Revert "Update category check for LOC calculation" by rmsharp · Pull Request #2 · rmsharp/methodology

rmsharp · 2026-03-31T02:43:51Z

Reverts #1

Adds documentation-project adaptations distilled from real-world use: 4 new failure modes (KJ5HST#20-23), Artifact Integrity section in SAFEGUARDS, build-equivalent step in BOOTSTRAP, documentation adaptation notes in TEMPLATE_WORKSTREAM, AUDIT_WORKSTREAM scope expansion to non-code artifacts, and the category="other" LOC exclusion in the dashboard.

Address PR #5 review feedback: - Check explicit @vitest/coverage package names first, use catch-all only as fallback for unknown variants — prevents redundant tags - Sync the category not in ("assets", "other") LOC fix from tools/ to starter-kit/ (missed in PR #2) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

PR #2 added failure modes KJ5HST#20-23 but references across CLAUDE.md, README.md, and HOW_TO_USE.md still said 19. Also completes the v2.1 release notes with PR #2's additions (Artifact Integrity, build equivalent step, doc project adaptations), fixes anachronistic v2.0 reference, adds KJ5HST#20-23 to the HOW_TO_USE summary table, and syncs the starter-kit dashboard LOC fix with the tools/ copy. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Classifies rad-con's 594 runner-only lines into universal / project-specific / already-canonical / noise. 6 universal upgrade candidates identified: 1. New Phase 3E Runtime Smoke Test (rad-con:205-217) 2. New FM KJ5HST#24 Build-passes-ship-it (rad-con:261) 3. "Structural control, not a suggestion" framing (rad-con:79) 4. Learning: Protocol discipline is perishable (rad-con:303) 5. Grep-inventory anecdote for Planning Sessions (rad-con:108) 6. rad-con Learnings #1, #2, #3, #5 generalized (rad-con:297-301) 509 of rad-con's 594 extra lines are the project-specific Session Learnings section (rad-con:329-838) — zero upstream value. 1 rad-con-internal bug in the Degradation Detection table noted but left alone per issue scope. Audit is the planning deliverable; the v2.3 content-release PR is a separate session (per FM KJ5HST#18 Planning-to-implementation bleed). Ref: KJ5HST#6

rmsharp · 2026-04-26T03:25:33Z

Closing as obsolete — both the original change (573122c) and the revert (d42d16b) are on stale branches and never landed in main. tools/methodology_dashboard.py has since been rewritten in v2.0 and again during the v2.2 upstream merge, so the revert is no longer relevant.

…KJ5HST#7) Issue KJ5HST#7 distilled 6 universal additions from a 90-session field audit of rad-con/SESSION_RUNNER.md. All approved by community feedback (rmsharp), with one wording revision to Learnings row #4. starter-kit/SESSION_RUNNER.md changes: - Phase 1B: added "structural control, not a suggestion" line - Planning Sessions: added concrete grep anecdote (10+ refs, 2 greps) - New Phase 3E: Runtime Smoke Test — old 3E (Commit) → 3F, old 3F (Report) → 3G - FM KJ5HST#24 APPENDED: Build-passes-ship-it — FMs 1–23 unchanged - Degradation Detection: row added for FM KJ5HST#24 - Learnings: 5 new rows (#2–#6) Row #4 wording: rmsharp flagged that "high-quality plans collapse multi-session work into a single session" reads as license to bend "1 and done." Reframed to put completion-criteria verification first and explicitly state that the rule does not bend for high-quality plans. The original lesson (verify by criteria, not by clock) survives; the erosion-bait headline is removed. Backward compatibility: FMs 1–23 not renumbered. Only "Phase 3E Commit" → "Phase 3F Commit" — adopters citing phase numbers in memory should update. Open coordination: PR KJ5HST#9 also claims v2.3 in file edits. Either KJ5HST#9 rebases to v2.4, or both fold into a single v2.3 release before tagging. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Addresses self-review findings #2 and #3. The two index rows previously understated their citation points: - /diagnose: listed only ITERATIVE_METHODOLOGY.md §Debugging Sessions and SESSION_RUNNER.md §Phase 3F. Also cited at ITERATIVE_METHODOLOGY.md Phase 6 step 8 (the master-framework version of the commit-cleanup gate, mirroring the Phase 3F gate in SESSION_RUNNER). - /grill-with-docs: listed only ITERATIVE_METHODOLOGY.md §Phase 2. Also cited at CONTEXT_TEMPLATE.md §Maintenance (the template's closing guidance on how the file is kept current). Both rows now list every place the skill is actually cited on the branch, so a reader using the index as a citation map gets the complete picture.

Fourth tutorial in the hands-on track; accumulates on the held T3 branch per the one-branch roadmap plan (PR KJ5HST#40 stays draft). - New docs/tutorials/T4_workstreams.md: choose the right workstream by matching the task to each lens's "When to Use" — the running todo-CLI becomes an Audit session that surfaces B1/B2 as findings (not fixes) — then spin up a custom RELEASE_WORKSTREAM.md from TEMPLATE_WORKSTREAM when none fits. Headline failure mode: FM KJ5HST#12 (workstream transfer amnesia), reinforced by FM KJ5HST#8/#2. - Cite-don't-restate throughout (workstream catalog, phases, gates, FM list linked, never forked); adopter-layout paths for learner commands, canonical two-up paths for doc links. - Wiring: tutorials README banner/curriculum/table (T4 -> published) + progressive bullet + running-example thread; T3 "Next" -> live T4 link. Gates: faithful GitHub-slug link/anchor checker clean; sample pytest 7/7 (F1/F2 still absent — learner's job); bin/sync --dry-run 0 tutorial paths (non-distribution invariant); bin/tests.sh 51/51. Adversarial 6-dimension review (refute-verified) + independent self-review; the self-review pass caught a Phase-1 campaign-check/no-workstream-fits conflation the workflow missed — fixed. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…on → consolidation) Adds the next roadmap tutorial in the docs/tutorials/ learning track. T6 runs the three campaign session types — planning, one execution unit, consolidation — in miniature against the running-example todo CLI, picking up the repository- wide hardening that T4's Audit session recommended ("bigger than one session"). Teaches recognition first: the three-part "is this actually a campaign?" test (and when NOT to reach for one), with the toy honestly flagged as a scaled-down structural rehearsal that telescopes what must, in practice, be separate sessions. Cites — does not restate — §Multi-Session Campaigns, TEMPLATE_CAMPAIGN, and the realized INHERITED_CODEBASE_FAMILIARIZATION_CAMPAIGN (the Audit-extending shape T4 set up). The per-unit boundary is reinforced as "1 and done" at campaign scale (FM KJ5HST#26 / FM #2); consolidation feeds Development sessions without fixing. Index updated: README status block, chain string (… → T4 → T6), threading paragraph, curriculum table (T6 → published); T4's Next pointer made a live link. Reviewed: deterministic anchor check (25/25 resolve) + a 6-lens adversarial workflow with a default-to-refuted verification pass. Two real defects caught and fixed before commit — a Phase 1B anchor pointing at the wrong section (self-review) and a miscitation of the 7-Dimension Audit Framework as the source of the issue-class names, now sourced to AUDIT §When to Use to match T4 (workflow, confirmed by 4 independent lenses). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Operator confirmed the deliverable at the Phase 0 gate and approved the approach at the Present gate. This is Phase 4 of docs/planning/upstream-read-set-pr-plan.md section 5: ITERATIVE_METHODOLOGY.md must shed >= 11,490 B (68,240 -> <= 56,750 B, the one-read cap). ADOPTER-FACING -- the file carries bin/_manifest.py:62 and this work adds a second distributed row beside it. CONTROL, MEASURED BEFORE ANY CHANGE. wc -c = 68,240. Upstream's copy is 68,247 and the two trees differ by a SINGLE LINE, so this is essentially virgin work that will port cleanly. THE PLAN'S 3.3 RESTS ON AN INVENTORY THAT DOES NOT EXIST. It states "Measured proposals total 12,358 B, which reaches the target", but MERGE-EROSION and EXTRACT-EV-DUPLICATE-PERFTABLE occur nowhere in this repository outside that paragraph and the S129 receipt quoting it, and both are named only as DEFECTS. git log -S traces each to fe906c2, the plan's own drafting commit. The proposals were therefore re-derived and re-measured this session rather than executed from the plan. THE THREE 3.3 DEFECTS, RE-VERIFIED RATHER THAN QUOTED. 1. Merging Protocol Erosion loses Prevention #1 -- confirmed. The runner's only "2 minutes" sentence (SESSION_RUNNER.md:369) is CONDITIONAL; IM's is UNCONDITIONAL, and no counterpart exists. 2. "Gaps identified" is corpus-unique -- confirmed, exactly one occurrence, ITERATIVE_METHODOLOGY.md:634. 3. An extraction needs a manifest row AND BOOTSTRAP.md updated -- confirmed, and the plan UNDERSTATES it: BOOTSTRAP names the manual-copy set at :72 AND :95, plus a tree diagram at :30. APPROVED APPROACH -- APPARATUS-EXTRACT. Move the six contiguous sections L541-861 (Knowledge Accumulation System, Honest Accounting Framework, Scope Validation System, Verification Hierarchy, Session Document Template, Performance Tracking; 14,647 B gross) into a new DISTRIBUTED sibling FRAMEWORK_APPARATUS.md -> docs/methodology/FRAMEWORK_APPARATUS.md, leaving a pointer stub. Net ~13,367 B, landing IM at ~54,873 B. Extraction deletes nothing, so defects #1 and #2 are unreachable by construction. THE ANCHOR PROOF, BECAUSE NO CHECKER CAN DO IT. bin/check-links:105 strips the #fragment and tests file existence only, so a broken anchor is invisible to tooling and the grep IS the entire defense. Every IM heading was slugified GitHub-style and every ITERATIVE_METHODOLOGY.md#... reference in every tracked .md resolved to a line number: 16 distinct fragments across 52 sites, ZERO resolving inside L541-861, ZERO already dangling. The nearest neighbour, #adapting-to-your-domain, resolves to L862 -- the line immediately after the span. WHAT THIS CLAIM FORBIDS ITSELF. No deletion of any kind. No touching Protocol Erosion (L497-540) or the Gaps identified row. The cross-reference rewrites at L189, L190, L290, L309 must alter ONLY the parenthetical: an adversarial reviewer found the drafted patch would have truncated their tails and destroyed three corpus-unique fragments -- the section 3.3 defect shape committed a third time by the fix for it. NO OUTWARD-FACING ACTION: Phase 5 opens the PR and needs the operator's explicit go-ahead, separately. Ledger note carried forward: HANDOFFS.md holds 6 receipts against the N=4 policy and is 74,404 B, 17,654 B over the one-read cap -- up from 5 receipts / 7,305 B over when the operator last chose to leave it at S129. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TGQniNR3JTuiHWU4g9eAzN

…nto a distributed sibling Phase 4 of docs/planning/upstream-read-set-pr-plan.md section 5. Approach APPARATUS-EXTRACT, approved by the operator at the Present gate. ITERATIVE_METHODOLOGY.md 68,240 -> 55,895 B: shed 12,345 B against a requirement of 11,490, landing 855 B under the 56,750 B one-read cap. The six contiguous apparatus sections at L541-861 -- Knowledge Accumulation System, Honest Accounting Framework, Scope Validation System, Verification Hierarchy, Session Document Template, Performance Tracking (14,647 B) -- moved VERBATIM into a new DISTRIBUTED sibling FRAMEWORK_APPARATUS.md -> docs/methodology/FRAMEWORK_APPARATUS.md, leaving a Reference Apparatus stub that names all six and links the file. NOTHING WAS DELETED, so section 3.3's defects #1 and #2 are unreachable by construction. Protocol Erosion (L497-540) was never in the span, so Prevention #1 is untouched; the corpus's only "Gaps identified" travelled with its block to FRAMEWORK_APPARATUS.md:108 and is still exactly one occurrence across the 23 distributed markdown sources. LOSSLESSNESS, PROVED FIVE WAYS. 1. all 318 removed content lines present in the sibling 2. the sibling's body byte-identical to the reconstructed span, modulo two documented edits (the trailing separator, and one back-reference) 3. IM's head and tail byte-identical outside the span 4. all four original parentheticals rewritten, no old form left 5. each of the 13 edited lines has prose identical to HEAD once the added pointer is stripped THE ANCHOR PROOF, BUILT MECHANICALLY BECAUSE NO CHECKER CAN DO IT. bin/check-links:105 strips the #fragment and validates existence only. Every IM heading was slugified GitHub-style and every ITERATIVE_METHODOLOGY.md#... reference in every tracked .md resolved to a line number: 16 distinct fragments across 52 sites, ZERO inside L541-861, ZERO already dangling. Re-run after the move: 66 references, 0 unresolved. #adapting-to-your-domain resolves to L862, the line immediately after the span -- which is what makes the boundary right rather than lucky. THE REACHABILITY REPAIR -- what the anchor proof could not see. An adversarial review found the pattern: the change updated every place that NAMES the six sections and no place that INVOKES what is in them. Those sections were reached by scrolling, not linking, so none of it can appear in a link-based proof. Nine pointers added at the points of need: Principles 4-7, Phase 2 step 8, Phase 6 steps 5-6 and its gate, and the self-referencing "Across the Full Series (Performance Tracking)" heading. Cost 901 B, taking headroom from 1,672 to 855 B. The trade is deliberate: a file under the cap whose content cannot be found is the failure this phase exists to prevent. DEFECT #3 IS A SIX-SITE PROBLEM, not the one the plan names. BOOTSTRAP.md :30, :72, :95 land here; README.md:75, README.md:79 and docs/tutorials/T1_setup.md:63 land in the follow-up commit. VERIFICATION. bin/tests.sh 288 passed / 1 failed, byte-for-byte the control taken before any edit: zero status flips across 285 shared assertions, the only two differing rows being the manifest count 26 -> 27, both passing. The one failure (github source dry-run) is the pre-existing network case, present in the control. check-links 105 links resolve; dashboard twins byte-identical; set(DISTINCTIVE) | set(AMBIGUOUS) == set(tracked_md) holds at 19; bin/sync into a scratch repo lands the sibling with all cross-file links intact. BLAST RADIUS, STATED RATHER THAN ABSORBED. SAFEGUARDS.md:49 caps a commit at five files; this is seven. The extraction, the manifest row, the dashboard doc-set constant (both twins) and BOOTSTRAP.md cannot be split -- any subset leaves bin/tests.sh red or ships an install that omits the file. The consumer documentation that merely describes the change is a separate commit. NO OUTWARD-FACING ACTION: Phase 5 opens the PR and needs the operator's explicit go-ahead, separately. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TGQniNR3JTuiHWU4g9eAzN

…corrected 11 reviewers against the frozen draft 10d4fbf: 8 claim-refuters, 3 completeness critics, 0 errors. Every finding re-verified here before being acted on. The answer did not change. The evidence largely did. Corrections, all recorded in the report's own Review corrections table: 1. The headline table was wrong in 3 of 11 cells and inflated the population 11 -> 3. The "Archived shards" line did not exist at eight of the eleven trims; it and the fold-in instruction both entered at eec1cbb. One positive, two negatives -- not an eleven-row correlation. My raw data said none->none and I transcribed it as "unchanged". 2. 273afff touches neither count line. The out-of-band correctors are 28b2361 and ec87d08 -- the commit Finding #2 faults S133 for citing. 3. "Never reads HEAD" / "false by construction" both overstated. The :166 else-branch reads HEAD and the working tree and FLIPS THE VERDICT TO GREEN, exit 0. Now Finding #6. 4. "five sibling v1.3.0 proofs pass" was inherited from S133, never re-derived. There are 7 v1.3.0 proofs; six siblings. 5. S110's "3,858 B -> 9,890 B" are CHARACTERS. Bytes are 3,880 -> 9,988 on my convention; two reviewers gave two others. Convention-sensitive, unnamed. 6. "differing only in the fields close-out writes" is wrong -- only session and date survive verbatim. 7. The charge against S88 is WITHDRAWN. "Regenerated" is the term of art for a declared field; S88 scoped it correctly. The issue is timing. 8. The charge against 9038e40 is WITHDRAWN. Its commit message states the fold-in and the recount; the generic sentence is emitted by the distributed tool at methodology_trim.py:1168. 9. "19 of 19, nothing skipped" was false: 22 shards, 19 proofs. Three archives have no proof and two of those were edited after archiving -- Finding KJ5HST#7. Strengthened by re-measurement: the sweep widened from 6 identities to 180 across all 11 trims (0 unreachable by session+date, 11 hash-misses, every one a pending->complete finalize); and the off-diagonal instrument pointed at Class A shows regeneration takes the two CHANGELOG proofs green. A better recommendation the draft never considered: starter-kit/HANDOFFS.md :115-116 already forbids a hand-maintained number. And "a trim commit touches nothing but the trim" is not achievable as worded -- .githooks/pre-commit forces CHANGELOG.md co-staging. bin/tests.sh 304 passed / 1 failed / 0 skipped -- identical to the Phase 0 baseline, zero status flips. docs/archive/ untouched. No distributed file touched. No outward-facing action. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01318np8PmwYj3gqhQBL9a4x

Revert "Update category check for LOC calculation"

d42d16b

rmsharp closed this Apr 26, 2026

rmsharp deleted the revert-1-rmsharp-patch-1 branch April 26, 2026 03:26

rmsharp mentioned this pull request May 2, 2026

Integrate Claude Code skills as optional tools per phase #6

Closed

4 tasks

Navigation Menu

Repository navigation

Revert "Update category check for LOC calculation" - #2

Revert "Update category check for LOC calculation"#2
rmsharp wants to merge 1 commit into
documentation-project-adaptationsfrom
revert-1-rmsharp-patch-1

rmsharp commented Mar 31, 2026

Uh oh!

rmsharp commented Apr 26, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

rmsharp commented Mar 31, 2026

Uh oh!

rmsharp commented Apr 26, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant