness at self-review) and Learning KJ5HST#8 (close-out-gate checklist propagation) from authoring-time self-checks to an explicit review-time lens: a review/audit pass now owes a whole-corpus sweep (not just the diff) for enumerations, worked examples, indexes, and count-claims a change made stale elsewhere. - SESSION_RUNNER.md: new Learning KJ5HST#10 - AUDIT_WORKSTREAM.md: new anti-pattern KJ5HST#9 "Diff-scoped blind spot", a new Verification Checklist bullet, and a note that /code-review, /review, /security-review are diff-scoped by design - ITERATIVE_METHODOLOGY.md: one sentence in Review/Audit Sessions citing the new Learning + the operative checklist step Motivated by the v3.3 doc-completeness gap (PR KJ5HST#54) that a clean 6-lens adversarial review missed. No new phase, gate, principle, or workstream; failure-mode count stays 27. Design adversarially checked by a 4-lens review before commit (2 clean, 2 raised real findings — a mis-anchored insertion point and a missing starter-kit/ citation prefix — both fixed). bin/tests.sh 84/84; bin/check-links clean. Closes KJ5HST#55
Version bumped v3.3 to v3.4 covering the completeness-critic lens (issue KJ5HST#55): CLAUDE.md "Current version" line + new §Versioning entry, README.md What's New. bin/tests.sh 84/84; bin/check-links clean.
rmsharp
added a commit
to rmsharp/methodology
that referenced
this pull request
Jul 9, 2026
# Conflicts: # CHANGELOG.md
rmsharp
added a commit
to rmsharp/methodology
that referenced
this pull request
Jul 9, 2026
Writes this session's HANDOFFS.md receipt (S5) and the CHANGELOG close-out entry recording PR KJ5HST#56 merge, the v3.4 tag + GitHub Release on both remotes, the fork-main sync (CHANGELOG union conflict resolved by wall-clock authorship order), and branch prune. bin/check-handoff OK; bin/tests.sh 84/84; bin/check-links clean; fork main 0 commits behind upstream/main.
rmsharp
added a commit
to rmsharp/methodology
that referenced
this pull request
Sep 7, 2026
… self-score 8 -> 7 The close-out at 00347a7 reported Phase 3 complete. It was 3 of 7 steps short: 3A (evaluate the predecessor), 3B (self-assess) and 3C (document learnings) were never performed, and the operator had to ask for them. bin/check-handoff returned 0 throughout, which is exactly what SESSION_RUNNER.md 3D says it means -- it asserts a receipt's presence and completeness, never its quality. 3C: Learning KJ5HST#56 appended to starter-kit/FRAMEWORK_LEARNINGS.md -- "A convention counted across a corpus can be a sample of the rule's OUTPUT, not the rule, and the lone exception is often the case the written policy was built around." check-learnings OK, 55 rows, contiguous 1..55, 0 over 1,500 B. The next number was 56 not 55: KJ5HST#14 is deliberately reserved (bin/check-learnings:74), so the checker's "contiguous 1..54" was a COUNT while the highest row was already 55. 3A/3B: written below the receipt's closing fence. check-handoff refused the first draft at 13,786 B over a 12,288 B per-record budget and said to cut the trailing prose first -- the budget is sized so the fenced requirements fit and the essays are what it excludes. Cut to 12,288 exactly, findings kept, prose sacrificed. self_score revised 8 -> 7. The trim itself is unchanged and still twice-proved; a phase reported complete while 3 of 7 steps were missing is not an 8, and the missing three are the ones that exist for someone other than this session. Predecessor S151 scored 7/10. 3F cross-reference sweep found two things. (a) Prepending this entry moved the CHANGELOG line numbers the receipt cites; :213/:299/:307 re-anchored to :254/:340/:348 and every one re-verified against the line it names. (b) NEW, recorded not fixed: HOW_TO_USE.md:774 says the failure-mode table has 27 rows; the highest FM is 28, CLAUDE.md:104 already says 28, the prose is live rather than frozen release notes, and HOW_TO_USE.md is TRACKED -- so the stale count ships to every adopter. bin/tests.sh 304 passed / 1 failed / 0 skipped -- identical summary, zero status flips; the four changed rows are derived counts (KJ5HST#56->KJ5HST#57, S152->S153) that confirm the additions landed. check-learnings / check-links / check-handoff all 0. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KSL144YZnmqbjfVhewTj1a
rmsharp
added a commit
to rmsharp/methodology
that referenced
this pull request
Sep 21, 2026
…ntence about it A config row is one object in JSON but many lines in git. The note BL-78 P2 replaced opened "max_tokens 6,777 SINCE S177" and four clauses later dated the BYTE pin beside it to S177 as well. git blame splits the two adjacent lines by 17 days: max_tokens is 0e8c6ac (2026-09-16, S177), max_bytes is beffbd0 (2026-08-30, S129), and git log -S on the literal returns that one commit. The misdating inverted the note's own characterization -- pinned at S129 the figure was the file's measured size, stable 53 days, exactly the "measurement dressed as a budget" the note insisted it was not. Three sessions costed, ratified and executed against that warrant without blaming the line. D3 discharged by refusal, rows considered named: KJ5HST#86, KJ5HST#87, KJ5HST#88, KJ5HST#54, KJ5HST#56. None meets (a) -- no gate, test or failure mode enforces their lessons; none meets (b) -- KJ5HST#89 narrows the provenance lesson rather than generalizing it; none meets (c) -- every artifact they are about still exists. Verified: check-learnings 75 rows, contiguous 15..89, 0 over 1,500 B. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
SESSION_RUNNER.md— new Learning feat: extend research-documentation workstream with anti-patterns 14-19 #10 (table was 1-9)AUDIT_WORKSTREAM.md— new anti-pattern feat: add Research Documentation workstream (v2.3) #9 "Diff-scoped blind spot" (list was 1-8), a new Verification Checklist bullet, and a note that/code-review//review//security-revieware diff-scoped by designITERATIVE_METHODOLOGY.md— one sentence in §Review/Audit Sessions citing the new Learning + the operative checklist stepCLAUDE.md§Versioning +README.mdWhat's New)Motivated by the v3.3 close-out-receipt work (#52/#53): a clean 6-lens adversarial review shipped, yet left
HOW_TO_USE.md, theREADME.mdrepo tree, and three tutorials describing close-out without the new receipt — none were in the diff, so no diff-scoped lens looked at them. A human had to ask "has documentation been fully updated?" (fixed in #54).No new phase, gate, principle, or workstream; failure-mode count stays 27. All three touched files are
bin/_manifest.py-distributed, so adopters receive this viabin/sync.Design verification
Drafted, then adversarially checked by a 4-lens review before commit (acceptance-criteria coverage, numbering/citation fact-check, a reflexive Learning-#7 self-check for other stale cross-references, and placement/precedent judgment) — 2 lenses clean, 2 raised real findings (a mis-anchored insertion point in
AUDIT_WORKSTREAM.md's Recommended Skills section; a citation missing thestarter-kit/path prefix used elsewhere in the repo), both fixed before commit.Test plan
bin/tests.sh84/84bin/check-linkscleanCloses #55
🤖 Generated with Claude Code