From 5eca08688a38f65f76d20d9ca8c6e398c51a1b70 Mon Sep 17 00:00:00 2001 From: bchanot Date: Sun, 1 Nov 2026 16:38:53 +0100 Subject: [PATCH] chore(memory): V1-2.2 closure, mutations rerun on final code, BDR-092, LRN-108, V1-2.3 contract --- .../2026-11-01-v1-2-2-interior-evidence.md | 20 +++++- .../2026-11-01-v1-2-2-mutations-interior.py | 12 ++-- .claude/memory/decisions.md | 12 ++++ .claude/memory/learnings.md | 7 +++ .claude/tasks/TODO.md | 18 +++++- .../contracts/2026-11-01-v1-2-2-interior.md | 2 +- .../2026-11-01-v1-2-3-worker-verification.md | 62 +++++++++++++++++++ 7 files changed, 123 insertions(+), 10 deletions(-) create mode 100644 .claude/tasks/contracts/2026-11-01-v1-2-3-worker-verification.md diff --git a/.claude/audits/2026-11-01-v1-2-2-interior-evidence.md b/.claude/audits/2026-11-01-v1-2-2-interior-evidence.md index 69d495b3..3a6ab85a 100644 --- a/.claude/audits/2026-11-01-v1-2-2-interior-evidence.md +++ b/.claude/audits/2026-11-01-v1-2-2-interior-evidence.md @@ -12,7 +12,7 @@ nonzero (the layer's rule), stroke out. Conservative widest-disc search (disc cr off, arcs add an f32 drift bound), refuses in doubt (0.03 px band measured). Caps: 4 096 vertices, 4 096 cells, 2²⁵ edge distances per sheet, over them refused. Arcs as browsers store them: nonzero radius below 2⁻²⁰ unreadable, endpoints equal in f32 draw no arc, the spread of one-ulp variants added to the deviation and their extremes to the bounds. Codes -`annotation.geometry.interior_too_small` / `verification.validation.marker_interior_too_small` (proposal). +`annotation.geometry.interior_too_small` / `verification.validation.marker_interior_too_small` (approved by the owner 2026-11-01). ## Threshold @@ -29,7 +29,7 @@ Equine at 320 widest disc: caudal 4.951 px, sacral 11.30, thoracic 11.47, cervic test on a 90 % ring, fill only and as shipped; equine zones take a real press. - Chromium touch adjustment: a tap up to ~10 px around a point's disc goes to the point (chromium-320/375, not webkit-375); every equine zone keeps a tappable interior (320: caudal 18/21 … cervical 195/195). -- Mutations: 29 killed, 2 equivalent (Z22 prune, Z25 unreachable). +- Mutations, rerun on the final code `c73415a` (closure): 29 killed, 2 equivalent (Z22 prune, Z25 unreachable); see Closure. - References: no frozen oracle, golden, N-1 copy, seed or PDF digest touched; app 566 incl. PDF suites pass. - Ignored cargo tests: one timing test (`pdf_serialize::l_report_with_real_rasters_costs_what_it_costs`) × 3 feature runs; passed in release (5.2 s). @@ -49,3 +49,19 @@ Equine at 320 widest disc: caudal 4.951 px, sacral 11.30, thoracic 11.47, cervic check:all passed: cargo 2 350 passed, 0 failed, 3 ignored (one timing test × 3 runs); core-wasm 65, supabase 128, app 566; e2e:ui 227 (incl. canvas-clicks 99 and interior-clicks 102, six projects); Docker app image, worker test and image. e2e:shell 86 on the isolated sandbox. WASM 580 428 B gzip -9 native, 580 522 in the app image (margin 19 478 B). + +## Closure (owner brief « Validation définitive V1-2.2 », 2026-11-01) + +- Owner verdict: VERIFIED on `c73415a` (BDR-092, V1-2 contract §17.7). Codes `interior_too_small` / + `marker_interior_too_small` approved. Six LOWs accepted and still tracked (TODO V1-2 section), never declared fixed. +- Verifier history: on `8dd2d78` ECARTS(3), three documentation gaps fixed in `03be75b` (the earlier result is not a + conformity). Fresh verifier on the closed state (`de2dfb3..e190dfc`): I1-I6 and I8 MET, I7 ECARTS(1): Z7, Z19, Z24 + of the mutation script anchored on code that left with `03be75b` / `86dce16`, so the recorded count was not + reproducible on the final HEAD. Fixed: anchors moved to the final code, whole script rerun on `c73415a`: 29 killed, + Z22 and Z25 equivalent. Z23 survived its integration suite there (the `f32` spread of `86dce16` refuses those arcs on + its own) and is killed by the unit test `a_tiny_arc_radius_is_not_read`; the script now runs it on `--lib`. The + final conformity therefore rests on this rerun plus the verifier's other findings (cargo 17 + 19 tests, Playwright + interior-clicks 102 on six projects, WASM 580 428 B native, references diff empty). +- Security: three passes PASS, no MEDIUM open on `86dce16`; `c73415a` = one LOW fix (sampling skipped). +- Gates on `c73415a`: check:all passed; e2e:shell 86 on the isolated sandbox; WASM 580 522 B gzip in the app image. +- Memory: LRN-108 (Chromium touch adjustment, observation only, never a selection contract). diff --git a/.claude/audits/2026-11-01-v1-2-2-mutations-interior.py b/.claude/audits/2026-11-01-v1-2-2-mutations-interior.py index 1e0e107a..b17aeff4 100644 --- a/.claude/audits/2026-11-01-v1-2-2-mutations-interior.py +++ b/.claude/audits/2026-11-01-v1-2-2-mutations-interior.py @@ -1,7 +1,9 @@ # V1-2.2 last MEDIUM (brief « fermeture du dernier MEDIUM », 2026-11-01), mutation run: Z = the # interior rule (anatomy::interior, path outline, frame placement, codes), plus the 2026-11-01 # decision mutants whose anchors this change moved (S1, S8, F16). Z23-Z25 cover the fixes after the -# security review of 8dd2d78. Each mutant is applied to its +# security review of 8dd2d78. Z7, Z19, Z24 re-anchored on the final code c73415a (verifier +# 2026-11-01: their first anchors left with 03be75b and 86dce16); Z23 run on the unit suite since 86dce16, whose +# f32 spread refuses the same arcs on its own in sheet_marker_visibility. Each mutant is applied to its # source file, its cargo suite runs, the file is restored. Run from the repo root. import subprocess F='crates/core/src/anatomy/frame.rs' @@ -17,7 +19,7 @@ muts=[ ('Z4 every point inside',I,' if inside { nearest } else { -nearest }',' if inside || true { nearest } else { -nearest }',VIS), ('Z5 closing edge dropped',I,' for (&a, &b) in closing.zip(subpath.iter()) {',' for (&a, &b) in closing.skip(1).zip(subpath.iter().skip(1)) {','--lib'), ('Z6 deviation not taken off',I,' let radius = clearance(outline, rule, centre) - outline.deviation;',' let radius = clearance(outline, rule, centre);','--lib'), - ('Z7 chord deviation not recorded',P,' self.outline.deviation = self.outline.deviation.max(bend * step * step / 8.0);\n','','--lib'), + ('Z7 chord deviation not recorded',P,' self.outline.deviation = self.outline.deviation.max(deviation);\n','','--lib'), ('Z8 one chord per curve',P,' let count = wanted.clamp(1.0, MAX_CURVE_CHORDS as f64) as usize;',' let count = wanted.clamp(1.0, 1.0) as usize;',VIS), ('Z9 budget never spent',I,' budget.work -= evaluated * edges;\n','','--lib'), ('Z10 vertex overflow ignored',I,' !outline.overflow && self.cells(edge_count(outline)) > 0',' self.cells(edge_count(outline)) > 0',VIS), @@ -29,12 +31,12 @@ muts=[ ('Z16 tolerance in units, not px',F,' let tolerance = INTERIOR_TOLERANCE_PX / scale;',' let tolerance = INTERIOR_TOLERANCE_PX;',VIS), ('Z17 closepath keeps the subpath',P,' self.quadratic_control = None;\n self.open_subpath();\n }',' self.quadratic_control = None;\n }','--lib'), ('Z18 cubic bend underestimated',P,' let bend = 6.0\n',' let bend = 1.0\n','--lib'), - ('Z19 arc bend on the smaller radius',P,' let count = self.chords(ellipse.rx.max(ellipse.ry) * ellipse.sweep * ellipse.sweep);',' let count = self.chords(ellipse.rx.min(ellipse.ry) * ellipse.sweep * ellipse.sweep);','--lib'), + ('Z19 arc bend on the smaller radius',P,' let radius = ellipse.rx.max(ellipse.ry);\n let spread',' let radius = ellipse.rx.min(ellipse.ry);\n let spread','--lib'), ('Z20 unmeasured accepted',F,' return Placement::InteriorUnmeasured;',' return Placement::InFrame;',VIS), ('Z21 cells past the cap',I,' || evaluated + 4 > affordable\n','','--lib'), # Security review of 8dd2d78 (MEDIUM tiny arc radius, LOWs), fixed after the first run. - ('Z23 tiny arc radius read',P,' if tiny(arc.rx) || tiny(arc.ry) {\n return None;\n }\n','',VIS), - ('Z24 arc f32 drift dropped',P,' radius * ARC_F32_DRIFT,',' 0.0,','--lib'), + ('Z23 tiny arc radius read',P,' if tiny(arc.rx) || tiny(arc.ry) {\n return None;\n }\n','','--lib'), + ('Z24 arc f32 drift dropped',P,' radius * ARC_F32_DRIFT + spread,',' spread,','--lib'), # Equivalent: bounds and outline share one reader, a path with bounds always has an outline. ('Z25 path without outline accepted (equivalent)',F,' _ => Placement::InteriorUnmeasured,',' _ => Placement::InFrame,',VIS), # Second security review (03be75b): f32 endpoint equality, arc f32 spread. diff --git a/.claude/memory/decisions.md b/.claude/memory/decisions.md index f57f21a2..fe90a9d4 100644 --- a/.claude/memory/decisions.md +++ b/.claude/memory/decisions.md @@ -686,3 +686,15 @@ Also kept by owner: per-document-instance prefix required on top of per-view pre **Alternatives**: cap on viewBox dims (rejected: bypass by scale); document owning main frame (heavier model change); TS-side check (rejected: Rust by default, one definition). **Extension 2026-11-01** (same rule, owner brief « décisions »): frame usable only within browser-faithful range (edges ≤ 2^20, span ≥ 1, aspect ≤ 100, far edge ≤ 256 spans); marker visible size ≥ 4 CSS px at 320 × 568 → `verification.validation.marker_too_small`. RUST.md §12.11, §12.13. + +## BDR-092 — V1-2.2 VERIFIED: 4 px interior rule, codes approved, six LOWs accepted and tracked (2026-11-01) +- **Status**: accepted +**Decision**: V1-2.2 VERIFIED by owner 2026-11-01 (brief « Validation définitive V1-2.2 »). Zone or drawn trigger (polygon, path) must fill disc of 4 px diameter at 320 × 568 preview, nonzero rule, stroke never counted; conservative measure, compute caps, refusal in doubt. Codes approved: `annotation.geometry.interior_too_small`, `verification.validation.marker_interior_too_small`. Equine accepted unchanged. + +**Rationale**: 4 px = approved point threshold; equine min widest disc 4.951 px (caudal). Conservative (under-estimate) measure never accepts shape browser fills less; browser probe on 3 engines confirms. Evidence `.claude/audits/2026-11-01-v1-2-2-interior-evidence.md`, RUST.md §12.14, V1-2 contract §17.7. + +**Accepted LOWs (tracked, NOT fixed, stay in TODO V1-2 section)**: (1) fully overlapping marker discs: no rule, human review must find them before publication, no claim of individual selectability; (2) arc rotation sub-pixel error, bounded, tests kept; (3) arc tangent to frame edge refused (conservative); (4) f32 arc margin is bounded estimate (≤ 0.13 unit measured on model); (5) verification cost of all-arc sheet (~1.8 s native worst); (6) relative path commands accumulated in f32 (contrived worst 6 % of span). Complex-sheet duration monitored at editor validations on real devices. + +**Alternatives**: larger threshold (6 px) refused equine; renderer-based measure (new dependency, WASM budget); area threshold (thin sliver keeps area). Rejected. + +**Status**: accepted diff --git a/.claude/memory/learnings.md b/.claude/memory/learnings.md index 4de80e17..486795f8 100644 --- a/.claude/memory/learnings.md +++ b/.claude/memory/learnings.md @@ -744,3 +744,10 @@ Supersedes [[LRN-100]]. **Context**: V1-2.2 frame range rules, disc a quarter of frame in 286 px box, painted-pixel count, WebKit 27.2 / Chromium 156 / Firefox 157 (Playwright 1.64.1), 2026-11-01. First span floor 2⁻²⁰ withdrawn on this measure; floor set to 1 unit (factor 32 from WebKit's last exact span 2⁻⁵). **Application**: bound SVG frame spans and offsets by browser f32 range before trusting geometry; derive floors from painted pixels in all three engines, never from f64 math alone. + +## LRN-108 — Chromium touch emulation sends taps ~10 px around point disc to the point; WebKit does not (observation, not contract) (2026-11-01) +**Pattern**: Chromium touch emulation redirects taps near small targets: in V1-2.2 tap grid, taps up to ~10 px around point marker's visible disc went to the point, not to zone under finger. Not seen under WebKit. Experimental, browser-dependent observation, never a selection guarantee. + +**Context**: V1-2.2 interior rule, Playwright 1.64.1 touch projects chromium-320 / chromium-375 vs webkit-375, equine main view, 2026-11-01. Taps at zone witness centre landed on junction points; zone taps recorded 18/21 (caudal) to 195/195 (cervical) at 320. Tool `.claude/audits/2026-11-01-v1-2-2-interior-tools/`. + +**Application**: Chromium touch tests press away from point discs (clearest fill point, margin) or ask `elementFromPoint`. Never use this tolerance as functional contract: explicit selection rules and Rust core guarantees stay authority (owner 2026-11-01). diff --git a/.claude/tasks/TODO.md b/.claude/tasks/TODO.md index 88a91e64..c8f3895d 100644 --- a/.claude/tasks/TODO.md +++ b/.claude/tasks/TODO.md @@ -614,13 +614,27 @@ Order (BDR-071): P0 docs + decisions → V1-1 clients and animals → V1-2 sheet - [x] N4 scale bypass: projected visible size rule (owner 2026-11-01: 4 px at 320 × 568, measured equine min 4.97 px), verdict + preview, closed code; numeric guard (finite, non-degenerate, value/transform bounds, f32 precision vs magnitude and frame), Rust vs browser edge cases - [x] N5 gates: cargo all features, clippy, mutations, wasm size, check:all on final HEAD, e2e:ui, e2e:shell if sandbox, C8 Docker + mutations - [ ] N6 fresh verifier + security on final HEAD (old MEDIUMs, f32/f64 residue); memory BDR + 2 LRN (dedup); checkpoint - - [ ] O last MEDIUM, zone without interior (brief « V1-2.2 : fermeture du dernier MEDIUM », 2026-11-01, on de2dfb3; contract `.claude/tasks/contracts/2026-11-01-v1-2-2-interior.md`) + - [x] O last MEDIUM, zone without interior (brief « V1-2.2 : fermeture du dernier MEDIUM », 2026-11-01, on de2dfb3; contract `.claude/tasks/contracts/2026-11-01-v1-2-2-interior.md`) - [x] O1 measure: flattened outline (polygon, path curves/arcs with a chord deviation bound), fill rule (layer = nonzero), widest inscribed disc search (deterministic, capped), projected at 320 × 568 - [x] O2 rule in `frame::annotation_placement` after the bounds rule; closed codes; verdict + preview; docs (ANNOTATION-RENDERING, RUST.md, V1-2 §6.7) - [x] O3 tests: back-and-forth, near-flat, self-intersection both rules, curves/arcs near zero, small legit, transforms/scales, threshold edges, equine; browser concordance fixture + spec on 3 engines - [x] O4 gates: wasm ≤ 600 000, mutations, check:all, e2e:shell sandbox, ignored tests documented, PDF unchanged; fresh verifier + security - [x] O5 memory: 3 LRN (WebKit snapping, Firefox client rects, WebKit tiny spans); LOW overlap recorded; checkpoint - - [ ] V1-2.3 drawings and worker verification (M6) · V1-2.4 wireframes (editor, review, admin entry) · V1-2.5 catalogue, admin space, rights, species, MFA · V1-2.6 editor level 1 · V1-2.7 preview, review, publication UI · V1-2.8 equine working copy and review preparation · V1-2.9 FR/EN acceptance and closure (owner checkpoints V1-2.3, V1-2.4, V1-2.6, V1-2.7, V1-2.8, V1-2.9 + each DEV operation of contract §16.11) + - [x] V1-2.2 CLOSED: VERIFIED by the owner 2026-11-01 (brief « Validation définitive V1-2.2 », code `c73415a`, BDR-092, contract §17.7); codes `interior_too_small` / `marker_interior_too_small` approved; LRN-108 (Chromium touch, observation only). Final verifier on the closed state: ECARTS(1) (three mutation anchors stale since 03be75b/86dce16) → re-anchored, script rerun on the final code (evidence « Closure ») + - [ ] V1-2.2 accepted LOWs, TRACKED, not fixed (owner 2026-11-01; never mark fixed without a fix and proof): + - [ ] fully overlapping marker discs: no rule; human review must find them before any publication (V1-2.7 review screen, V1-2.8 equine preparation) + - [ ] arc rotation sub-pixel error (bounded, tests kept) + - [ ] arc exactly tangent to a frame edge refused (conservative bounds) + - [ ] `f32` arc margin a bounded estimate (≤ 0.13 unit measured on a model) + - [ ] verification cost of an all-arc sheet (~1.8 s native): watch complex-sheet durations at the editor validations on real devices (V1-2.6) and the worker at-caps timing (V1-2.3) + - [ ] relative path commands accumulated in `f32` (contrived worst 6 % of the span) + - [ ] V1-2.3 worker verification (M6), owner go 2026-11-01 (brief « Validation définitive V1-2.2 »); contract `.claude/tasks/contracts/2026-11-01-v1-2-3-worker-verification.md`; sandbox only, DEV untouched, no merge, STOP at owner checkpoint (server workflow V1-2.1 to V1-2.3) + - [ ] P1 M6 SQL: claim / input / complete (structure check, failed verdict without groups) / fail, closed worker codes, attempt limit, audit; pgTAP allowed + denied; types; sandbox test-db + - [ ] P2 child `verify`: framed stdin input, core verdict, output cap, sandbox (rlimits, seccomp re-derived); spawner limits per command + - [ ] P3 `run --jobs sheet-verify`: backend RPCs + Storage downloads, failure classification §6.5, fake-backend tests per row; compose service under `worker` profile + - [ ] P4 sandbox e2e: pass, each failure, stale/malformed, avatars and purges unaffected, publication gate, working copy in place, at-caps timing; upgrade rehearsal; e2e:avatar unchanged + - [ ] P5 docs, gates (rust:check, worker:test, worker:image, check:all, e2e:shell), fresh verifier + security, memory, checkpoint + - [ ] V1-2.4 wireframes (editor, review, admin entry) · V1-2.5 catalogue, admin space, rights, species, MFA · V1-2.6 editor level 1 · V1-2.7 preview, review, publication UI · V1-2.8 equine working copy and review preparation · V1-2.9 FR/EN acceptance and closure (owner checkpoints V1-2.3, V1-2.4, V1-2.6, V1-2.7, V1-2.8, V1-2.9 + each DEV operation of contract §16.11) ## Release readiness track (owner 2026-10-25, distinct from the functional roadmap; nothing here authorises a release or exposure) Inventory: `.claude/audits/2026-10-25-functional-inventory.md`. Each item: treat or qualify before any release or public exposure; not fixed now unless an active critical flaw is shown. diff --git a/.claude/tasks/contracts/2026-11-01-v1-2-2-interior.md b/.claude/tasks/contracts/2026-11-01-v1-2-2-interior.md index 92bd0b49..2cb4d881 100644 --- a/.claude/tasks/contracts/2026-11-01-v1-2-2-interior.md +++ b/.claude/tasks/contracts/2026-11-01-v1-2-2-interior.md @@ -15,7 +15,7 @@ Owner brief « V1-2.2 : fermeture du dernier MEDIUM », 2026-11-01, on `de2dfb3` radius away from every edge (implicit closing edges included); deterministic search, capped; over the caps the annotation is refused (fail closed), never accepted. - I3 Closed codes, the approved ones unchanged: issue `annotation.geometry.interior_too_small`, - verdict `verification.validation.marker_interior_too_small` (implementer's proposal, to confirm). + verdict `verification.validation.marker_interior_too_small` (implementer's proposal; approved by the owner 2026-11-01, brief « Validation définitive V1-2.2 »). Verdict and preview agree. The document and the drawings never rewritten. - I4 Tests: back-and-forth path, near-flat polygon, self-intersection under both rules, curves and arcs of near-zero area, small legitimate filled shapes, drawing transforms and extreme scales, diff --git a/.claude/tasks/contracts/2026-11-01-v1-2-3-worker-verification.md b/.claude/tasks/contracts/2026-11-01-v1-2-3-worker-verification.md new file mode 100644 index 00000000..ebbbc246 --- /dev/null +++ b/.claude/tasks/contracts/2026-11-01-v1-2-3-worker-verification.md @@ -0,0 +1,62 @@ +# Contract: V1-2.3 worker verification (M6) + +Owner brief « Validation définitive V1-2.2 », 2026-11-01 (V1-2.2 VERIFIED at `c73415a`, BDR-092): +start V1-2.3 along the validated plan, V1-2 contract §16.4 (objective, prerequisites, tests, +acceptance), §6.4 (server verdict), §6.5 (failures), §12.3 (M6). Branch +`feature/v1-2-anatomical-sheets`. Sandbox stack only; DEV and production untouched; no merge; +no V1-2.4; STOP at the owner checkpoint (server workflow V1-2.1 to V1-2.3). + +## Criteria + +- W1 M6 `sheet_verification_jobs`, forward-only, service role only (revoked from `public`, + `anon`, `authenticated`), `security definer`, `search_path = ''`: + `claim_sheet_verifications(limit)` (clamped, `for update skip locked`, rounds `verifying` with no + live lease, token + `locked_at` set, `attempts` incremented, an expired lease counted, the attempt + limit closing the round `verification_failed` / `unavailable`); + `get_sheet_verification_input(review_id, lease_token)` (the frozen document and the base as text, + the referenced drawing rows registered for the sheet: asset id, sha256, byte size, bucket, object + key; nothing for a stale token or status); + `complete_sheet_verification(review_id, lease_token, verdict)` (accepted only while `verifying` + with the live token; binding = the round's sheet, version, revision, review; a passed verdict's + group keys = `general`, the document's view and zone ids and the base's removed ones, its drawings + = the document's asset ids with the rows' sha256; a failed verdict accepted without groups or + drawings, its code from the closed list; else `stale` or `malformed`, nothing written; moves to + `open` or `verification_failed` / `failed`; audited); + `fail_sheet_verification(review_id, lease_token, code)` (closed worker codes; claimable again + below the limit, `verification_failed` / `unavailable` at it; audited). + pgTAP allowed and denied cases; `database.types.ts` regenerated; publication still refused + without an approved round whose passed verdict is current. +- W2 Child `verify`: the same binary, empty environment, rlimits and seccomp installed before + reading stdin; reads one framed input (binding, rows, document, base, drawing bytes) under the + per-command input cap; runs `faunosteo_core` `compute_verdict`; writes the verdict JSON under the + output cap. The parent never parses the document or the drawings. Seccomp allowlist derived again + for this path (Docker `test` stage). +- W3 Spawner per command: input cap, output cap, deadline and memory per command (`decode` + unchanged); `verify` sized on the at-caps fixture (about 13 MiB input). +- W4 Mode `run --jobs sheet-verify`: claim, input, Storage downloads (missing object = drawing + missing for the core, never a worker failure), child, completion; transient, timeout, resource and + crash outcomes reported by `fail_sheet_verification` with a closed code; a broken sandbox stops + the worker. Avatar and purge processing unchanged in the default mode; the verification mode runs + no sweep. RPC body cap of this mode raised to the verdict size. +- W5 Compose service of its own under the `worker` profile (read-only, `cap_drop: ALL`, same + hardening as `worker`); sandbox start of the verification worker; never started on DEV. +- W6 Tests: `pnpm worker:test` (sandbox, seccomp derivation), unit and fake-backend job tests for + each §6.5 row, hostile corpus through the child, e2e on the sandbox stack (clean sheet passes, + each failure and its exit, stale or malformed completion writes nothing, avatars and purges + unaffected, publication refused without a current passed verdict, the sandbox working copy + verified in place with its shared drawing, verification time at the caps within budget), upgrade + rehearsal from the DEV state (Git fixture of the DEV drawing, §12.3), `pnpm e2e:avatar` unchanged. +- W7 Docs describe only what is implemented: V1-2 contract §6.4, §6.5, §12.3, §16.4; STORAGE.md + worker section; RUST.md; SUPABASE.md; DEPLOY.md / compose notes; CLAUDE.md. +- W8 Gates: `pnpm rust:check`, `pnpm worker:test`, `pnpm worker:image`, sandbox `test-db` and + `types --check`, `pnpm check:all` on the final HEAD, `pnpm e2e:shell` on the sandbox; fresh + verifier and security reviews on the final HEAD. + +## Proposals for the owner (checkpoint) + +Closed worker failure codes, attempt limit (five), lease duration, the service name: chosen by the +implementer, listed at the checkpoint for confirmation. + +## Out of scope + +DEV and production (every §16.11 operation), screens, V1-2.4 onwards, merge, deploy.