docs(release): v3.5 — capability-tiered review by rmsharp · Pull Request #58 · KJ5HST/methodology

rmsharp · 2026-07-09T04:56:08Z

Summary

Version bump v3.4 → v3.5 covering the capability-tiered review elective addition merged in #57.

CLAUDE.md — "Current version" line + new §Versioning entry narrating the full change
README.md — new What's New in v3.5 section

Cite-don't-restate: the full narrative lives in CLAUDE.md §Versioning "v3.5". No behavioral content changes here — PR #57 already shipped the actual addition; this only adds the release narration.

Test plan

bin/tests.sh — 84/84 passed
bin/check-links — clean

🤖 Generated with Claude Code

Bump v3.4 -> v3.5: CLAUDE.md "Current version" line + new §Versioning entry; README.md What's New. Covers the capability-tiered review elective addition (BL-7, PR KJ5HST#57) already merged to main. Cite-don't- restate: the full narrative lives in CLAUDE.md §Versioning "v3.5". Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Merge upstream/main (6b06fd4) into fork main. Resolved one union conflict in CHANGELOG.md by wall-clock order: the "Released v3.5" entry (just happened) lands above the prior "[BL-7] Fork close-out" entry (happened earlier, already on fork main before this release work started). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

…-force, 245,254 B -> 42,938 B The receipt ledger is back inside its retention policy for the first time since S132: 27 receipts -> 4, and methodology_trim.py --check now exits 0, "trigger does not fire", where at Phase 0 it exited 1. 23 records (2026-08-12 -> 2026-09-04) frozen in docs/archive/HANDOFFS-through-2026-09-04.md with its .verify.sh. Two operator decisions, asked one at a time, neither inferred: the S129 deprioritisation released at the Phase 0 gate, and the --force put back separately because srf-red-refusal-adjudication.md 2 says of this exact file "the trim must not be forced" and 11.2 supersedes that only in part. A go-ahead for the deliverable is not a go-ahead for a flag that overrides a standing written refusal. The force is scoped to this file, this regime, this evidence; the next trim asks again. The gate was found because the predecessor's recipe does not run: --write alone exits 2 and writes nothing, and the refusal is cut-independent. That is Learning KJ5HST#57, authored by S153 in that same session, applied to its own handoff. Both commit shapes were built and the proof run on each. Folded inside the trim commit the shipped .verify.sh fails "L2 FRONT MATTER lost 1 line(s)"; folded in the next commit it exits 0. New Learning KJ5HST#58. Distinct from BL-50, which is the writer's L2 refusing at write time. Three budgets refused an edit and all three times the cut was taken, never the constant: the 7,168 B header reserve (front matter 7,570 -> 7,096, so 72 B clear against the 46 B inherited, with the bytes taken out of claims that had gone false), the 1,500 B learning-row cap (KJ5HST#58 rewritten 1,511 -> 1,455 B), and the 12,288 B record cap (the 3A/3B essays cut to fit at 12,208 B). A comment edit carrying KJ5HST#58's rule into the front matter was measured, applied and then REVERTED because it left 7 B of headroom. Handed forward as a question, not decided: the header reserve is at 99% and blocked two correct edits this session, while the A1 fit leaves 21,504 B of slack. Raising it is arithmetically free, which is exactly why S153 forbade doing it silently. The trade is the operator's. Verification, the proof run after committing four times with a negative control: verify.sh exit 0 on the raw trim in a clone, after the fold in a clone, and on the live tree; the one-commit control failed (1). Write-time L1_OK/L2_OK/L3_OK/P1A_OK, 27 records partitioned byte-identical, stale count regenerated 24 -> 4. bin/tests.sh 304 passed / 1 failed / 0 skipped exit 1 read bare, row-for-row against the Phase 0 baseline: 303 rows both sides, zero status flips, zero added, zero removed. check-links / check-learnings / check-handoff all 0. bin/model-report reads the new shard (14 files, archived 48, the new shard 5), so the S133 glob fix held and this trim did not gut it. CUT_STRADDLES_DAY not overridden. Both live ledgers backed up outside the repo and shasum -c clean after every run. Self-score 8/10; predecessor S153 scored 8/10. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01De6mNLbCkAyLpmt4pEbF7N

…for the row Separate commit by design: folded into the trim commit, the shipped .verify.sh fails L2 FRONT MATTER (Learning KJ5HST#58, measured at S154 by building both shapes). The fold trades methodology_trim.py's ~447 B pointer block for one ~182 B table row, but the trim's regenerated fields had already pushed the front matter to 7,286 B against the 7,168 B HEADER_RESERVE_BYTES the operator chose to keep at S155. So the row is paid for in text, not in the constant: - the note that "validates only the newest receipt", stated here until S127, was false -- a correction of a sentence this file no longer contains - a shard-drift example's commit sha and the phrase "corrected three sessions later", where "for three sessions once" carries the same fact - four phrases reworded shorter across the retention and resync paragraphs Front matter 7,286 -> 7,159 B, 9 B inside the reserve. The archive table's counts are corrected with the new row: 12 trims / 139 receipts -> 13 / 141, and the "currently holds 4" field is true again. The NEXT TRIMMING SESSION comment now carries the own-commit rule S154 could not fit. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0193vQmoYCLGUYwRpcniWkZi

…B -> 42,923 B The remedy the operator selected from S155's close-out. 6 receipts -> 4, and the file fits in one read again: 19,040 tokens on the doubled-file meter (38,081 halved), 76.2% of the 25,000-token cap, down from 28,287 at claim. 22,613 B of ceiling headroom, so the next session can write a receipt -- which at claim time it could not. The warrant was corrected mid-session: methodology_trim.py's own TRIGGER_READ line rules the token overage "for reference and not as a fault", because a newest-on-top ledger truncates its oldest records. The forcing condition was 2 B of byte headroom. No gate: SRF 0.1096 vs aa1c476, dry run exit 0, no --force needed or asked for -- recorded next to S154's 3.7022 refusal so the two do not read as one. The post-claim re-derive changed the cut from 1 record to 2. Learning KJ5HST#58 verified rather than cited: the fold went in as its own commit and the frozen proof was run twice, exit 0 both times. New Learning KJ5HST#60: a file's bytes_per_token is a snapshot of the content it held when measured, not a property of the file -- HANDOFFS.md metered 2.3164 B/token at 65,531 B and 2.2543 at 42,923 B the same day, the latter below the 2.27 floor that context_budget.py:68 calls conservative. The config was NOT edited; with two conflicting measurements the right value is a judgment. Suite 304/1/0 exit 1, 305 rows row-for-row against S155's final run, zero status flips. 0 skipped -- Test 34's three-receipt floor survived. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0193vQmoYCLGUYwRpcniWkZi

…22 B -> 38,884 B methodology_trim.py v1.5.0 --file HANDOFFS.md --cut 4 --write, re-derived after the S158 claim commit. Archives S154 and S153 (both 2026-09-08) to docs/archive/HANDOFFS-through-2026-09-08.md with its .verify.sh; retains S158, S157, S156, S155. SRF 0.7414 vs c581ac4, no SRF_RED, no --force. L1/L2/L3/P1A all OK at write time. The generated pointer block is left in place: folding it into the archive table here would make the shipped proof fail L2 (Learning KJ5HST#58). The fold is the next commit. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018E7BB3NbBVYU7wXNmNkbg7

… with four spent clauses The trim commit 000a843 left methodology_trim.py's pointer block in place; folding it there makes the shipped .verify.sh fail L2 (Learning KJ5HST#58). Here it becomes one table row (448 B out, 190 B in); counts 13/141 -> 14/143. The front matter sat 9 B under the 7,168 B reserve, so the row needed 181 B. Four spent clauses paid 201 B: "where S152's trim moved it", the residue "it is not newest-only. Since S133", the "read 6" anecdote, and S94's 235/1 -> 229/6 evidence (BL-40 (b) still cited). Front matter 7,607 -> 7,148 B, 20 B spare. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018E7BB3NbBVYU7wXNmNkbg7

…10 B -> 38,737 B methodology_trim.py --file HANDOFFS.md --cut 4 --write, re-derived after the S160 claim: S156 and S155 (both 2026-09-09) archived to docs/archive/HANDOFFS-through-2026-09-09.md; S160, S159, S158 and S157 retained. SRF 0.9095 vs 000a843, no --force. L1/L2/L3/P1A OK at write. CUT_STRADDLES_DAY: S157 also carries 2026-09-09, so the shard name is a span label. The pointer block is folded in the next commit, not this one (Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014PNNUm2iQYQhp6hgU5cvzm

… by dropping its proof column The trim commit cd52df7 left methodology_trim.py's pointer block in place; folding it there makes the shipped .verify.sh fail L2 (Learning KJ5HST#58). Here it becomes one 125 B table row (the 448 B block out), and the table's proof column (922 B) is dropped on the operator's decision: each of its 14 cells was asserted equal to its own row's shard path plus .verify.sh before removal -- the rule HANDOFFS.md:46 already states. Counts 14 trims / 143 receipts -> 15 / 145. Front matter 7,596 -> 6,351 B, 817 B under the 7,168 B reserve. Verified before commit, in a --no-local clone with this content committed: bin/tests.sh 304/1/0, 305 rows, 0 status flips vs the claim-commit baseline. 12 of 15 shard proofs exit 0; the other 3 fail with byte-identical output at 7a6ea4b, before any S160 change (BL-36 x2, and 9038e40's in-commit fold). bin/check-links cannot see this file -- it validates distributed files only -- so its 19 relative links were resolved directly: 0 unresolved. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014PNNUm2iQYQhp6hgU5cvzm

…2 B -> 50,737 B methodology_trim.py --file HANDOFFS.md --cut 4 --write, on the operator's request under the retention policy at HANDOFFS.md:8 (5 receipts vs 4). S157 (2026-09-09) moves to docs/archive/HANDOFFS-through-2026-09-09-2.md, the name disambiguated because -09-09 is taken. SRF 0.9849 vs cd52df7: no SRF_RED, no --force. L1/L2/L3 OK at write. The pointer block is folded into the archive table in the next commit (Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K59TajFbZkHc3fPJehFaxC

…FS.md back to four receipts The trim commit 16ac3fc left methodology_trim.py's generated pointer block in the front matter, because folding it there makes the shard's proof fail L2 (Learning KJ5HST#58). This commit folds it into the archive table as one row (456 B out, 129 B in): 16 trims, 146 receipts, each row cross-checked against the fence count of the shard it names. Front matter 6,480 B of the 7,168 B reserve; live file 50,423 B. The S161 receipt's item (1) records the HANDOFFS.md half as done. Proofs in a --no-local clone: the new shard's exits 0; 13 of 16 pass, and the 3 that fail are the known pre-existing -08-02, -08-09 (BL-36) and -08-25. Suite: 304/1/0, 0 status flips against the Phase 0 baseline. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K59TajFbZkHc3fPJehFaxC

…71 B -> 41,455 B, forced past SRF_RED on the operator's decision methodology_trim.py --file HANDOFFS.md --cut 4 --force --write. S159 and S158 (both 2026-09-10) move to docs/archive/HANDOFFS-through-2026-09-10.md; four receipts remain, the stub of this session's among them. L1/L2/L3/P1A OK at write. --force is the operator's decision at S163 (Q2 A), scoped to this trim: SRF_RED fired at 1.3309 against 16ac3fc, because the last trim archived one receipt (8,065 B) and the growth since exceeds it -- the on-schedule refusal srf-red-refusal-adjudication.md s11.1 proves no retention policy can avoid. Without it this session's receipt could not fit under CEILING_BYTES (5,988 B of room; the smallest record here was 9,106 B). The pointer block is folded into the archive table in the next commit, not this one (Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…FS.md 17 trims, 148 receipts The pointer block methodology_trim.py appended for docs/archive/HANDOFFS-through-2026-09-10.md becomes one table row (n = 2, the shard's own fence count). Its own commit, after the trim (Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…36 B -> 37,186 B methodology_trim.py --file HANDOFFS.md --cut 4 --write, under the retention policy at HANDOFFS.md:15 (six blocks after the S165 claim; it keeps four). S161 and S160 (both 2026-09-11) move to docs/archive/HANDOFFS-through-2026-09-11.md; four remain, this session's stub among them. L1/L2/L3/P1A OK at write. No --force: SRF 0.9383 against ae44dec after the claim, under the 1.0 refusal (SRF_RED returns at 61,471 B; the claim left the file at 60,236 B). The pointer block is folded into the archive table in the next commit, not this one (Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…FS.md 18 trims, 150 receipts The pointer block methodology_trim.py appended for docs/archive/HANDOFFS-through-2026-09-11.md becomes one table row (n = 2, the shard's own fence count). Its own commit, after the trim d9ace03 (Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…66 B -> 38,813 B methodology_trim.py --file HANDOFFS.md --cut 4 --write, under the retention policy at HANDOFFS.md:15 (six blocks after the S167 claim; it keeps four). S163 (2026-09-15) and S162 (2026-09-14) move to docs/archive/HANDOFFS-through-2026-09-15.md; four remain, this session's stub among them. L1/L2/L3/P1A OK at write. CUT_STRADDLES_DAY: the name is a span label, since three retained receipts share 2026-09-15. No --force: SRF 0.8755 against d9ace03 after the claim, under the 1.0 refusal (SRF_RED returns at 60,236 B; the claim left the file at 57,366 B). The pointer block is folded into the archive table in the next commit, not this one (Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…FS.md 19 trims, 152 receipts The pointer block methodology_trim.py appended for docs/archive/HANDOFFS-through-2026-09-15.md becomes one table row (n = 2, the shard's own fence count). Its own commit, after the trim 5049845 (Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…51 B -> 35,001 B methodology_trim.py --file HANDOFFS.md --cut 4 --write, under the retention policy at HANDOFFS.md:15 (six blocks after the S169 claim; it keeps four). S165 and S164 (both 2026-09-15) move to docs/archive/HANDOFFS-through-2026-09-15-2.md -- the name disambiguated, the earlier shard untouched; four remain, this session's stub among them. L1/L2/L3/P1A OK at write. CUT_STRADDLES_DAY: the name is a span label. No --force: SRF 0.8644 against 5049845 after the claim, under the 1.0 refusal. The pointer block is folded into the archive table in the next commit, not this one (Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…FS.md 20 trims, 154 receipts The pointer block methodology_trim.py appended for docs/archive/HANDOFFS-through-2026-09-15-2.md becomes one table row (n = 2, the shard's own fence count). Its own commit, after the trim 0ccfce8 (Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…2 records to a frozen shard, 40,816 B -> 15,874 B Applies the N=1 retention policy: S175's Phase 0 counted three receipts, above two. S173 and S172 move to docs/archive/HANDOFFS-through-2026-09-16-2.md with its emitted proof; L1, L2 and L3 hold, and verify.sh exits 0 against the working tree before this commit. --cut 2 rather than the policy text's --cut 1 because this session's claim stub is a fourth record; the cut archives the same two receipts a Phase 0 --cut 1 would have (24,942 B either way). --force as the policy states: SRF_RED refuses every on-schedule retention trim by construction (BL-59). The fold of the pointer block into docs/HANDOFFS_ARCHIVE_INDEX.md follows in its own commit (Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The pointer block methodology_trim.py wrote for docs/archive/HANDOFFS-through-2026-09-16-2.md (trim 4679cca) becomes one row in docs/HANDOFFS_ARCHIVE_INDEX.md and leaves HANDOFFS.md. The first fold since S174 moved the table out of the front matter: 4,475 B inside the trim commit, back to 4,019 B after this one. Its own commit, after the trim (Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…ARNINGS.md The last of resync decision D1 (A)'s follow-through. The runner's Phase 3C sends a canonical-repo session to starter-kit/FRAMEWORK_LEARNINGS.md, which on this fork is now upstream's file; a short CLAUDE.md subsection, placed outside the table row upstream rewrites at M3, routes fork learnings to docs/FORK_LEARNINGS.md and fixes the citation form. The archive index's Learning KJ5HST#58 link follows. Corrects the previous ledger entry's per-number breakdown of the dashboard test sites (totals unchanged). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…9 records to a frozen shard, 68,571 B -> 16,296 B Applies the N=1 retention policy at R1's end, as the resync plan's R1 step 5 and S175's handoff name it: the M1 merge brought upstream's eight receipts S21, S19-S13, so the file held 11. --cut 2 keeps this session's stub and S175's receipt; S174 and the eight upstream receipts move to docs/archive/HANDOFFS-through-2026-09-16-3.md with its emitted proof. L1, L2 and L3 hold, and verify.sh exits 0 against the working tree before this commit. Every commit before this one is already recorded in CHANGELOG.md. --force as the policy states: SRF_RED refuses every on-schedule retention trim by construction (BL-59). The fold of the pointer block into docs/HANDOFFS_ARCHIVE_INDEX.md follows in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

084ba1b's pointer block becomes one row of docs/HANDOFFS_ARCHIVE_INDEX.md and leaves HANDOFFS.md, in its own commit (fork Learning KJ5HST#58). The fold comment's Learning KJ5HST#58 citation now reads fork Learning KJ5HST#58 -- a live instruction in the front matter that the citation commit's inventory skipped with the ledgers. Front matter 4,024 B against the 7,168 B reserve. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…4 records to a frozen shard, 47,200 B -> 16,224 B Applies the N=1 retention policy at R2's end: stages M3 and M4 brought upstream's S22, completed S20 and S23, so the file held 6. --cut 2 keeps this session's stub and S176's receipt; S175, S23, S22 and S20 move to docs/archive/HANDOFFS-through-2026-09-16-4.md with its emitted proof. verify.sh exits 0 against the working tree before this commit; check-handoff --all passes on the 2 retained receipts and --archived on the shard's 4. Every commit before this one is already recorded in CHANGELOG.md. --force as the policy states (SRF_RED refuses every on-schedule retention trim by construction, BL-59). The pointer block folds into docs/HANDOFFS_ARCHIVE_INDEX.md in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…2 records to a frozen shard, 36,666 B -> 16,401 B Applies the N=1 retention policy: S178's close-out left 3 receipts and this session's claim made 4. --cut 2 keeps this session's stub and S178's receipt; S177 and S176 move to docs/archive/HANDOFFS-through-2026-09-16-5.md with its emitted proof. verify.sh exits 0 against the working tree before this commit; check-handoff --all passes on the 2 retained receipts and --archived on the shard's 2. Every commit before this one is already recorded in CHANGELOG.md. --force as the policy states (SRF_RED refuses every on-schedule retention trim by construction, BL-59). FRONTMATTER_FIELD_ABSENT is the expected notice (BL-60). The pointer block folds into docs/HANDOFFS_ARCHIVE_INDEX.md in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…2 records to a frozen shard, 33,267 B -> 13,883 B Applies the N=1 retention policy: S180's close-out left 3 receipts and this session's claim made 4. --cut 2 keeps this session's stub and S180's receipt; S179 and S178 move to docs/archive/HANDOFFS-through-2026-09-17.md with its emitted proof. verify.sh exits 0 against the working tree before this commit; check-handoff --all --allow-pending passes on the 2 retained receipts and --archived on the shard's 2. Every commit before this one is already recorded in CHANGELOG.md. --force as the policy states (SRF_RED refuses every on-schedule retention trim by construction, BL-59). FRONTMATTER_FIELD_ABSENT is the expected notice (BL-60). The pointer block folds into docs/HANDOFFS_ARCHIVE_INDEX.md in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…2 records to a frozen shard, 34,289 B -> 17,712 B Applies the N=1 retention policy the file's own header states, authorized by the operator at this session's Phase 0 picker. S182's close-out left 3 receipts and this session's claim made 4; --cut 2 keeps this session's stub and S182's receipt, and S181 and S180 move to docs/archive/HANDOFFS-through-2026-09-17-2.md with its emitted proof. The shard name carries a -2 suffix: SHARD_NAME_DISAMBIGUATED, because docs/archive/HANDOFFS-through-2026-09-17.md was written by S181's trim earlier today. A shard's date is a span label, not a unique key; the earlier shard is untouched. verify.sh exits 0 against the working tree before this commit; check-handoff --all --allow-pending passes on the 2 retained receipts and --archived on the shard's 2. CUT_STRADDLES_DAY and FRONTMATTER_FIELD_ABSENT are the expected notices (BL-60). --force as the policy states (SRF_RED refuses every on-schedule retention trim by construction, BL-59). The pointer block folds into docs/HANDOFFS_ARCHIVE_INDEX.md in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

One row at the bottom of docs/HANDOFFS_ARCHIVE_INDEX.md, block deleted from the ledger. HANDOFFS.md 15,309 B -> 14,853 B. Its own commit by design: inside the trim commit the shipped .verify.sh fails L2 (fork Learning KJ5HST#58). Re-run against f049525 after the fold -- L1, L2/front-matter, L3 all OK, 3 records = 2 retained + 1 archived. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

One row at the bottom of docs/HANDOFFS_ARCHIVE_INDEX.md, block deleted from the ledger. HANDOFFS.md 15,074 B -> 14,618 B. Its own commit by design: inside the trim commit the shipped .verify.sh fails L2 (fork Learning KJ5HST#58). Run against 1273a90 before the fold -- L1, L2/front-matter, L3 all OK, 3 records = 2 retained + 1 archived. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

One row at the bottom of docs/HANDOFFS_ARCHIVE_INDEX.md, block deleted from the ledger's front matter. HANDOFFS.md 14,911 B -> 14,455 B. Its own commit by design: inside the trim commit the shipped .verify.sh fails L2 (fork Learning KJ5HST#58). Run against 4a0914c before this fold: L1, L2/front-matter, L3 all OK, 3 records = 2 retained + 1 archived. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…4,602 B The retention policy (keep ONE, trim above TWO) fell due the moment this session's Phase 1B claim made three receipts, exactly as S201, S202 and S203 each predicted in turn. Depth --cut 2, not the front matter's --cut 1: BACKLOG-DETAIL.md:1995 measured --cut 1 as not viable (bin/tests.sh goes red, 285/14, at one receipt). methodology_trim.py --file HANDOFFS.md --cut 2 --force --write, archiving S202's receipt to docs/archive/HANDOFFS-through-2026-09-20-5.md. Stated and expected: SRF (the force is warranted -- SRF_RED refuses every on-schedule retention trim by construction, BL-59), CUT_STRADDLES_DAY, SHARD_NAME_DISAMBIGUATED and FRONTMATTER_FIELD_ABSENT. L1/L2/L3 OK on 3 records partitioned. The trimmer's ~448 B pointer block is folded into the shard index in its own commit, per the note in this file's front matter (inside the trim commit the shipped .verify.sh fails L2, fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

One row appended to docs/HANDOFFS_ARCHIVE_INDEX.md for docs/archive/HANDOFFS-through-2026-09-20-5.md; the trimmer's 3-line block deleted from HANDOFFS.md front matter. 14,602 B -> 14,146 B. Its own commit by design: inside the trim commit the shipped .verify.sh fails L2 (fork Learning KJ5HST#58). The verifier was run against the trim commit a533938 before this fold -- L1, L2/front-matter, L3 all OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…4,956 B Written by methodology_trim.py v1.5.0 --file HANDOFFS.md --cut 2 --force --write. The claim commit made three receipts, one above the policy depth of two, exactly as S204's handoff predicted; the trim is its own action after the Phase 0 report, never inside Phase 0. L1/L2/L3 OK, P1A_OK 64 -> 65. Stated and expected, not failures: SRF (the retention trim is refused by SRF_RED by construction, BL-59), CUT_STRADDLES_DAY (the cut is positional, three receipts share 2026-09-20), SHARD_NAME_DISAMBIGUATED (-6, the sixth shard ending that day), FRONTMATTER_FIELD_ABSENT (the hand-maintained receipt count was removed deliberately at S172). The trimmer's ~448 B pointer block is left in place here and folded into docs/HANDOFFS_ARCHIVE_INDEX.md in its own commit: inside this commit the shipped .verify.sh fails L2 (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

One row at the bottom of docs/HANDOFFS_ARCHIVE_INDEX.md for docs/archive/HANDOFFS-through-2026-09-20-6.md; the 3-line block deleted from HANDOFFS.md's front matter. 14,956 B -> 14,500 B. Its own commit by design -- inside the trim commit 219ce5b the shipped .verify.sh fails L2 (fork Learning KJ5HST#58). That verifier was run against 219ce5b before this fold: L1, L2/front-matter, L3 all OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…5,110 B methodology_trim.py --file HANDOFFS.md --cut 2 --force --write, after the dry run. S206's claim made three receipts, one above the policy depth of two. Archived S204's receipt into docs/archive/HANDOFFS-through-2026-09-20-7.md. Stated and expected, as at every recent trim: SRF (non-positive boundary), CUT_STRADDLES_DAY, SHARD_NAME_DISAMBIGUATED (-7), FRONTMATTER_FIELD_ABSENT. Shipped verify.sh: L1, L2/front-matter, L3 OK (3 = 2 retained + 1 archived). The pointer block is folded into the shard index in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

One row at the bottom of docs/HANDOFFS_ARCHIVE_INDEX.md for docs/archive/HANDOFFS-through-2026-09-20-7.md; the 3-line block deleted from HANDOFFS.md's front matter. 15,110 B -> 14,654 B. Its own commit by design -- inside the trim commit d387bca the shipped .verify.sh fails L2 (fork Learning KJ5HST#58). That verifier was run against d387bca in a --no-local clone before this fold: L1, L2/front-matter, L3 all OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…lines, not records From the BL-82 measurement: git's merge=union kept real ledger entries whole but fused two real receipts into one block with two session: lines, exit 0, because receipts share their fence lines. Countermeasure: merge two real records from the ledger's history and run its checker; exit 0 is not "whole". 1,142 B. D3 discharged by explicit refusal: KJ5HST#31, KJ5HST#33, KJ5HST#41, KJ5HST#58 and KJ5HST#82 considered; none meets (a), (b) or (c). check-learnings: 76 rows, 15..90, 0 over budget. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…5,670 B methodology_trim.py --file HANDOFFS.md --cut 2 --force --write, after the dry run. S207's claim made three receipts, one above the policy depth of two. Archived S205's receipt into docs/archive/HANDOFFS-through-2026-09-20-8.md. Stated and expected, as at every recent trim: SRF (non-positive boundary), CUT_STRADDLES_DAY, SHARD_NAME_DISAMBIGUATED (-8), FRONTMATTER_FIELD_ABSENT. Dry run and write both reported L1, L2, L3 OK (3 = 2 retained + 1 archived). The shipped verify.sh runs against this commit in a --no-local clone before the pointer block is folded into the shard index, in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The trimmer's 3-line pointer block for HANDOFFS-through-2026-09-20-8.md becomes one row in docs/HANDOFFS_ARCHIVE_INDEX.md and leaves the ledger's front matter (15,670 B -> 15,214 B). Its own commit because inside the trim commit the shipped verify.sh fails L2 (fork Learning KJ5HST#58); that verifier ran against 003a243 in a --no-local clone first: L1, L2/front-matter, L3 OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…4,504 B methodology_trim.py --file HANDOFFS.md --cut 2 --force --write, after the dry run. S208's claim made three receipts, one above the policy depth of two. Archived S206's receipt into docs/archive/HANDOFFS-through-2026-09-20-9.md. Stated and expected, as at every recent trim: SRF (non-positive boundary), CUT_STRADDLES_DAY, SHARD_NAME_DISAMBIGUATED (-9), FRONTMATTER_FIELD_ABSENT. Dry run and write both reported L1, L2, L3 OK (3 = 2 retained + 1 archived). The shipped verify.sh runs against this commit in a --no-local clone before the pointer block is folded into the shard index, in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The trimmer's 3-line pointer block for HANDOFFS-through-2026-09-20-9.md becomes one row in docs/HANDOFFS_ARCHIVE_INDEX.md and leaves the ledger's front matter (14,504 B -> 14,048 B). Its own commit because inside the trim commit the shipped verify.sh fails L2 (fork Learning KJ5HST#58); that verifier ran against c9853b8 in a --no-local clone first: L1, L2/front-matter, L3 OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…4,362 B methodology_trim.py --file HANDOFFS.md --cut 2 --force --write, after the dry run. S209's claim made three receipts, one above the policy depth of two. Archived S207's receipt into docs/archive/HANDOFFS-through-2026-09-20-10.md. Stated and expected, as at every recent trim: SRF (non-positive boundary), CUT_STRADDLES_DAY, SHARD_NAME_DISAMBIGUATED (-10), FRONTMATTER_FIELD_ABSENT. Dry run and write both reported L1, L2, L3 OK (3 = 2 retained + 1 archived). The shipped verify.sh runs against this commit in a --no-local clone before the pointer block is folded into the shard index, in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The trimmer's 3-line pointer block for HANDOFFS-through-2026-09-20-10.md becomes one row in docs/HANDOFFS_ARCHIVE_INDEX.md and leaves the ledger's front matter (14,362 B -> 13,902 B). Its own commit because inside the trim commit the shipped verify.sh fails L2 (fork Learning KJ5HST#58); that verifier ran against adb9d8f in a --no-local clone first: L1, L2/front-matter, L3 OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…5,859 B methodology_trim.py --file HANDOFFS.md --cut 2 --force --write, after the dry run. S210's claim made three receipts, one above the policy depth of two. Archived S208's receipt into docs/archive/HANDOFFS-through-2026-09-20-11.md. Stated and expected, as at every recent trim: SRF (non-positive boundary), CUT_STRADDLES_DAY, SHARD_NAME_DISAMBIGUATED (-11), FRONTMATTER_FIELD_ABSENT. Dry run and write both reported L1, L2, L3 OK (3 = 2 retained + 1 archived). The shipped verify.sh runs against this commit in a --no-local clone before the pointer block is folded into the shard index, in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The trimmer's 3-line pointer block for HANDOFFS-through-2026-09-20-11.md becomes one row in docs/HANDOFFS_ARCHIVE_INDEX.md and leaves the ledger's front matter (15,859 B -> 15,399 B). Its own commit because inside the trim commit the shipped verify.sh fails L2 (fork Learning KJ5HST#58); that verifier ran against 19d9e42 in a --no-local clone first: L1, L2/front-matter, L3 OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…5,895 B methodology_trim.py --file HANDOFFS.md --cut 2 --force --write, after the dry run. S211's claim made three receipts, one above the policy depth of two. Archived S209's receipt into docs/archive/HANDOFFS-through-2026-09-20-12.md. Stated and expected: SRF (non-positive boundary), SHARD_NAME_DISAMBIGUATED (-12), FRONTMATTER_FIELD_ABSENT. Unlike recent trims there is no CUT_STRADDLES_DAY: S209's receipt is dated 2026-09-20 and both retained ones 2026-09-21, so this cut is a day seam. Dry run and write both reported L1, L2, L3 OK (3 = 2 retained + 1 archived). The shipped verify.sh runs against this commit in a --no-local clone before the pointer block is folded into the shard index, in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The trimmer's 3-line pointer block for HANDOFFS-through-2026-09-20-12.md becomes one row in docs/HANDOFFS_ARCHIVE_INDEX.md and leaves the ledger's front matter (15,895 B -> 15,435 B). Its own commit because inside the trim commit the shipped verify.sh fails L2 (fork Learning KJ5HST#58); that verifier ran against 03194df in a --no-local clone first: L1, L2/front-matter, L3 OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…7,369 B methodology_trim.py --file HANDOFFS.md --cut 2 --force --write, after the dry run. S212's claim made three receipts, one above the policy depth of two. Archived S210's receipt into docs/archive/HANDOFFS-through-2026-09-21.md. Stated and expected: SRF (non-positive boundary), CUT_STRADDLES_DAY (all three receipts are dated 2026-09-21, so the shard name is a span label), FRONTMATTER_FIELD_ABSENT. No SHARD_NAME_DISAMBIGUATED: the name is new. Dry run and write both reported L1, L2, L3 OK (3 = 2 retained + 1 archived). The shipped verify.sh runs against this commit in a --no-local clone before the pointer block is folded into the shard index, in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The trimmer's 3-line pointer block for HANDOFFS-through-2026-09-21.md becomes one row in docs/HANDOFFS_ARCHIVE_INDEX.md and leaves the ledger's front matter (17,369 B -> 16,921 B). Its own commit because inside the trim commit the shipped verify.sh fails L2 (fork Learning KJ5HST#58); that verifier ran against d899bd0 in a --no-local clone first: L1, L2/front-matter, L3 OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…6,745 B methodology_trim.py --file HANDOFFS.md --cut 2 --force --write, after the dry run. S213's claim made three receipts, one above the policy depth of two. Archived S211's receipt into docs/archive/HANDOFFS-through-2026-09-21-2.md. Stated and expected: SRF (non-positive boundary), CUT_STRADDLES_DAY (all three receipts are dated 2026-09-21, so the shard name is a span label), SHARD_NAME_DISAMBIGUATED (HANDOFFS-through-2026-09-21.md was taken by S212's trim, so this shard is -2), FRONTMATTER_FIELD_ABSENT. Dry run and write both reported L1, L2, L3 OK (3 = 2 retained + 1 archived). The shipped verify.sh runs against this commit in a --no-local clone before the pointer block is folded into the shard index, in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The trimmer's 3-line pointer block for HANDOFFS-through-2026-09-21-2.md becomes one row in docs/HANDOFFS_ARCHIVE_INDEX.md and leaves the ledger's front matter (16,745 B -> 16,289 B). Its own commit because inside the trim commit the shipped verify.sh fails L2 (fork Learning KJ5HST#58); that verifier ran against aec024f in a --no-local clone first: L1, L2/front-matter, L3 OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…7,555 B methodology_trim.py --file HANDOFFS.md --cut 2 --force --write, after the dry run. S214's claim made three receipts, one above the policy depth of two. Archived S212's receipt into docs/archive/HANDOFFS-through-2026-09-21-3.md. Stated and expected: SRF (non-positive boundary), CUT_STRADDLES_DAY (all three receipts are dated 2026-09-21, so the shard name is a span label), SHARD_NAME_DISAMBIGUATED (HANDOFFS-through-2026-09-21.md was taken, so this shard is -3), FRONTMATTER_FIELD_ABSENT. Dry run and write both reported L1, L2, L3 OK (3 = 2 retained + 1 archived). The shipped verify.sh runs against this commit in a --no-local clone before the pointer block is folded into the shard index, in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The trimmer's 3-line pointer block for HANDOFFS-through-2026-09-21-3.md becomes one row in docs/HANDOFFS_ARCHIVE_INDEX.md and leaves the ledger's front matter (17,555 B -> 17,099 B). Its own commit because inside the trim commit the shipped verify.sh fails L2 (fork Learning KJ5HST#58); that verifier ran against 3740ec2 in a --no-local clone first: L1, L2/front-matter, L3 OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…7,666 B methodology_trim.py --file HANDOFFS.md --cut 2 --force --write, after the dry run. S215's claim made three receipts, one above the policy depth of two. Archived S213's receipt into docs/archive/HANDOFFS-through-2026-09-21-4.md. Stated and expected: SRF (non-positive boundary), CUT_STRADDLES_DAY (all three receipts are dated 2026-09-21, so the shard name is a span label), SHARD_NAME_DISAMBIGUATED (HANDOFFS-through-2026-09-21.md was taken, so this shard is -4), FRONTMATTER_FIELD_ABSENT. Dry run and write both reported L1, L2, L3 OK (3 = 2 retained + 1 archived). The shipped verify.sh runs against this commit in a --no-local clone before the pointer block is folded into the shard index, in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The trimmer's 3-line pointer block for HANDOFFS-through-2026-09-21-4.md becomes one row in docs/HANDOFFS_ARCHIVE_INDEX.md and leaves the ledger's front matter (17,666 B -> 17,210 B). Its own commit because inside the trim commit the shipped verify.sh fails L2 (fork Learning KJ5HST#58); that verifier ran against fc1f05a in a --no-local clone first: L1, L2/front-matter, L3 OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…4,071 B The S216 claim made three receipts; the retention policy (N=1, trim above two; operator, S172) makes this trim owed after the Phase 0 report. Approved at the Phase 0 picker. methodology_trim.py --cut 2 --force, dry run first: S214's receipt to docs/archive/HANDOFFS-through-2026-09-21-5.md, with the expected SRF, CUT_STRADDLES_DAY, SHARD_NAME_DISAMBIGUATED and FRONTMATTER_FIELD_ABSENT notices; L1/L2/L3 OK in-tool. The pointer block is folded into the shard index in its own commit (fork Learning KJ5HST#58). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The trimmer's 3-line pointer block for HANDOFFS-through-2026-09-21-5.md becomes one row in docs/HANDOFFS_ARCHIVE_INDEX.md and leaves the ledger's front matter (14,071 B -> 13,615 B). Its own commit because inside the trim commit the shipped verify.sh fails L2 (fork Learning KJ5HST#58); that verifier ran against 1c1c4e3 in a --no-local clone first: L1, L2/front-matter, L3 OK. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

rmsharp merged commit 6b06fd4 into KJ5HST:main Jul 9, 2026

rmsharp deleted the release/v3.5 branch July 9, 2026 04:59

Navigation Menu

Repository navigation

docs(release): v3.5 — capability-tiered review - #58

docs(release): v3.5 — capability-tiered review#58
rmsharp merged 1 commit into
KJ5HST:mainfrom
rmsharp:release/v3.5

rmsharp commented Jul 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

rmsharp commented Jul 9, 2026

Summary

Test plan

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant