HAC-341: pre-freeze cleanup — framing, contrast, and one stale name - #28
Merged
Conversation
Four surgical corrections from an external audit. No evidence semantics changed; scene-manifest.json, media/hac-334 and experiments/ are untouched and the view model moves by exactly one label. Cloud L1 rendered `total 105 <= 130` beneath EXECUTED. The value is real — the protected target's own recorded bound — but an inequality against the same bound as the HAC-330 joint constraint, one proof-class switch away from `120 <= 130`, can be chained by a fast reader into "same experiment, confirmed on cloud". That is the reading the claim boundary denies and the one SB-06 works hardest to prevent. The invariant moves to L2 under Protected mutation, with a note that it is not the HAC-330 joint bound. L1 now carries the packet's own caption, which reinforces the EXECUTED vs OBSERVED distinction instead of competing with it. The gate for that asks where `effect.invariant` is consumed rather than blacklisting the current number, so changing the value cannot evade it — and it also fails if the field stops being rendered anywhere, because demoting evidence must not mean deleting it. The adapter labelled hop 6 "ALLOW + authorization receipt" while the decision card and the storyboard both said "ALLOW + receipt". The longer phrase is not false — src/authorization/receipt.ts calls it exactly that — but two names for one artifact on one surface invites reading the longer one as a lifecycle state these packets never emitted. Rendered labels now agree; prose describing the artifact is untouched, including the frozen HAC-333 factualBindings entry. Light COUPLED measured 4.01:1 against the sunken surface, under the 4.5:1 floor for the 8.5-10px label and chip text carrying it. L 0.52 measures 5.15:1 on sunken and 5.48:1 on card; hue and chroma unchanged, dark peer (~8.6:1) untouched. Fixed at the token so the storyboard inherits it. The gate pins the measured lightness and fails if hue or chroma move without a fresh browser measurement, or if the cockpit drifts from the token. Also: cloud L3 links the existing IL-DIAG-012 rather than authoring another architecture artifact, and deliberately not from local L3 — that asset is class B and would be a cross-class leak. Both colour limitations are now documented in-repo rather than only in the PR trail. Five orphaned capture PNGs from earlier passes removed; only IL-COCK-013 changed dimensions (804 -> 817, the new L2 section). 7 new negative proofs. 437 tests, 9 gates, typecheck, build.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
The visual-suite determinism gate caught this, which is what it is for. media/hac-334/bin/build-visual-model.mjs reads media/hac-341/evidence/view-model.json as its source for the cloud path, so renaming hop 6 moved two leaves in the visual model — assets[3] path label and assets[4] node — and the two masters that draw them, IL-DIAG-011 and IL-DIAG-012. Their PDFs and rasters follow. Diffed leaf by leaf before rebuilding: exactly two leaves differ, both the same label. No geometry, no colour, no evidence value moved. The colour token did not reach HAC-334 at all; those masters carry resolved hex. So my earlier claim that media/hac-334 was untouched no longer holds, and the correction is to say so rather than to special-case the gate. A single source of truth means derived artifacts follow it. Also: compare the token's parsed chroma and hue with a tolerance instead of exact float equality (SonarCloud S1244).
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
NO EVIDENCE SEMANTICS CHANGED.
scene-manifest.jsonandexperiments/**are 0 changes. The view model moves by exactly one label.Four surgical corrections from an external contract audit, plus cleanup. No redesign.
P1 — cloud L1 invited HAC-330/HAC-340 stitching
Cloud L1 rendered
total 105 <= 130beneathEXECUTED. The value is real — the protected target's own recorded bound — but an inequality against the same bound as the HAC-330 joint constraint, one proof-class switch away from120 <= 130, can be chained by a fast reader into "same experiment, confirmed on cloud." That is precisely the reading the claim boundary denies and the one SB-06 works hardest to prevent.The invariant moves to L2 → Protected mutation, with revision-before/after and an explicit note that it is not the HAC-330 joint bound. L1 now shows the packet's own caption — "Effect. Distinct from observation." — which reinforces the
EXECUTED/OBSERVEDdistinction instead of competing with it.The gate is semantic, not a blacklist. It asks where
effect.invariantis consumed (via therenderCloudsource region), so changing the number cannot evade it — and it also fails if the field stops rendering anywhere, because demoting evidence must not become deleting it.P1 — light COUPLED under the contrast floor
Measured 4.01:1 against the sunken light surface, under the 4.5:1 floor for the 8.5–10px label and chip text that carries the state.
Hue and chroma unchanged, so the state still reads as the same blue. Dark peer was already ~8.6:1 and is untouched. Fixed at
assets/tokens/colors.cssso the storyboard inherits it. The gate pins the measured lightness and fails if hue or chroma move without a fresh browser measurement, or if the cockpit's re-declaration drifts from the token.P2 — one name for the decision artifact
The adapter labelled hop 6
ALLOW + authorization receiptwhile the decision card and the storyboard both saidALLOW + receipt.The longer phrase is not false —
src/authorization/receipt.tscalls it exactly that, and the audit's original P0 reading was retracted on that basis. But two names for one artifact on one surface invites reading the longer one as a lifecycle state these packets never emitted. Rendered labels now agree. Prose describing the artifact is untouched, including the frozen HAC-333factualBindingsentry — the gate governs rendered labels, not descriptions.P2 — architecture link, and documented limitations
Cloud L3 now links the existing
IL-DIAG-012. No new architecture artifact, no storyboard reordering —seq.architecturealready sits correctly at step 8. Deliberately not added to local L3: that asset is class B and would be a cross-class leak.Both colour limitations are now documented in-repo rather than only in the PR trail: the shared Decision/Effect hue (the grammar has no ALLOW hue, and the only unused candidate is the forbidden authorization green), and the contrast measurement above.
Cleanup
Five orphaned capture PNGs from earlier passes removed — the manifest referenced only four. Only
IL-COCK-013changed dimensions (804 → 817, the new L2 section).Verification
9 gates PASS · 437 tests · typecheck · build. 7 new negative proofs, all confirmed to fire: invariant returning to L1, invariant deleted rather than demoted, decision-hop label drift,
authorization receiptreintroduced in a rendered surface, coupled lightness raised, hue/chroma changed without re-measure, cockpit drifting from the token.Browser-verified on the branch at 1440×900: contrast 5.15:1, cloud L1 free of
105 <= 130, L2 carries it, L3 links architecture, local L3 does not, no console errors.The full viewport/zoom/reduced-motion/keyboard matrix was already run on merged main and is treated as project evidence; the changed surfaces re-ran through the freshness and package gates, which forced the recapture.