Tranche C — the close made honest · design true · the engine dogfooded#3
Open
mkbabb wants to merge 7 commits into
Open
Tranche C — the close made honest · design true · the engine dogfooded#3mkbabb wants to merge 7 commits into
mkbabb wants to merge 7 commits into
Conversation
Makes B's seven asserted-not-met gates TRUE + biting, each re-verified by a
checked-in, re-runnable instrument that passes (the integrity foundation C
exists to establish). No engine source touched; the design/dogfood waves build
on a now-honest close.
S1 — real <main> landmark (EditorShell.vue): route (a) — <main class="grid
place-items-center place-self-stretch"> occupies the grid's center cell;
display:contents (which stripped both the box AND the implicit landmark role)
is gone, the aria-label band-aid dropped. Byte-identical layout, real landmark
in the a11y tree.
S2 — occlusion gate HARD + controls-OPEN axis (occlusion-gate.mjs +
scripts/lib/demo-driver.mjs): dock-over-content promoted from advisory note to
a failing assertion (content-rect intersection, per-scene dockFloatAllowed);
both controls:{closed,open} axes; verified reddens on KF_OCCLUSION_INJECT=cube.
One real occlusion surfaced (square/mobile) → NAMED W2_PENDING_OCCLUSION
allowance with a self-cleaning stale-check (removal trigger: W2 mobile
work-area/dock-reserve fix), NOT a silent exemption.
S3 — π RM probe + contrast (capture.mjs): --reduced-motion captures 6 frames/
scene over the named window, asserts final-frame non-empty, emits a per-surface
contrast table; the hard rest-frame assertion is gated behind KF_RM_HONORED=1
(default off) — flips true at W3 when the demo honors reduced-motion.
S4 — the >50ms LoAF bench gate (playwright.bench.ts + bench/loaf-scene.html):
the expect(true) stub is gone; the gate drives a 200-cell AnimationGroup
composite, reads window.__kfLoaf, and fails on >50ms main-thread blocking —
the observer's real 2nd consumer (overfitting closed). Wired into ci.yml.
S5 — inv β reconciled honestly (package-lock + W6.md + FINAL.md + ci.yml):
disposition (b) — npm tolerates the dangling optional ../glass-ui link
non-fatally and the library graph never dereferences it; the false
'cleanly skips' / 'glass-ui-absent lockfile' prose corrected to the artefact.
S6 — lighthouse A11y(demo-owned)=100 + SEO≥90 open-panel gate (lighthouse-gate.mjs
+ ci.yml): drives the OPEN-panel editing state (not the splash); HARD on any
a11y audit outside two NAMED allowance buckets (bucket-glassui → ASK-3;
bucket-w2 {image-alt,color-contrast} → W2) + SEO<90. Full A11y=100 binds when
bucket-w2 empties at W2. CI stays green W1→W2.
All gates verified locally green; the three deferred hard-assertions (π RM
rest-frame → W3; A11y=100-full + square/mobile occlusion → W2) are NAMED with
removal triggers, not silently passed. Part of the constellation dock+animation
convergence (HUB/docs/constellation/DOCK-ANIMATION-CONVERGENCE.md).
…tays) The W1 LoAF >50ms bench gate reddened on CI: the shared GitHub VM runs ~6x slower than real hardware, so the 200-cell composite's loop legitimately blocked ~132ms there (local: 20ms) — and a single absolute threshold cannot separate that from the bite-test's 120ms injected block. Fix: CI runs a runner-calibrated 48-cell composite (KF_LOAF_COUNT, still crosses the YIELD_BATCH=32 boundary; loop worst ~30ms, under the UNCHANGED strict 50ms; the 120ms inject still reddens). The full 200-cell yield-stress is the local 'npm run bench' authority. The threshold is NOT relaxed — only the stress size is sized to the runner (cf. B 5fa76b4).
The headline design wave A booked and B re-deferred: the demo stops forking glass-ui's φ type-ladder and speaks ONE design language. Net-deletion adoption of already-shipped glass-ui utilities/tokens. - φ-ladder migration: every instrument-serif + raw display rung across 23 files retired onto the semantic glass-ui ladder (.text-display-4 hero, .text-title/ .text-subheading, .text-heading section titles, bare .dock-label for dock registers). Sweep clean: `grep -rn instrument-serif demo/` (excl dist) = 0; the utils.css alias deleted. Hero raw display rungs = 0. - --font-display formalized to Instrument Serif in style.css @theme (the prerequisite — the ladder fell to Georgia without it; zero new font payload). - --dock-menubar-reserve derived from the dock primitives (fixes the invalid controls-pane max-height calc); --spring-snappy GENERATED by the engine (springLinearStops({response:.35,dampingFraction:.65}) — the demo's spring CSS is an output of the engine it ships); --work-area-vertical-bias replaced with an explicit 0.42/0.58 optical-balance slack pair. - Off-token migrations (S3): CSSCodeEditor cartoon-shadow → .cartoon-surface; SquareScene .square-box fork → canonical .demo-box (theme-aware halo); EasingTarget unscoped global .glass-card leak → component-scoped .easing-target (JS retargeted). The occlusion driver's square subjectSelector updated to .demo-box accordingly. Verified: gh-pages build green, demo typecheck clean, occlusion gate green (the square/mobile allowance carries to W3 — mobile has zero slack, so the fix is work-area SIZING not the vertical-bias), lighthouse open-panel gate green. The spring/desktop color-contrast (bucket-w2) + the ~128-site text-sm/xs leaf tail (F6) are the remaining W2 closes. Hero hint opacity → text-muted-foreground (an a11y close, mute-by-colour). inv-16: only keyframes.js written.
…tion + π full The shop-window now runs on its own engine. No hand-rolled rAF loop survives that a shipped light engine already is; reduced-motion is honored; the cross-scene swap is restored by dogfooding (not the <Transition> B removed for cause). - AnimationVisualizer coast → SmoothProgress (velocity) + SpringProgress seated to the boundary, carried by RAFPlayback.drive (auto-stops on settled). The FRICTION/VELOCITY_EPSILON/coastRafId bookkeeping deleted. - useSpringDemo loop → RAFPlayback.loop (one shared clock, the _gen guard); the re-derived dt seed gone. useEasingDemo ping-pong → NumericAnimation(direction: "alternate") via resolveEasing(name); the dummyAnimation shim deleted. - useRafLoop → a thin reactive skin over RAFPlayback.loop (inherits _gen restart safety). The 2 justified exceptions (Three.js present loop; the 2 one-shot post-paint schedulers) recorded. - Reduced-motion honored: respectReducedMotion:true in defaultAnimationOptions (was absent → false); the cube idle-bob gated in @media (prefers-reduced-motion: no-preference). VERIFIED: the C.W1 π RM probe now PASSES HARD (KF_RM_HONORED=1 — reduced-motion paints its rest frame on cube + spring) — π binds at FULL. - Cross-scene swap restored: SpringProgress (the "smooth" preset) drives an opacity/scale fade on a SIBLING style binding to the keyed <Suspense> host — the async loader UNTOUCHED (no <Transition>/<KeepAlive>, so the blank-scene re-break cannot recur; demo-smoke PASS). 18 lines of dead .scene-* CSS deleted. - inv ζ gate: scripts/proof-dogfood.mjs greps demo rAF against the 3-file justified-exception allowlist + FAILS on any residual (mutation-tested). PASS. - The mobile work-area SIZING fix (the square/mobile occlusion): --work-area-max -height capped at min(64rem, 100dvh − --dock-band-reserve) (cycle-free reserve) so the work-area reserves the dock band. A residual ≥35%-overlap persists on the SMALLEST subject (the optical-split under-reserve closed; the controls-grid starving the row open) — accurately re-documented + booked as a focused square-scene mobile-composition pass; the occlusion gate stays HARD elsewhere. Verified: build + typecheck + demo-smoke (inv γ) + occlusion (inv δ, modulo the named allowance) + proof:dogfood (inv ζ) + the π RM probe (full) all green. inv-16: only keyframes.js written.
… tickDt · fail-explicit total
Finishes the engine unification to its contract-edges. The transposition deletes
divergence (the stepper/playback core is net −20); the boundary detectors + the
proof tests are additive-by-nature (gate-hardening, not alias/legacy).
- S1 — play/drive/loop folded to ONE generation-guarded `_run` core; `drive`
inherited the `_gen` guard it lacked (the unguarded double-schedule class is
now structurally impossible). Public signatures unchanged.
- S2 — ONE canonical `tickDt(dt: ms): number` across every stepper.
SmoothProgress.tick() (frame-dependent no-arg) DELETED; SpringProgress.tick
(seconds) demoted to private `_stepSeconds`; Tickable.tickDt fixed `:void`→
`:number`. The internal seconds callers (springLinearStops/springTimingFunction)
+ the demo dogfood callers (useSpringDemo, AnimationVisualizer — the W3↔W4 seam)
migrated to `tickDt(ms)`. A test asserts tickDt(16) advances the same physical
amount on smooth + spring (no units foot-gun).
- S3 — setColorSpace/setHueMethod joined the fail-explicit `parseOption` seam
(throw AnimationOptionError on a malformed PRESENT value; only `undefined`
defaults) — the contract is now total (was nine-of-eleven). The colorSpace test
used the invalid "srgb" (value.js's sRGB space is "rgb") relying on the old
silent-accept bug → updated to a valid space ("oklab").
- S4 — the default-easing css-twin: VERIFIED then WITHHELD. A single cubic-bezier
CANNOT reproduce the piecewise Penner easeInOutCubic (standard twin drifts
~0.0247; the best symmetric-fit floor ~0.0208 — both above the 1e-2 tolerance,
proven by grid-search). The default stays rAF-only, faithful by omission; the
test is a standing gate (reds if a faithful twin is ever found / an unfaithful
one shipped). The perf lever is a proven "no", not a punt.
- S5 — Timeline._advance dedup (finalizeProgress + the duplicate setTarget gone;
one setTarget, one branch on the time-step variant).
- S6 — proof:boundary residual false-negatives closed (bare side-effect import,
subpath specifier, direct `export const` light export — each reddens the gate
when introduced, self-tested); `rolldown` declared in devDependencies; the CI
demo gate's glass-ui clone PINNED to v3.2.0 (no moving-HEAD reproducibility
hole); the dts byte-check asserts the 15 public symbols.
- S7 — dev.sh/deploy.sh: KILLed with rationale (the demo deploy is a dev-machine
concern; CI already gates the build) — the terminal call, no third punt.
Verified: build:lib + 320 tests (incl. the new S1/S2/S3/S4 gates) + proof:boundary
(hardened) + gh-pages (the seam fixed) all green. inv-16: only keyframes.js.
Closes the two demo-owned a11y leaves S6's bucket-w2 held, so they are now REQUIRED to pass (the gate reddens if either regresses): - color-contrast — the spring scene's .settled-badge was green --color-progress text on its own 14% tint = 1.97:1. Push the badge text toward --foreground (theme-aware: dark in light mode, light in dark) so the green hue survives but the lightness moves to ≥4.5:1 — VERIFIED 0 violations in light + dark via axe color-contrast. spring/mobile + spring/desktop now lighthouse a11y=100. - image-alt — never fired (no demo <img> lacks alt); the precautionary bucket entry removed so a future missing alt bites the gate, not silently. The only remaining lighthouse a11y allowance is bucket-glassui (button-name / label / aria-input-field-name — the glass-ui LabeledField label-association, ASK-3, OUTWARD/inv-16). The full A11y=100 score binds when glass-ui ships it. inv-16: only keyframes.js written.
… 7 overclaims The close ceremony, run under inv ε (the invariant C establishes: the close cannot overclaim). Every gate FINAL records MET resolves to a checked-in, re-runnable instrument shown to PASS. - FINAL.md reconciles B's seven asserted-not-met gates as now-corrected (the a11y landmark, inv δ, π, the harness, the LoAF 2nd consumer, inv β, the A11y/SEO gate) — inv ε's first application; + C's verified gate table + invariants (α hardened, β honest, γ, δ HARD, ε, ζ). - audit/pi.md records π at FULL (the reduced-motion rest-frame probe PASSES HARD with KF_RM_HONORED=1; the contrast table measured) — no tooling-contingency floor, the inv-ε sequencing (W1 authored → W3 honored → flips true) discharged. - audit/DELTA.md pairs the AFTER capture (18 shots, 0 console errors — the checked-in harness re-runs from the repo) with the per-page intended change + the gate evidence; the gate suite is the regression authority. - .changeset/tranche-c.md (major) renders C's published surface (the W4 engine residuals) + folds the unpublished B 3.1.0 changeset — one provenance-signed publish for both. The publish leg (changeset version → tag → release.yml) is user-domain, confirm-first. Tranche C closes: the close honest, the design language whole, the shop-window on its own engine. inv-16: only keyframes.js written.
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
1 participant
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.
Tranche C — keyframes.js' third tranche. Plan: `docs/tranches/C/C.md`.
C's first duty is integrity (inv ε — the close cannot overclaim): make every
B-asserted gate that was asserted-not-met actually TRUE + biting, by a
checked-in, re-runnable instrument. Then the deferred headline (the φ-ladder
design system), the gestalt the shop-window missed (the demo dogfoods its own
engine), and the engine's residual contract-edges.
Part of the constellation dock + animation convergence
(`fourier-analysis/docs/constellation/DOCK-ANIMATION-CONVERGENCE.md`): the
slides spring-dogfood + the glass-ui-AT dock slices are sibling arms.
Waves
0b51cbd. The 7 asserted-not-metgates TRUE + biting, each re-verified locally green: real
<main>landmark;occlusion HARD + controls-OPEN axis + inject-bite; π RM probe + contrast;
the >50ms LoAF bench gate (the observer's 2nd consumer); inv β disposition-(b)
honest prose; lighthouse A11y(demo-owned)/SEO open-panel gate with 2 named
allowance buckets. Three deferred hard-assertions NAMED with removal triggers
(π RM rest-frame → W3; A11y=100-full + square/mobile occlusion → W2).
The npm publish leg (changeset → tag → release) is user-domain, confirm-first.