docs(lineage): three §3a rows from the 2026-09-09 harvest, the Zoekt row updated, and readmedriftcheck E2b — every README copy of the pair must agree - #86
Conversation
…deburn, markitdown — and the Zoekt row brought up to date
Each row names the lesson and the file it lives in, per the table's own honesty rule. tgrep: a search tool must not make
you learn a second grep (line anchors, line-at-a-time scanning, the libc++ multiline fault), and state what a cut COST
(corpus_pruned_dirs=); the crossover priced and both halves of the margin refuted. codeburn: measure the thing you claim
to optimise in the unit you claim it in, and mark estimates — the tokenaudit instrument, the loop ledger, and the
negative lesson that cost is the wrong unit. markitdown: state the privilege model of every shell-out — applied to
ripwire's own git calls (githarden), and honestly surveyed-not-folded for the prose tier. The Zoekt row's "where it
lives" had been false since P3 removed the posting-list index on 2026-07-27; it now says what survives (the per-file
prefilter) and points at the tgrep row for the resident form's price.
Counts: 43 -> 46 repositories at every spelling readmedriftcheck derives — nine sites, not eight: LINEAGE's disjointness
sentence carries the number twice ("46 folded plus 237 surveyed — not 46 picked out of 237"), and arm E10 caught the
second one.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…gree, with a second-copy mutation control (E2) extracted the FIRST "<M> repositories and <P> papers" and stopped, and README prints the pair twice (the <summary> line near the top, the bolded sentence in the honesty section ~1,600 lines down), so the second copy was a published number with no instrument on it — the merge-clean-but-wrong shape the 2026-09-09 landing round met four times. (E2b) takes `sort -u` over ALL matches and asserts one distinct pair (a future third copy that disagrees fails too; the copy count is reported, never pinned). (E2c) mutates ONLY the second copy in a temp file — the one (E2) can never see — asserts the mutation took, and re-runs the identical extraction; it must report two distinct pairs. Observed RED from a bash run against a tree whose second copy drifted: (E2) green, (E2b) red. An arm on an existing gate: the gate count does not move. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
Caution Review failedThe pull request is closed. ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Advanced Run ID: 📒 Files selected for processing (4)
Cache: Disabled due to data retention organization setting Knowledge base: Disabled due to 📝 SummarySummary by CodeRabbit
WalkthroughThe change updates repository lineage counts from 43 to 46 across documentation and presentation text. It expands lineage entries for three repositories, revises the Zoekt description, and adds checks for inconsistent repeated README counts. ChangesLineage count and validation
Estimated code review effort: 2 (Simple) | ~10 minutes Suggested reviewers: ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
The last landing of the 2026-09-09 harvest round. Documentation only — no gate added, no source touched.
Three §3a rows for the repositories this round studied: tgrep, codeburn, markitdown. Each row records what was taken and what was refuted, which is the point of the ledger — a tool that gave a lesson is named, and so is one that gave none.
The Zoekt row was stale and is brought up to date.
43 → 46 repositories, derived at all nine spellings.
readmedriftcheckarm E1 counts the §3a body rows and re-derives every published spelling against that count — README's pair, the "46 folded plus 237 surveyed" restatement, the wrapped second number in that same sentence, LINEAGE's bold header triple, and both of the deck's long and short forms. E6 proves §3a and §3b stay disjoint with a planted control, so the folded and surveyed sets add rather than nest. E8 proves all 88 cited paths exist on the tree — which is why this lane could not land until tgrep's files were on main.A new arm, and the reason it exists. While deriving the nine spellings, the lane found a site its own gate did not cover:
README.mdcarries the lineage pair twice (lines 21 and 1648) and E2 read only the first. It reported that as hand-checked rather than shipping it under a green E2 — eight of nine derived, one named as not.E2bnow takessort -uover every<M> repositories and <P> papersmatch and asserts a single distinct pair, reporting the copy count without pinning it, so a future third copy that disagrees also fails.E2cis the control, and it is the part worth having: it mutates only the second copy and requires(E2)to stay green while(E2b)goes red — reproducing the old blindness rather than merely proving the new arm fires. Observed red from bash both ways.Verified before landing: tip contains current main; CI run 34407178796 26/26 green on the exact tip; gate count 569, main's own and untouched — this lane adds no gate; lineage count derived on the final tip, not carried;
qschemetripcheckgreen so untouched;ripwirepubliccheckarms 1/2/5 clean.The rebase over #77 stopped once, on
present/deck5_ripwire_build.js, where #77's gate-count line sits adjacent to this lane's lineage line. Resolved by keeping both claims with both instruments beside them:Closes the harvest round's four landings: codeburn (#79), githarden + textdocs (#80), tgrep (
a2973177) and this.🤖 Generated with Claude Code