chore(memory): R5 owner decisions BDR-068, R5.1b/R5.2 contract, TODO, journal, LRN-071, EVAL-027

This commit is contained in:
bchanot committed 2026-10-22 21:24:05 +02:00
1 parent 6afc1e0a37
commit 5e114da2ce
7 files changed
+75 -4

No files matched your search

+8
View File
@@ -509,3 +509,11 @@ Decision: J1 + J2 declared PASS as a whole on iPhone Safari iOS (no per-check sp
Alternatives: wait for per-check iPhone measures (rejected by owner: device-level declaration suffices for merge, measures = follow-up); regenerate goldens on any gap (forbidden).
Recorded: RUST.md §10.15 (risk table + checks), PDF_IMPORT_HARDENING.md §7, R4_J1/J2_RESULTS + MINI_RECETTE, CHANGELOG headers, TODO R4.13.
Status: active; limits stay: no release 0.2.0, no deploy, no DNS/server change, no R5 before next owner signal, D2/D4/D5 deferred.
## BDR-068 | 2026-10-22 | R5 owner decisions: D-R5 = RFC 8785 of validated doc, aliases, F1 fix, byte-exact port, frozen SVG bytes
Context: owner validated R5.1 (« Validation R5.1, décision D-R5 et lancement du portage Rust »).
Decision: D-R5 = option B, convention `sha256-rfc8785-validated-v1` = sha256 of RFC 8785 canonical JSON of the validated document (defaults filled, canonical model), real RFC 8785 (UTF-16 key sort, ES number print, UTF-8, no Unicode normalisation, I-JSON refusals, Rust/JS deterministic) proven by official vectors + edge cases + independent differential; `v1` freezes the construction and validation rules (later defaults/schema never change v1 identity retroactively). Alias mechanism approved: Report v3 `sheet.documentSha256` stays 64 hex; new reports stamp current convention; old reports open with no rebase on identical document; aliases only from trusted frozen historical documents, never from an archive or bundle; unknown fingerprint ≠ official sheet; real anatomical change still migrates; no archive/fixture/oracle rewritten. F1 fix authorised first. Port byte-exact by default, differences declared one by one; fingerprint-bearing artifacts regenerated once at R5.6 only. Historical SVG bytes (horse.svg) kept, never re-serialised. `.prettierignore` on Rust-written generated paths, narrow, with a reproducible check. Upload tool reuses `ureq` (no reqwest by default) at R5.5. Slicing R5.1b to R5.7 validated; stop after R5.2.
Alternatives: A (Rust emitted bytes, re-ties identity to a serialiser), C (Rust reproduces Prettier bytes, keeps formatter chain); one-time visible rebase of every old archive (rejected: transparent alias possible).
Status: active. R5.1b + R5.2 done 2026-10-22; documentSha256, seed, report contracts untouched until R5.6; no merge, no package deletion, no 0.2.0, no deploy.
+7
View File
@@ -198,3 +198,10 @@ Action: owner checkpoint; floor 21 DELETED_TEST (R4.8 deletions, approved in bri
Method: 2 analyzer agents (pipeline inventory 43 files, fingerprint census: producers, consumers, 99 PDFs, docs), both refused to write files (role rule) → orchestrator saved reports; generator test (`MAKE_R5_ORACLES=1`) + freeze re-run green 6/6, anatomy 137, domain 262; Node scripts (matrix, PGlite jsonb) + scratchpad Rust binary on core crate: digests compared value by value; tie census 200k values.
Anomalies: R2 oracle test self-referential on number printing (LRN-070) → F1 undetected since R2; docs contradict (BDR-045 vs RUST.md §10.10 / TODO "will change once") → cross-reference added; analyzer inventory counted "10 anomalies" wrong in orchestrator draft (actual 27, records 825 not 690, 16 ledger artifacts) → corrected from oracle counts before commit.
Action: F1 fix proposed, owner go; D-R5 recommendation B + aliases; checkpoint delivered, STOP.
## EVAL-027 | 2026-10-22 | R5.1b + R5.2 gates: check:all, security, verifier
Method: check:all green 21:16:39 (app 434, core-wasm 58, domain 262, anatomy 141, supabase 40, curation suites, worker + images); security-auditor semgrep 129 rules / 36 files → PASS, 0 HIGH, 5 LOW; fresh verifier on contract `2026-10-22-rust-r5-f1-importer.md` → CONFORME 12/12 (ran cargo workspace + curation suites + import-equine --check itself, red-before proven by rustc probe).
Anomalies: LOW report.rs `Map` index panics on hand-built ImportResult → fixed (get chain) after verdict, importer suite + --check rerun green; LOW quadratic vertebrae scan, symlink-following `listed()` in --check, lint allows (accepted, operator inputs). check:all ran on pre-commit tree (code identical to 03bdb5e). Verifier floor note: plan B.3.5 "(proposal, not implemented)" heading from R5.1, intended.
Action: checkpoint R5.2 delivered, STOP before R5.3.
+4
View File
@@ -322,3 +322,7 @@ Owner: J1 + J2 PASS on computer, Android, real iPhone Safari iOS (declared as a
## 2026-10-22 (R5.1 freeze + D-R5 study)
Owner: R4 bilan validated, R5.1 authorised (freeze TS pipeline, study D-R5, stop before Rust pipeline + fingerprint change). Branch `feature/rust-r5-pipeline-core` from 9c0117a. Two analyzers (inventory, documentSha256 census) → `.claude/audits/R5_*.md`. Oracles `oracles/r5/{pipeline,ids,fingerprints}.json` via `r5.pipeline-oracles.test.ts` (MAKE_R5_ORACLES=1; 6 cases, determinism two runs, ledger vs committed). Study scripts `scripts/r5-fingerprints.mjs`, `packages/supabase/scripts/r5-jsonb-fingerprint.mjs`; throwaway Rust experiment (scratchpad) reproduced every draft digest byte-for-byte, JCS 40 lines no crate, RFC vectors pass; jsonb round trip keeps B only. Found F1: `js_number` tie rounding (Grisu up vs ES6 even; `840.1692504882812`, 7.4 % of float32 values; `zmij` matches) hidden by self-referential R2 oracle test → LRN-070, EVAL-026, not fixed (owner go). Plan `2026-10-22-rust-r5-pipeline-core.md`, RUST.md §11, TODO R5.1 ticked. Recommendation: B on validated form + alias mechanism (old `27c7…` stays valid, no visible migration). STOP at checkpoint. No Rust pipeline, no fingerprint change, no merge.
## 2026-10-22 (R5.1b F1 fix, R5.2 legacy importer in Rust)
Owner validated R5.1, decided D-R5 = B `sha256-rfc8785-validated-v1` + aliases, F1 go, byte-exact port, frozen SVG bytes, ureq at R5.5 → BDR-068. R5.1b aed6b1e: `js_number` digits from `zmij` (edge only, no new crate), V8 oracle 44,322 rows (1,552 mismatches before → 0), `zod_oracle` now checks oracle bytes (red on old printer), rust:check + wasm + check green, no oracle/golden/digest moved. R5.2 03bdb5e: `crates/core/src/curation` (feature `curation`, 3,187 lines: Prettier string width table transcribed, JSON layout rules, table alignment, slugify via icu_normalizer NFD, uuids, legacy importer, IMPORT-REPORT) + `import-equine --check|--out`; legacy-import/ byte-exact (6 files), pipeline.json legacy parts equal, 10 synthetic TS cases, 968 Prettier probes; declared refusals: emoji width, malformed SVG, wrong-typed catalog nodes. Docs 82f1664 (RUST.md §11.9-11.11, CHANGELOG), contract `2026-10-22-rust-r5-f1-importer.md`. Gates: check:all green 21:16:39, security PASS (5 LOW, report Map-index panic fixed), verifier CONFORME 12/12 → EVAL-027. STOP after R5.2 checkpoint.
+5
View File
@@ -524,3 +524,8 @@ Apply: when an implementation is swapped under a test that spied the old one, re
Found by R5.1 study. `js_number` (core json.rs) takes shortest digits from `format!("{:e}")` (Grisu): exact tie between two shortest candidates → Grisu picks upper, ECMAScript/Python/`zmij` pick even. `840.1692504882812` (= 840.16925048828125, float32-exact, Horse.svg legacy cx) → Rust `…813`. Rate: 7,401 / 100,000 float32-exact values in 0..2000, 769 / 100,000 random doubles, `zmij` (serde_json 1.0.154 formatter, locked) 0. Draft seed has no tie → every R2/R3/R4 suite green, but `sheet_contract.rs` `zod_oracle` computes expected = `to_js_pretty(oracle.output)` → same function both sides → blind. 134 JSON fixtures Node vs Rust pretty: 2 differ (legacy baseline, that value).
Apply: shortest-float port → take digits from `zmij`/ryu-class crate, never `{:e}`; tie vectors (float32-exact values) in unit tests; oracle expected must be the recorded JS text, never re-printed by the port under test. Any R5 importer emitting legacy coordinates needs the fix first.
## LRN-071 | 2026-10-22 | Prettier byte parity for generated JSON/Markdown is cheap: four layout rules + one width table, oracle-probed
R5.2 reproduced Prettier 3.8.3 `parser: json` over `JSON.stringify(v, null, 2)` and markdown table alignment in ~350 lines Rust, byte-exact on 968 synthetic probes + 7 seed files + 3 reports. Rules that mattered: objects always expanded (stringify newline after `{`), array breaks when >1 element and all elements arrays >1, or child breaks, or flat form + trailing comma > 80 cols; broken all-number arrays use `fill`, others one per line; `printNumber` drops `e+`; width = length if all bytes 0x20..0x7F else per code point (controls/combining/VS16 = 0, East Asian wide/fullwidth = 2) after RGI emoji replaced by spaces. Emoji table not reproduced → refuse (`UnsupportedWidth`), declared. Probe generator in TS (vitest, `MAKE_R5_ORACLES=1`) sweeping depth × key width × count × digits around the 80-col boundary found both bugs (`e+21` form, width fast path) before any real file did.
Apply: any port that must match a formatter's bytes: sweep the boundary with synthetic probes written by the formatter itself; transcribe tables from the installed package (same table lives in `index.mjs` and plugin), never from memory; refuse unreproduced classes loudly instead of approximating.