Skip to content

Repository navigation

LEDGER.md

Latest commit

 

History

History
3288 lines (3263 loc) · 234 KB

LEDGER.md

File metadata and controls

3288 lines (3263 loc) · 234 KB
Raw

Out-of-Band Ledger

Some layers of the platform are allowed to lag behind feature work: the primary development pass ships the feature, and a separate catch-up pass tops these layers off later. This file is the checkpoint record for those passes.

Each section records:

  • Checkpoint — the commit (SHA, date) through which the category is known to be caught up. "Caught up through X" describes what was audited, not where the fix landed. Cross-repo categories list one checkpoint per repo.
  • Scope — what the pass owns.
  • Gap detection — how to compute outstanding work since the checkpoint.
  • Log — one line per completed pass, newest first.

Running a catch-up pass: compute the gap over <checkpoint>..HEAD, do the work, verify it, then update the checkpoint and append a log line.

I18n strings

  • Checkpoint: noisemaker ed47815 / noisedeck c674516 (2026-10-01)
  • Scope: two translation surfaces:
    • Noisemaker effect catalogs shaders/effects/strings.{de,es,fr,it,ja,pt}.json. The English catalog strings.en.json is generated from effect definitions (npm run strings) and drift-tested by npm run test:shaders:i18n.
    • Noisedeck UI catalogs under app/js/i18n/locales/{de,es,fr,it,ja,pt}/. The matching en/ slices are the hand-authored source of truth. Missing locale keys fall back to English at runtime, so gaps on both surfaces are invisible without the checks below.
  • Gap detection:

    1. Noisemaker — every key present in strings.en.json and absent from a locale file needs a translation. From shaders/effects/:

      node -e 'const fs=require("fs");const en=Object.keys(JSON.parse(fs.readFileSync("strings.en.json","utf8")));for(const l of["de","es","fr","it","ja","pt"]){const t=new Set(Object.keys(JSON.parse(fs.readFileSync("strings."+l+".json","utf8"))));const m=en.filter(k=>!t.has(k));if(m.length)console.log(l+":",m.join(", "))}'

      Keys stay in the same (sorted) order as the English catalog. Match each locale's existing conventions — parameter labels lowercase, effect names capitalized, description tone per locale — and reuse the file's existing translation for a term before inventing a new one. Run npm run test:shaders:i18n to enforce exact key order and non-empty values across all six translated catalogs, in addition to generated-English drift and runtime fallback behavior.

    2. Noisedeck — from its repository root, run the catalog-parity test. Its diff names every missing or extra leaf key by locale:

      node --test --test-timeout=60000 tests/i18n.node-test.js

      Add keys to the matching area slice in English order. The test enforces non-empty string values plus every placeholder, plural leaf, and markup tag from the English catalog.

  • Log:
    • 2026-10-01 — caught up through noisemaker ed47815 / noisedeck c674516: audited noisemaker range f24b525..ed47815 (zero diffs in shaders/effects/; all 3,677 keys remain in exact parity across all seven catalogs) and noisedeck range de9c3ba..c674516 (five commits: 28d5a3f "Match loop duration to loaded video clips" added media.loopLimit and media.matchLoop, translated in all six locales in the same commit; the remaining four commits touched no catalog strings). The delivery trigger's end SHA 9d5a61a was already an ancestor of the prior checkpoint de9c3ba, so nothing new was assumed from the forced range. Verified via Noisemaker gap detection (0 missing keys across all 6 locales), npm run test:shaders:i18n (5/5), the Noisemaker JS test suite (all files pass, 0 failures), Noisedeck tests/i18n.node-test.js (6/6), tests/export-dialog-i18n.node-test.js, tests/menu-shortcuts-accelerators.node-test.js, and tests/downstream-workflow.node-test.js (13/13 combined), and tests/gift-menu.spec.js on Chromium (2/2; Firefox/WebKit do not launch in this sandbox — the same spec run fails 10/10 on an unrelated spec too, an environment limitation, not a catalog defect). No catalog edits were required.
    • 2026-09-30 — caught up through noisemaker f24b525 / noisedeck de9c3ba: audited noisemaker range 6827390..f24b525 (zero diffs in shaders/effects/; all 3,677 keys remain in exact parity across all seven catalogs) and noisedeck range 69eadc5..de9c3ba (six commits: f48a5c5 "Group app menu actions into nested submenus" added six UI string keys — menus.app.help, menus.app.account, menus.file.export, menus.file.customEffects, menus.program.editIn, menus.program.importExport — translated in all six locales in the same commit; 9d5a61a "Keep canvas-related dialogs within the controls region" added export.common.upgrade, also translated in all six locales in the same commit; the remaining four commits touched no catalog strings). Verified via Noisemaker gap detection (0 missing keys across all 6 locales), npm run test:shaders:i18n (5/5), the Noisemaker JS test suite (200/200 passed across 9 files), Noisedeck tests/i18n.node-test.js, tests/export-dialog-i18n.node-test.js, tests/menu-shortcuts-accelerators.node-test.js, and tests/downstream-workflow.node-test.js (19/19 combined). No catalog edits were required.
    • 2026-09-29 — caught up through noisemaker 6827390 / noisedeck 69eadc5: audited noisemaker range 5b81e04f..6827390 (zero diffs in shaders/effects/; all 3,677 keys remain in exact parity across all seven catalogs) and noisedeck 9eabfb2..69eadc5 (the delivered trigger 0a2a7012..69eadc5 was force-pushed and 0a2a7012 is unreachable; the observed delivered range is the single commit 69eadc5). Its only UI string addition, menus.giftNoisedeck, was already translated across all six UI locales in the same commit. Verified via Noisemaker gap detection (0 missing keys across all 6 locales), npm run test:shaders:i18n (5/5), Noisedeck tests/i18n.node-test.js (6/6), and tests/standalone-account-menu.node-test.js (12/12). No catalog edits were required.
    • 2026-09-23 — caught up through noisemaker 5b81e04f / noisedeck a5d448b2: audited noisemaker range dd38fdd2..5b81e04f and noisedeck 697d2df9..a5d448b2 (covering work queue item #425 trigger 0a2a7012..df632d0a). In noisemaker, zero effect definitions or parameters were added or removed, leaving all 3,677 keys in exact parity across all seven catalogs (en and de, es, fr, it, ja, pt). In noisedeck, sync.previewPaused was added to English (34295fbf) and translated across all six UI locales (55fa1a24), merged in df632d0a; all UI string leaf keys remain in exact parity. Verified via Noisemaker gap detection (0 missing keys across all 6 locales), npm run test:shaders:i18n (5/5), Noisemaker JS test suite (63/63 passed), Noisedeck tests/i18n.node-test.js (6/6), tests/export-dialog-i18n.node-test.js (2/2), tests/menu-shortcuts-accelerators.node-test.js (7/7), and tests/downstream-workflow.node-test.js (1/1). No catalog edits were required.
    • 2026-09-22 — caught up through noisemaker dd38fdd2 / noisedeck 697d2df9: audited noisemaker range 7ea31be3..dd38fdd2 (covering work queue item #392 trigger 2f855c9c..36a519a2 through HEAD) and noisedeck 4e412eb5..697d2df9. In noisemaker, commit 36a519a2 added render/renderLandscape3d.filtering, render/renderLandscape3d.filtering.isosurface, and render/renderLandscape3d.filtering.voxel across all seven effect catalogs (en and de, es, fr, it, ja, pt), bringing the active catalog to 3,677 keys in exact parity. Noisedeck had zero UI string additions or deletions across the range. Verified via Noisemaker gap detection (0 missing keys across all 6 locales), npm run test:shaders:i18n (5/5), Noisemaker JS test suite (65/65 passed), Noisedeck tests/i18n.node-test.js (6/6), tests/export-dialog-i18n.node-test.js (2/2), tests/downstream-workflow.node-test.js (1/1), and tests/menu-shortcuts-accelerators.node-test.js (7/7). No catalog edits were required.
    • 2026-09-21 — caught up through noisemaker 7ea31be3 / noisedeck 4e412eb5: audited noisemaker range 1086890d..7ea31be3 (covering work queue item #312 trigger c68fb3c8..2f855c9c) and noisedeck e1146981..4e412eb5. In noisemaker, commit 2f855c9c retired filter/bc, filter/colorspace, and filter/hs, removing their 15 effect string definitions across all seven catalogs (en and de, es, fr, it, ja, pt), bringing the active catalog to 3,674 keys in exact parity. Noisedeck had zero UI string additions or deletions across the range. Verified via Noisemaker gap detection, npm run test:shaders:i18n (5/5), Noisedeck's tests/i18n.node-test.js (6/6), export-dialog-i18n (2/2), downstream-workflow (1/1), and the Noisemaker JS test suite (57/57 passed). No catalog edits were required.
    • 2026-09-20 — caught up through noisemaker 1086890d / noisedeck e1146981: audited noisedeck range 73538a8f..e1146981 (covering work queue item #293 trigger adef85dd..0a2a7012) and noisemaker da1b6985..1086890d. UI string additions in noisedeck (startup.statuses.enabled, startup.statuses.midiNoDevices, startup.statuses.midiPortFailed in 0a2a7012) were already translated across all six UI locales (de, es, fr, it, ja, pt); all 3,689 Noisemaker effect strings remain in exact key parity with zero missing keys across all six translated catalogs. Verified via Noisemaker gap detection, npm run test:shaders:i18n (5/5), Noisedeck's tests/i18n.node-test.js (6/6), export-dialog-i18n (2/2), startup-backend (13/13), and the Noisemaker JS test suite (58/58 passed). No catalog edits were required.
    • 2026-09-17 — caught up through noisemaker da1b6985 / noisedeck 73538a8f: audited noisedeck range 47b145e0..73538a8f (covering work queue item #194 trigger adef85dd) and noisemaker c68fb3c8..da1b6985. The only UI string addition in the range (menus.app.downloadApp in adef85dd) was already translated in all six UI locales (de, es, fr, it, ja, pt); all 3,689 Noisemaker effect strings remain in exact key parity with zero missing keys across all six translated catalogs. Verified via Noisemaker gap detection, npm run test:shaders:i18n (5/5), Noisedeck's tests/i18n.node-test.js (6/6), export-dialog-i18n, standalone-account-menu, and the Noisemaker JS test suite (58/58 passed). No catalog edits were required.
    • 2026-09-16 — caught up through noisemaker c68fb3c8 / noisedeck 47b145e0, closing the gap left by incomplete work queue item #156 (below) as well as this task's own nominal range (noisemaker ff1bfbc1..c68fb3c8, noisedeck 458f7758..47b145e0) — audited from the stale checkpoint forward rather than trusting #156's own narrative, per this file's "audit the target's actual state" rule. Both gap-detection commands came back clean: the noisemaker key-diff found zero missing keys in any of the six locales, and noisedeck's tests/i18n.node-test.js passed 6/6 with no catalog-parity failures. Spot-checked that the round's three new effects (synth3d/heightmap3d, render/renderLandscape3d, points/heightGrid) already carry real, non-empty translations across all six noisemaker locales (e.g. synth3d/heightmap3d → "Höhenkarte 3D" / "Mapa de alturas 3D" / "Carte de hauteur 3D" / "Mappa di altezza 3D" / "3D高さマップ" / "Mapa de altura 3D"), not just present-but-empty stand-ins. npm run test:shaders:i18n passed 5/5. No translation work was needed on either surface — #156's own catalog work (3,689 English keys, all six locales complete) was already correct and complete; only this checkpoint had never been advanced to reflect it. #156's Log entry below is left untouched; its remaining non-i18n items (browser/runtime fixes) are outside this section's scope.
    • 2026-09-12 — work queue item #156 remains incomplete. Audited Noisemaker 7d6cc7127457..ff1bfbc1740c and Noisedeck 94f4d0021133..458f77583394. Removed four obsolete landscape-control keys from all seven effect catalogs. All six translations contain all 3,689 English keys. Noisedeck requires no catalog changes. Corrected Noisedeck's audio fixtures, cache teardown handling, browser-specific clipboard permissions, and Remap test interactions. Bounds checks use the accepted pointer coordinates and still require exact uniform values. Its latest Node suite reported 873 passes, one existing conditional skip, and no failures. All 40 Chromium audio tests passed. Affected browser checks reported 120 passes and three existing skips across Chromium, Firefox, and Linux WebKit with a virtual display. The WebKit viewport test exceeded its 45-second budget with two CPUs, then passed in 42.4 seconds with four CPUs and unchanged assertions. The full browser suite then exposed a Noisemaker state-expression serialization defect. The compiler and unparser fix passed the language, runtime, i18n, prescribed JavaScript, and parity-evidence checks. Browser replay of seed 110457923 passed all 500 mutations against the fixed source, with no skipped mutations. The published engine still needs this fix before the full browser suite can complete. Both I18n checkpoints remain unchanged. The operator authorized autonomous publication. The compiler fix is pushed as c68fb3c8. Shader CI exposed the same Linux landscape pixel failure as ff1bfbc1. A same-size WebGPU presentation copy was interpolating adjacent texels. A focused GPU regression failed before selecting the nearest sampler for same-size copies, then passed on Mac and Linux. Scaled copies still use linear filtering. The Linux landscape camera test now passes exact cross-backend parity. Setting the existing test resolution before the first run also resolved the controls audit timeout; all 100 Linux audit cases passed. Mac camera parity, runtime, JavaScript, and all registered parity evidence checks passed. The follow-up fix is pushed as 9a31314c. Its CI camera checks passed, but the controls audit timed out on WebGPU. A second CI run with diagnostics (b02b42ec) instead timed out during initial editor loading. The audit now starts with a simple program, selects the required 256x256 resolution, and uses a separate 60-second compilation wait. Its five-second UI action limit and exact pixel checks remain unchanged. This test fix is pushed as 5fea6395; all 100 local Linux audit cases passed. Its CI run was canceled after repeated control timeouts; it did not pass. Further harness inspection found that loading the reference page started an unrelated asynchronous homepage renderer. The reference now uses an empty same-origin fixture. The editor selects its test resolution before its first frame and stops before each program change. A full audit of these changes is in progress. That audit passed 69 cases, then rapid WebGPU mode changes settled on the older selection. A completed compile callback was restoring its old DSL state while a later compile was queued. The UI now lets only the latest request restore controls and report completion. A deterministic regression failed before the fix and passed afterward. All 100 browser audit cases then passed with unchanged source hashes. Language, runtime, i18n, and lint checks passed. The fix is pushed as bb158566. Its CI passed the earlier gates, then timed out waiting for screenshot stability within five seconds. The capture now brings the editor forward and has a separate 30-second deadline; its pixel checks and control-action limits are unchanged. All four focused presentation cases passed. This harness fix is pushed as 6f60c815. Its CI reached the same screenshot stability timeout at 30 seconds. The audit now captures the visible canvas rectangle and explicitly verifies unchanged geometry across capture. All four presentation cases passed. This change is pushed as f37bd787; exact-commit CI and shader release verification remain pending. A Noisedeck browser preflight passed 211 cases before exposing obsolete startup-carousel assertions. Startup now shows static resource slides; its tests now verify paging and no preview canvases. The gallery retains its live-context checks. All 15 focused cases passed across Chromium, Firefox, and Linux WebKit. Full released-engine validation is pending. A later preflight exposed a cached HTTP 429 for a Godot kit file that currently returns HTTP 200. The shared test cache now stores only successful responses and lets waiting requests recover after an uncacheable error. All five cache regressions passed, including stale-error recovery and lock contention. The real Godot archive-cancellation test then passed.
    • 2026-09-18 — verified the state left by the entry above. All seven noisemaker fixes (c68fb3c8 through f37bd787) are merged to main; CI has passed on every push since, most recently 688c5146. The noisedeck fixes (origin-cache eviction of uncacheable responses, disposed-response handling after page close, browser-specific clipboard grants, pointer-rounding tolerance in the remap bounds checks, and the resource-carousel test rewrite) are committed on preview at c48fe396: full node suite 1046/1047 (one pre-existing skip), Chromium 73/73 (one pre-existing skip), Firefox clipboard fix passing. WebKit did not run locally — this machine's Playwright WebKit build is frozen and rejects a protocol setting Playwright now sends, unrelated to the change — so it still needs its usual Linux CI pass once pushed. This closes out the remaining non-i18n items from the 2026-09-12 entry; nothing from that entry is still open.
    • 2026-09-07 — caught up through noisemaker d30d1045 / noisedeck 94f4d002: audited all 20 live-capture settings, status, permission, recording-limit, and export-warning strings; all six UI translations were already complete. All six effect catalogs retain all 3,639 English keys. Generated-English drift, key order, placeholders, markup, fallback, and locale checks passed, along with Noisedeck's full 806-test Node suite and both affected toolbar/settings browser tests. No catalog edits were required.
    • 2026-09-07 — caught up through noisemaker d30d1045 / noisedeck 00cb87d2: all six effect catalogs contain all 3,639 English keys. Audited the new Remap editor and MIDI CC, NRPN, MPE, and audio-channel copy plus the removed Sync retry string; all six UI locales were already complete. Generated-English drift, exact key order, non-empty values, placeholders, markup, fallback, and async locale checks passed, as did Noisedeck's full 806-test Node suite. No catalog edits were required.
    • 2026-09-03 — caught up through noisemaker 691eea16 / noisedeck 0f3917fb: verified that the updated Cell description is present in all six translated effect catalogs and that Noisedeck's insert-zone, MIDI-device, and audio-device strings are translated in all six UI locales. Revalidated exact key order, non-empty values, placeholders, markup, generated-English drift, runtime fallback, and async locale selection; no catalog edits were required in this pass.
    • 2026-08-31 — caught up through noisemaker c38316a1 / noisedeck 114d64f7: Noisemaker's generated English catalog and all six translated effect catalogs remained complete. Noisedeck's translations for unloaded user effects, storage save failures, and paid-feature load failures were present in all six locales; revalidated exact key parity, non-empty values, placeholders, markup, and async locale selection. No catalog edits were required in this pass.
    • 2026-08-29 — caught up through noisemaker 19c15cc0 / noisedeck eb3c3de6: Noisemaker's generated English catalog and all six translated effect catalogs were already complete; added 312 Noisedeck translations (52 new UI leaves in each of six locales) for shader-pipeline export, collaboration, completion, save-failure, and unloaded-user-effect copy. Added durable Noisemaker key-order/non-empty-value checks and Noisedeck key-order, non-empty-value, placeholder, and markup integrity checks.
    • 2026-08-19 — caught up through 1ee891a2: audited the i18n catch-up range; the English catalog was unchanged and all six locale catalogs remain complete, so no additional translations were needed in this pass.
    • 2026-08-14 — caught up through 712ac2cf: audited the i18n catch-up range; the English catalog was unchanged and all six locale catalogs are now complete, so no additional translations were needed in this pass.
    • 2026-08-14 — catch-up, 1,566 strings: 261 missing keys across all six locales, covering the new and expanded filter/chrome through filter/wind effect blocks.
    • 2026-07-10 — initial top-off, 37 strings: filter/parallax block (5 keys, de/fr/it/ja/pt), filter/dither.type.errorDiffusion and filter/lighting.heightMap (all six locales).

Large-format tiling

  • Checkpoint: noisemaker e1fbfd0d / noisedeck ec401c94 (preview branch), 2026-10-08
  • Scope: every effect must be classified for Noisedeck's large-format (tiled print) export. Tile-aware effects consume the global tileOffset and fullResolution uniforms in both GLSL and WGSL when their coordinates depend on the full canvas (packed WGSL layouts may need explicit uniformLayout slots); purely local kernels are also safe when their maximum source footprint fits inside the 1024 px overlap. Effects that cannot render tiled belong in one of noisedeck's deny-lists: app/js/utils/hasStatefulEffects.js (state textures cannot re-render at print resolution) or app/js/utils/hasUpscaleOnlyEffects.js (would seam at the 1024 px tile overlap). An effect in none of the three states silently produces seamed prints.
  • Gap detection: list effects added since the checkpoint (git log --diff-filter=A --name-only <checkpoint>..HEAD -- 'shaders/effects/*/*/definition.js') and classify each: make it tile-aware, or route it into the appropriate deny-list. Verify tile-aware claims with noisedeck's seam harness (tests/large-format-seams/).
  • Log:

    • 2026-10-08 — caught up through noisemaker e1fbfd0d / noisedeck ec401c94: gap detection from 058d15dc found zero added effects. The range (the WebGPU-matches-WebGL2 campaign through 1.0.271) changed 34 definitions and the shaders of 98 effects; 58 of the 111 are in the harness scope and 53 are in a deny-list. Noisedeck's seam harness, run against bundles built at e1fbfd0d, gave 56/58 seam-pass on GLSL and the same on WGSL. render/meshRender passes only trivially (its default program has no mesh). filter/spookyTicker was in neither deny-list and is not tile-aware: its ticker rows are measured from the input texture's edge in tile-local coordinates, so the harness reports it cannot-test and every tile would draw its own band. Noisedeck ec401c94 routes it to hasUpscaleOnlyEffects.js and the harness's UPSCALE_ONLY_FUNCS; its deny-list test pins it (113/113 classifier and harness tests, 1385 Node tests).

    • 2026-10-03 — caught up through noisemaker 058d15dc / noisedeck 7cac7e92: gap detection from 1d581ffa found zero added effects. Changed definitions were filter/degauss, render/pointsEmit, and synth3d/cellularAutomata3d. Degauss now binds tileOffset and fullResolution in its packed WGSL layout; both shader backends use global coordinates and clamp the tiled displacement against a 256-pixel budget. The Noisedeck seam harness passed filter/degauss on both GLSL and WGSL against the published Noisemaker 1.0.221 bundle. The two simulation definitions changed only resetOnChange UI metadata; both remain in hasStatefulEffects.js. The subsequent filter/octaveWarp hash rewrite did not change its spatial sampling, and it remains in hasUpscaleOnlyEffects.js. No Noisedeck classifier edit was needed. Verified 84/84 Noisedeck classifier and seam enumerator tests and the Noisemaker JS suite. The checkout's older ignored dist/ gave a false WGSL seam result because it lacked degauss's new uniform layout; live 1.0.221 bundles passed, and dist/ was restored after the check.

    • 2026-09-22 — caught up through noisemaker 1d581ffa / noisedeck 697d2df9: gap detection (git log --diff-filter=A) identified zero new effects in range eef25e91..1d581ffa (covering work queue item #393 trigger 2f855c9c..36a519a2 through HEAD). Audited the one effect with changed definition in the range: render/renderLandscape3d (36a519a2), which added a pruned isosurface filtering mode. Both perspective and isometric isosurface raymarch paths derive global UV coordinates from tileOffset and fullResolution across GLSL and WGSL, preserving its tile-aware classification. No additions to hasStatefulEffects.js or hasUpscaleOnlyEffects.js required. Verified Noisemaker landscape tests (npm run test:shaders:landscape, 13/13 passing with dual WebGL2/WebGPU parity attestations for synth3d/heightmap3d and render/renderLandscape3d), Noisemaker JS test suite (65/65 passed), and Noisedeck classifier and seam harness tests (83/83 passed across has-stateful-effects.node-test.js, has-upscale-only-effects.node-test.js, and tests/large-format-seams/enumerator.node-test.js).

    • 2026-09-21 — caught up through noisemaker eef25e91 / noisedeck 4e412eb5: gap detection (git log --diff-filter=A) identified zero new effects in range ead42a5d..eef25e91 (covering work queue item #313 trigger f2506d21..2f855c9c). Audited the effects changed in the range: commit 2f855c9c retired filter/bc, filter/colorspace, and filter/hs, none of which were in noisedeck's deny-lists (hasStatefulEffects.js or hasUpscaleOnlyEffects.js). In noisedeck, zero classifier changes occurred across d6fd477a..4e412eb5. Verified 83/83 classifier and seam harness tests in noisedeck (has-stateful-effects.node-test.js, has-upscale-only-effects.node-test.js, enumerator.node-test.js). No additions to hasStatefulEffects.js or hasUpscaleOnlyEffects.js required.

    • 2026-09-18 — caught up through noisemaker ead42a5d / noisedeck d6fd477a: gap detection (git log --diff-filter=A) identified zero new effects in range ff1bfbc1..ead42a5d. Audited the two effects with changed definitions in the range: synth3d/heightmap3d and render/renderLandscape3d (f2506d21). Their definitions changed only defaultProgram to use discrete write/read pipelines instead of inline surface parameter calls; neither shader implementation (.glsl / .wgsl) or spatial coordinate handling was modified. Both effects remain tile-safe under their established classifications (heightmap3d volume bake pass upstream of canvas tiles; renderLandscape3d global-UV projection from tileOffset/fullResolution). No additions to hasStatefulEffects.js or hasUpscaleOnlyEffects.js required. Verified 83/83 classifier and seam harness tests in noisedeck (has-stateful-effects.node-test.js, has-upscale-only-effects.node-test.js, enumerator.node-test.js).

    • 2026-09-15 — caught up through noisemaker ff1bfbc1: classified the perspective/depth-sort/defocus round's 3 new and 7 changed effects. points/heightGrid (new: writes agent state via MRT like the other points/* behaviors already listed) needed a new hasStatefulEffects.js entry, mirrored in the seam harness's enumerator.js. The other 9 needed no routing change: synth3d/ heightmap3d (new) is a volume-atlas bake pass upstream of the canvas tile boundary, the same structural class already documented for palette3d; render/renderLandscape3d (new, plus this round's own definition change for its perspective camera) correctly derives one tile-aware uv from fullResolution/tileOffset shared by both its isometric and new perspective branches, verified in both GLSL and WGSL; render/pointsRender and render/pointsBillboardRender were already deny-listed and this round's new perspective/depth-sort/ defocus passes don't change that; synth/remap's full zone-compositor rewrite preserved the tile-local-source-sampling / global-polygon- coordinate invariant fixed in the 2026-09-07 pass below; filter/ chrome, filter/grade, and mixer/alphaMask's premultiplied-alpha/ gradient fixes didn't touch spatial handling and remain tile-safe. Verified 23/23 classifier + harness-mirror tests (has-stateful-effects.node-test.js, enumerator.node-test.js). Ran noisedeck's live GLSL seam harness (headless Chromium) for mixer/alphaMask (pass) and synth/remap (trivial/uniform default DSL); could not run it for the two new 3D-landscape effects — this checkout has no synced app/js/noisemaker/vendor/, and the harness's live-CDN fallback doesn't resolve newly-published effect bundles for its effect loader. Their classification rests on direct source verification, the same determinant method this file's own scope note and hasUpscaleOnlyEffects.js's header prescribe for structural-safety claims, not live seam-harness pixels — flagged for a real render once a synced vendor/ checkout is available. Also found, untouched (out of scope): noisedeck's preview branch carries a second, older stash (fix(fullscreen): restore canvas dimensions on exit for any entry path) and a large set of uncommitted working-tree changes matching the dangling work queue item #156 I18n narrative — both predate this pass and were left exactly as found.

    • 2026-09-07 — caught up through noisemaker 827b2e91: no new effects; rechecked the changed Remap definition and Text shader. Remap's active zones exposed a GLSL seam: source surfaces are tile-local, but sampling used global UV. Corrected the source lookup to use the packed resolution slot; polygon coordinates remain global and WGSL already uses local source sampling. Remap now has a zero-pixel source displacement and needs no deny-list entry. Both backends match full-frame captures exactly at 2048x2048 and 2048x3000 using Noisedeck's seam harness; four untiled before/after captures remain byte-identical. Added a nontrivial active-zone regression, renewed source-bound strict Remap parity evidence, and passed 33 tile checks, 126 harness/classifier tests, Shade compilation/structure, and the full non-parity JS suite. Text's external-overlay tile regression also passes. No Noisedeck routing changes were required.

    • 2026-09-04 — caught up through noisemaker ee523ab9 / noisedeck 539ee089: audited 25 added artistic filters and seven changed effect definitions. Classified 23 additions as tile-safe, routed spinBlur and pondRipples to the upscale-only path, and kept extrude tiled with a pinned 768 px maximum source-dependency bound. Repaired text WGSL to sample its external overlay in global output space and added a behavioral WebGPU full-frame/offset-tile regression. Stopped Noisedeck's animation loop during controlled captures to eliminate a WebGPU harness race and consolidated the diagnostic deny-list onto the production-synced mirror. Revalidated 124 harness/classifier tests, 31 cross-backend tile checks, all 203 shader structures, and the full shader-language suite; no stateful routing changes were required.

    • 2026-07-10 — post-ship catch-up: parallax tile-clamped + synth WGSL ports (noisemaker 45a34489); effects added since ship routed into deny-lists, seam-harness hardening (noisedeck a12ee01e..75262325).

    • 2026-05-16 — feature shipped (noisemaker f1b0a919, noisedeck 913b722b).

Documentation

  • Checkpoint: noisemaker badf1f9e (2026-10-08)
  • Scope: the Sphinx docs under docs/ (published to docs.noisemaker.app by .github/workflows/docs-site.yml) and the per-effect shaders/effects/*/*/help.md files rendered by the live Effect Reference. help.md is normally written in-band with the effect; this pass catches stragglers and decides which shipped features merit a narrative guide.
  • Gap detection:
    1. Effects missing help.md: compare shaders/effects/*/*/definition.js against the corresponding help.md paths.
    2. Features shipped since the checkpoint that merit a guide under docs/shaders/features.rst (pattern: docs/shaders/cubemaps.rst).
    3. Narrative statements invalidated by recent commits (grep the affected terms in docs/).
  • Log:

    • 2026-10-08 — caught up through badf1f9e: audited a1329096..badf1f9e (212 commits, including the WebGPU-matches-WebGL2 campaign through 1.0.271). All 210 definitions have help.md, and all 1,258 documented parameters across 200 effects match their definitions (node shaders/tests/test_effect_help_params.mjs). No shipped feature needs a narrative guide: the engine changes are parity fixes, WebGPU device limits and canvas uploads, and host APIs. Two host-facing gaps were closed: docs/shaders/integration.rst now lists renderer.pipeline.whenAsyncInitsSettled() (cabe6ee3) for one-off captures of async CPU effects, and docs/shaders/pipeline.rst now says that WebGPU also records ERR_MISSING_RENDER_TARGET for MRT, storage-surface and copy-output targets (5578cf7c). Canvas sizing from width/height (b805b365), getMediaSteps() (632820a8) and oscKind.noise2d (4e046a2d) were documented in-band; the 128³ volume ceiling now holds on WebGPU too (6a772479), so no statement changed. sphinx-build -b dirhtml docs passes; its 19 warnings are the existing unregistered dsl lexer.

    • 2026-10-04 — caught up through a1329096: audited 63cfbb7..a1329096, including work queue item 804's observed 378c8bf8..058d15dc delivery. Both range endpoints are ancestors of the audited mainline tip. The shipped changes are portable-effect registration, live parameter aliases and reseed hints, degauss tile coordinates, and octaveWarp hash conversion; the latter two are filter fixes, not new authoring features. The portable registration path and live-parameter behavior were documented in-band in docs/shaders/integration.rst, effects.rst, and pipeline.rst. Effect coverage is unchanged: all 210 definitions have help.md, and all 1,258 documented parameters across 200 effects match their definitions (test_effect_help_params.mjs). The degauss and octaveWarp help files remain consistent with their definitions; no new feature guide or Sphinx narrative correction is needed. A fresh Sphinx dirhtml build succeeded with 18 existing warnings; docs static-path tests passed 4/4, ESLint passed, and the non-parity JS suite passed.

    • 2026-10-01 — caught up through 63cfbb7: audited the job's trigger range dd4606e..63cfbb7 (work queue item 704, delivered as a02f73e..63cfbb7; the trigger's start dd4606e sits in the non-contiguous history but is an ancestor of published main). The range is already fully documented: the GAP-007 final leg (e24c844, a0e9bbf), the in-band pipeline.rst alignment (e105344), and the test-only GAP-010 commits (16c1997, ed47815) were audited by the prior pass whose own checkpoint advance is this range's end commit 63cfbb7, and a02f73e touches only LEDGER i18n checkpoints. Gap detection re-run: all 210 shaders/effects/*/*/definition.js have their help.md (zero missing). No new end-user feature in the range, so no docs/shaders/features.rst guide; no narrative statements invalidated by a02f73e/63cfbb7 (LEDGER-only + the prior pass's own docs edits). Verified the Sphinx docs build locally (sphinx-build -b dirhtml docs docs/_build/dirhtml: build succeeded, 18 pre-existing warnings, 0 errors), docs static asset paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint, clean), and the non-parity JS test suite (node scripts/run-js-tests.js --skip-parity, 0 failures). Measured results above are archived in the job's run evidence (Automated job d441d8a4, evidence noisemaker-ledger-docs-704.txt: clean full Sphinx build after removing docs/_build/dirhtml — the "1 warning" incremental re-run in that file predates the clean rebuild and does not reflect a from-scratch build).

    • 2026-10-01 — caught up through ed47815: audited range 01e9d62..dd4606e (work queue item 698; the trigger's start 01e9d62 had been force-pushed out of upstream main, so the objects were fetched directly by SHA and the observed delivery ranges c2a19c70..dd4606e, a0e9bbffc..e24c844f, e24c844f..e105344 were diffed locally; all of them plus the later test-only commits 16c1997/ed47815 are ancestors of published main at a02f73e, 44 commits in the audited range). Gap detection: all effect definitions still have their help.md (zero missing across shaders/effects/*/*/definition.js), so no help.md writes were needed. No new end-user feature shipped in the range that merits a docs/shaders/features.rst guide: the range's substantive changes are GAP-007..GAP-032 fixes (diagnostics union legs, audio capture, harness and metric tooling) plus GAP-026, which makes the already-documented lifecycle-hook authoring feature actually run in the production renderer — its documentation home stays docs/shaders/effects.rst (Lifecycle Method Contract) and docs/shaders/pipeline.rst, not a new features page. Invalidated-statement audit edits: docs/shaders/pipeline.rst §14.2 no longer claims the pipeline has no effect-lifecycle mechanism — it now documents the managed-effect set initLifecycleEffects() rebuilds (onInit once per pipeline lifetime, per-frame onUpdate with fallback-bound returned uniforms, dispose-time onDestroy joining the dispose error path), matching GAP-026's initLifecycleEffects()/ _invokeUpdateHooks()/_withRuntimeUniforms()/dispose() in shaders/src/runtime/pipeline.js; docs/shaders/effects.rst's lifecycle contract now states the production invocation and the fallback uniform-binding semantics (a returned key binds only when the pass does not already resolve it, verified against _withRuntimeUniforms()), and the dimension grammar now lists the validator-accepted 'input'/'resolution' keywords that resolve to screen size without a fallback diagnostic (a0e9bbff), in both the dimensionSpec JSON schema description and the resolveDimension pseudocode. GAP-007's runtime-leg statements were already aligned in-band by e105344 (pipeline.rst + llms-full.txt); the audio capture, replaceEffect, preflight, and harness-metric changes invalidated no narrative statements (the MCP tool descriptions in docs/coding-agents.rst are unchanged by the range, and replaceEffect/ preflight remain engine-internal APIs with no Sphinx page to correct). Verified the Sphinx docs build locally (sphinx-build -b dirhtml docs docs/_build/dirhtml: build succeeded, 18 pre-existing warnings, 0 errors), docs static asset paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint, clean), and the non-parity JS test suite (node scripts/run-js-tests.js --skip-parity, 0 failures). The live noisemaker-site and noisemaker-docs-site deployments serve ed47815 per deployment-meta.json, so the pipeline.rst/effects.rst corrections in this commit reach the published docs with the next docs-site build.

    • 2026-09-26 — caught up through 01e9d62: audited range 6a0af04d..01e9d62 (work queue item 586). The range contains exactly one commit, 01e9d62, which is the prior Documentation pass's own Sphinx edits (GAP-006 texture-pooling opt-in and GAP-007 structured ShaderDiagnostic union in docs/shaders/pipeline.rst and docs/shaders/compiler.rst) — no engine changes. Gap detection confirmed zero missing help.md files across all 210 effect definitions and all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). No new features shipped in the range, so no docs/shaders/features.rst guide required. Invalidated-statement audit: spot-checked the documented symbols (ShaderDiagnostic/ERR_SHADER_MISSING in shaders/src/runtime/backends/diagnostics.js and the backends, buildTexturePoolingPlan/getResourcePlan/releaseRegroupedTextures/ applyTextureAliases in shaders/src/runtime/pipeline.js) and grepped the affected terms in docs/ — every remaining "virtual ID"/"allocation map" statement is qualified by the texturePooling: true opt-in and matches the source; nothing invalidated. Verified the Sphinx docs build locally (sphinx-build -b dirhtml docs docs/_build/dirhtml: build succeeded, 18 pre-existing warnings, 0 errors), docs static asset paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and the non-parity JS test suite (node scripts/run-js-tests.js --skip-parity, 201 tests, 0 failures). Published-source CI and deployment evidence observed via the GitHub API: docs-site.yml succeeded at 01e9d62 (Verify docs asset paths and Dispatch static-site-release jobs both success; run 36213112409), downstream.yml succeeded at 01e9d62 (run 36213112296), and the live docs.noisemaker.app deployment serves git_hash: 01e9d620d7ccc700cc8926a91145702dc69a8eeb per deployment-meta.json — the deployed Sphinx HTML includes this range's pipeline.rst/compiler.rst edits. The follow-up checkpoint-advance commit 3968f6c touches only LEDGER.md, which is outside the docs-site.yml push-trigger paths (docs/** with *.md excluded), so no workflow run applies to it and the deployment at 01e9d62 already serves all this pass's .rst content.

    • 2026-09-26 — caught up through 6a0af04d: audited range 6c3f9a26..6a0af04d (work queue item 571, covering trigger 8eeb7b5a..6a0af04d delivered as 6c3f9a26..428ea29..9574362..6a0af04d, audited linearly in the local checkout). Gap detection confirmed zero missing help.md files across all effect definitions. Shipped features documented in the Sphinx docs: GAP-006 texture-pooling consumption (the texturePooling: true opt-in, Pipeline.buildTexturePoolingPlan() grouping/aliasing, the identical-plain-2D-spec / producing-write / partial-and-non-clearing-write guard including the viewport-without-clear case from 9574362, and Pipeline.getResourcePlan()) in docs/shaders/pipeline.rst Sections 2 and 7; GAP-007 structured diagnostics (one ShaderDiagnostic union with backend/stage/parsed messages, the added ERR_SHADER_MISSING code row, and the enumerated still-unstructured paths) in docs/shaders/pipeline.rst Section 11 and docs/shaders/compiler.rst Failures and Diagnostics (with a new pipeline-error-codes anchor). No new standalone guide required under docs/shaders/features.rst — both features are pipeline/runtime internals, matching the prior pass's precedent. Invalidated statements: corrected docs/shaders/pipeline.rst claims that the Pipeline creates textures by virtual ID and never uses the allocation map as a pool. Verified the Sphinx docs build locally (sphinx-build -b dirhtml docs docs/_build/dirhtml: build succeeded; only the pre-existing pipeline-3d.rst dsl lexer warning, unchanged), docs static asset paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and the non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-25 — caught up through 6c3f9a26: audited range 69d83b80..6c3f9a26 (work queue item 557, covering 69d83b80..8eeb7b5a and publication evidence at 6c3f9a26). Gap detection confirmed zero missing help.md files across all 210 effect definitions and all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Shipped features: GAP-004 texture policies (mipmaps, persistent, 3D filter) in docs/shaders/effects.rst and docs/shaders/pipeline.rst; GAP-005 pass fields (name, type, clear, samplerTypes, viewport) and dynamic dimension viewport resolution (viewportResolved) documented in docs/shaders/effects.rst (passSpec schema updated with type, clear, samplerTypes, viewport, and conditions) and docs/shaders/pipeline.rst (Section 9.2 added for pass-field propagation and viewport resolution). No new standalone guide required under docs/shaders/features.rst as pass execution controls belong to core pipeline/effect schemas. Invalidated statements: corrected docs/shaders/pipeline.rst Section 9 to state that the expander copies authored conditions for per-frame dynamic pass skipping. Verified the Sphinx docs build locally (sphinx-build -b dirhtml docs docs/_build/dirhtml: build succeeded, 0 errors), docs static asset paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-25 — caught up through 69d83b80: audited range fa4b2f02..69d83b80 (work queue item 538, delivered as the observed ranges 2f47612c..63349a7d and fa4b2f0..69d83b80; the trigger end is a force-push/non-contiguous delivery, so the range was diffed directly in the checkout). Gap detection confirmed zero missing help.md files across all 210 effect definitions and all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). The range's Documentation-scope changes are this pass's own prior Sphinx edits (69d83b80: GAP-003 validateEffectDefinition in docs/shaders/pipeline.rst and docs/shaders/effects.rst) plus 63349a7's docs/releases.rst rewrite for the retired per-push SNAPSHOT release (checked against the remaining workflow narrative — accurate); no invalidated narrative statements remained. The range's engine commits (a021a28, 62eb56f, 2f47612, GAP-004 texture policies and backend mip handling) and closure docs (85ded3a) were audited under the AI development contract section. Invalidated-statement audit of GAP-004 updated docs/shaders/effects.rst (textureSpec schema now documents mipmaps, persistent, and 3D filter) and docs/shaders/pipeline.rst (resize-behavior section documents both 2D allocation policies, mip-chain regeneration, persistent resample, and 3D filtering; the blanket "does not blit old surface content" sentence no longer holds for persistent textures), with no new standalone narrative guide: the policy documentation lives in the existing docs/shaders/effects.rst and docs/shaders/pipeline.rst sections rather than a new docs/shaders/features.rst-linked page. Verified the Sphinx docs build locally at the exact source (sphinx-build -b dirhtml docs with docs/sphinx-requirements.txt: build succeeded, 18 pre-existing warnings, 0 errors), docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity). Published-source CI and deployment evidence observed via the GitHub API: docs-site.yml succeeded at 69d83b80 (run 36164682935) and at 63349a7d (run 36174749632), with downstream.yml also succeeding at both SHAs (36164683019, 36174749672); the live docs.noisemaker.app deployment serves git_hash: 63349a7d6b95da1e97f35b8570872db82c55d7da per deployment-meta.json — 63349a7d descends from 69d83b80 (verified with git merge-base --is-ancestor), so the deployed Sphinx HTML includes the 69d83b80 GAP-003 .rst edits. The GAP-004 texture-policy .rst documentation added later in this pass is not yet in that deployment; it publishes with this documentation commit's own docs-site.yml run.

    • 2026-09-25 — caught up through fa4b2f02: audited range cfccdf96..fa4b2f02 (covering work queue item #534 trigger 60b90af3..9d3474df through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Updated Sphinx docs in docs/shaders/pipeline.rst and docs/shaders/effects.rst to reflect GAP-003 runtime effect definition validation (validateEffectDefinition in shaders/src/runtime/effect-validator.js), correcting earlier abridged descriptions. Confirmed zero new narrative feature guides required under docs/shaders/features.rst. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass), structure suite (npm run test:shaders:structure, 206/206 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-25 — caught up through cfccdf96: audited range 240740dd..cfccdf96 (covering work queue item #522 trigger 60b90af3..5f4b7c43 through HEAD). Range consists entirely of documentation, ledger, and AI contract updates (5a58a325, c85db3f0, 5f4b7c43, cfccdf96). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Confirmed zero new narrative feature guides required under docs/shaders/features.rst. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-25 — caught up through 240740dd: audited range bbdeb56c..240740dd (covering work queue item #510 trigger 4891b995..60b90af3 through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Invalidation audit updated docs/shaders/language.rst to document subchain argument separator rules, strict-mode option (subchainArguments: 'strict'), and added P008 (Unknown subchain argument key), P009 (Duplicate subchain argument key), and P010 (Missing ',' between subchain arguments) to the DSL diagnostics table. Audited subchain-argument validation contract (66b2c721), registered differential gate baseline (240740dd), and AI contract checkpoint (3886ecfa); confirmed zero new narrative feature guides required under docs/shaders/features.rst. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-24 — caught up through bbdeb56c: audited range aa96726d..bbdeb56c (covering work queue item #496 trigger c9ee8a04..4891b995 through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Invalidation audit updated docs/shaders/language.rst to add P007 (Invalid call expression) to the DSL diagnostics table. Audited structured call-form and expectation diagnostics (4891b995), scanner source coordinate derivation (9fa1a221), numeric coercion array coordinate derivation (fca611fd), and GAP-002 closure provenance citations (8a21c9ca/bbdeb56c); confirmed zero new narrative feature guides required under docs/shaders/features.rst. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-24 — caught up through aa96726d: audited range 30c47030..aa96726d (covering work queue item #482 trigger c9ee8a04..13853dff through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Confirmed zero narrative statements invalidated and zero new narrative feature guides required under docs/shaders/features.rst. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-24 — caught up through 30c47030: audited range c7e9e09d..30c47030 (covering work queue item #474 triggers c9ee8a04..13fa8b54, GAP-002 subchain validation evidence 741333cb, and site styling 30c47030). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Invalidation audit updated docs/shaders/language.rst to add P006 (Invalid subchain) to the DSL diagnostics table. Confirmed zero new narrative feature guides required under docs/shaders/features.rst. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-24 — caught up through c7e9e09d: audited range 65ff358f..c7e9e09d (covering work queue item #455 triggers through 823bbff1 and contract checkpoint advance c7e9e09d). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Verified P005 in docs/shaders/language.rst; confirmed zero new narrative feature guides required. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-24 — caught up through 65ff358f: audited range 7d0be45c..65ff358f (covering work queue item #442 triggers 9b88e567..c9ee8a04 and cc1ba268 through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Invalidation audit updated docs/shaders/language.rst to add P005 (Invalid output operation) to the DSL diagnostics table. Audited structured output validation diagnostics (7a54ab38, P005 for invalid render targets, expression-context writes, and invalid write/write3d target arguments), AI development contract checkpoint advance (c9ee8a04), site header strip styling (285e50f5), and GAP-002 output validation reconciliation (65ff358f); no additional narrative feature guides under docs/shaders/features.rst required. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-23 — caught up through 7d0be45c: audited range ec952244..7d0be45c (covering work queue item #433 triggers 0766743e..9b88e567 and 5b81e04f through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Invalidation audit updated docs/shaders/renderer-output.rst to document the optional deferRender() sink method and renderer.deferredFrameCount shipped in 5b81e04f. Audited classicNoisedeck/glitch zero-amount guards and parity attestation (9b88e567); no additional narrative feature guides under docs/shaders/features.rst required. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-23 — caught up through ec952244: audited range 731b76e4..ec952244 (covering work queue item #417 triggers 9e188535..0766743e through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Invalidation audit updated docs/shaders/language.rst to add P004 (Invalid search directive) to the DSL diagnostics table. Audited structured search directive diagnostics (0766743e, P004 in parse(tokens) / compile(source)), scoped texture size preservation on setUniform (e2874c85/3a32b198), and classicNoisedeck/noise zero-refract optimization and parity attestation (fde2ea40/e11f0767); none required additional narrative feature guides. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-23 — caught up through 731b76e4: audited range eb46047b..731b76e4 (covering work queue item #402 triggers 44bc4ed4..9e188535 and 532ed647..e32a5a4a through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Invalidation audit updated docs/shaders/language.rst to add L003 (Unterminated comment), L004 (Output surface reference out of range), and P003 (Invalid automation arguments) to the DSL diagnostics table and remove duplicate S005 entry. Audited automation argument diagnostics (e32a5a4a, P003 in parse(tokens) / compile(source) for osc(), midi(), audio()), documented in llms-full.txt and docs/plans/active-framework-gap.md. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass), ESLint (npm run lint), and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity).

    • 2026-09-22 — caught up through eb46047b: audited range 2779b409..eb46047b (covering work queue item #394 trigger e5bd2013..44bc4ed4 through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,258 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs, incremented from 1,257 due to filtering added to renderLandscape3d in 36a519a2). Invalidation audit updated docs/shaders/pipeline-3d.rst to document renderLandscape3d's new smooth isosurface mode alongside voxel mode. Audited parser expectation diagnostics (44bc4ed4, P001/P002 in parse(tokens) / compile(source)), already documented in llms-full.txt and docs/plans/active-framework-gap.md. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass) and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity, 65 tests pass).

    • 2026-09-22 — caught up through 2779b409: audited range e5bd2013..2779b409 (covering work queue item #374 trigger e5bd2013..643b2be1 through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,257 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Audit of commits in the range (52ac841b recording GAP-002 column verification, b0364da0 docs ledger advance, a0ff705a contract ledger advance, 643b2be1 exposing structured DSL lexer diagnostics, and 2779b409 recording GAP-002 lexer verification) confirmed no narrative statements in docs/ were invalidated and no new narrative feature guides under docs/shaders/features.rst were required. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass) and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity, 64 tests pass).

    • 2026-09-22 — caught up through e5bd2013: audited range c9ca39ba..e5bd2013 (covering work queue item #343 trigger 62bd24bc..e5bd2013 through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,257 documented parameters across 200 effects match definitions (node shaders/tests/test_effect_help_params.mjs). Audit of commits in the range (68d37721 excluding builtins from mutation introspection, 256d2d46 closing GAP-022 in docs plan, 3001db91 ledger coverage, and e5bd2013 preserving source columns in DSL diagnostics) confirmed no narrative statements in docs/ were invalidated and no new narrative feature guides under docs/shaders/features.rst were required. Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass) and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity, 64 tests pass).

    • 2026-09-21 — caught up through c9ca39ba: audited range 9d522a31..c9ca39ba (covering work queue item #327 trigger 50b8f909..62bd24bc through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Verified all 1,257 documented parameters across 200 effects match definitions. Invalidation audit confirmed no further documentation statements invalidated (DSL output surface range enforcement in docs/shaders/language.rst already landed in 62bd24bc). Verified docs static paths (node --test test/docs-static-paths.test.js, 4/4 pass) and non-parity JS test suite (node scripts/run-js-tests.js --skip-parity, 64 tests pass).

    • 2026-09-21 — caught up through 9d522a31: audited range f61ac073..9d522a31 (covering work queue item #315 trigger 2f855c9c..50b8f909 through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions. Invalidation audit identified that the DSL grammar definition for OutputRef in docs/shaders/language.rst allowed unbounded digits ('o' Digit+); updated it to OutputDigit ('0'…'7') and documented compile-time range enforcement (resolved by GAP-001 in 50b8f909). Verified node --test test/docs-static-paths.test.js (4/4 pass), node shaders/tests/test_effect_help_params.mjs (1257 documented parameters across 200 effects matching definitions), and node scripts/run-js-tests.js --skip-parity (57/57 passed).

    • 2026-09-21 — caught up through f61ac073: audited range ac8a8d90..f61ac073 (covering work queue item #306 trigger 8ddf9e6e..2f855c9c through HEAD). Gap detection confirmed zero missing help.md files across all 210 effect definitions (accounting for the removal of expired effects bc, colorspace, and hs in 2f855c9c). Invalidation audit identified an outdated warning in docs/shaders/renderer-output.rst regarding frame-export queue reconfiguration/close accounting; corrected it to document that canceled pending frames are now accounted as dropped (resolved by GAP-028 in 0139e958). Verified node --test test/docs-static-paths.test.js (4/4 pass), node shaders/tests/test_effect_help_params.mjs (1257 documented parameters across 200 effects matching definitions), and node scripts/run-js-tests.js --skip-parity (57/57 passed).

    • 2026-09-20 — caught up through ac8a8d90: audited range

      beabda38..ac8a8d90. Gap detection confirmed zero missing help.md files across all 213 effect definitions. Commits in the range consist of documentation and ledger synchronizations: active framework gap resolution (b8332f16, closing GAP-031), prior Documentation checkpoint and Sphinx narrative updates (8ddf9e6e), contract snapshot re-audit (1086890d), and I18n strings checkpoint advancement (ac8a8d90). None invalidated existing Sphinx narrative in docs/ or required new feature guides under docs/shaders/features.rst. Verified node --test test/docs-static-paths.test.js (4/4 pass) and node shaders/tests/test_effect_help_params.mjs (1263 documented parameters across 203 effects matching definitions). No documentation gaps to close.

    • 2026-09-20 — caught up through beabda38: audited range f90a4d2a..beabda38. Gap detection confirmed zero missing help.md files across all 213 effect definitions. Shipped commits in the range include borrowed VideoFrame texture upload support (2df19feb), dynamic surface/texture format recreation on format changes (6e0166ce, GAP-030), and static integer 1..16 channel enforcement for legacy MIDI modes (beabda38, GAP-031). Narrative audit identified an invalidated statement in docs/shaders/midi-audio.rst regarding legacy note mode channel handling, which was corrected to reflect static integer 1–16 validation across all channel-based modes; also documented borrowed VideoFrame upload support in docs/shaders/integration.rst. Verified node --test test/docs-static-paths.test.js (4/4 pass) and node shaders/tests/test_effect_help_params.mjs (1263 documented parameters across 203 effects matching definitions).

    • 2026-09-19 — caught up through f90a4d2a: audited range ead42a5d..f90a4d2a. Gap detection confirmed zero missing help.md files across all 213 effect definitions. Shipped commits in the range include the compiler phase-2 harness nonzero exit status fix (f1d2b46a, GAP-023), related active framework gap status resolution (f90a4d2a), starter-position documentation in llms-full.txt (f3d6f9a6), agent doc unifications, and symlink ban enforcements. None invalidated existing Sphinx narrative in docs/ or required new feature guides under docs/shaders/features.rst. Verified node --test test/docs-static-paths.test.js (4/4 pass) and node shaders/tests/test_effect_help_params.mjs (1263 documented parameters across 203 effects matching definitions). No documentation gaps to close.

    • 2026-09-18 — caught up through ead42a5d: audited range 5a142567..ead42a5d. Gap detection identified zero missing help.md files across all 213 effect definitions. Shipped commits in the range include the WebGPU frame export row-inversion orientation fix (5ceb97ba/688c5146), defaultProgram discrete-chain updates for synth3d/heightmap3d and render/renderLandscape3d (f2506d21), and associated parity attestations (ead42a5d). None invalidated existing Sphinx narrative in docs/ or required new feature guides under docs/shaders/features.rst. Verified node --test test/docs-static-paths.test.js (4/4 pass). No documentation gaps to close.

    • 2026-09-16 — caught up through 5a142567: the range (c68fb3c8..5a142567) contains no new or changed effect definitions and no other docs/ edits — its only doc-relevant content is this section's own prior catch-up commit (5a142567 itself, the four files logged below) plus unrelated dependency bumps (Playwright, eslint, ruff, types-requests) and browser/CI test-harness fixes (WebGPU presentation sampling, landscape capture timing, renderer pruning audit) that touch no Sphinx narrative or effect-help content. Confirmed via git diff --stat c68fb3c8..5a142567 -- docs/ 'shaders/effects/*/*/definition.js' 'shaders/effects/*/*/help.md' (only the four already-logged files) and re-ran the production dirhtml Sphinx build: same 18 pre-existing warnings, no new ones. No gaps to close.

    • 2026-09-15 — caught up through c68fb3c8: the range added three effects (synth3d/heightmap3d, render/renderLandscape3d, points/heightGrid) landing a native voxel-heightfield landscape renderer, plus a perspective camera and depth-sorted alpha/aperture defocus for pointsBillboardRender, and a full rewrite of synth/remap's zone compositor. Wrote the two missing help.md files (heightmap3d, renderLandscape3d); heightGrid, pointsRender/pointsBillboardRender, and remap already carried accurate in-band help.md updates. Corrected docs/shaders/smrticles.rst: pointsRender's parameter table was missing posZ/fieldOfView (present in the definition since before this checkpoint, so this was a pre-existing gap, not new this round) and still described viewMode as flat/ortho only; added a pointsBillboardRender reference section (shape modes, blend modes, depth sort, defocus) that did not exist despite the effect predating this checkpoint, replaced a stale note claiming billboards/textured sprites were "planned for future releases" when pointsBillboardRender already ships them, and added heightGrid to the behavior-middleware table. Corrected docs/shaders/pipeline-3d.rst: added heightmap3d to the generators list and renderLandscape3d to the renderers table, both previously absent. Confirmed the help-table checker passes (1,263 documented parameters across 203 effects) and the production-format Sphinx dirhtml build passes with the same 18 pre-existing warnings as the prior checkpoint (all Pygments lexer name 'dsl' is not known, unrelated to this pass).

    • 2026-09-07 — caught up through d30d1045: verified help coverage for all 210 effects and 1,223 documented parameters across the 200 effects recognized by the table checker. Documented lossless Remap vertex formatting in the language, effect-authoring, and effect-help references. Re-audited the in-band MIDI CC, CC14, NRPN, MPE, audio-channel, and capture capability guides; no additional feature guide was needed. The existing non-parity suite, six Remap round-trip tests, and four static-asset checks passed. The production-format Sphinx dirhtml build passed with 18 warnings in unchanged documentation and configuration.

    • 2026-09-04 — caught up through acb26f80: verified help.md file coverage for all 210 effect definitions and ran the existing parameter-table checker across the 200 definitions it recognizes. Corrected the affected shader-language narrative for mandatory namespace search, built-in I/O, current generators and namespaces, and schema-valid argument and partial-application examples. Aligned the compiler guide with its actual parse, validate, expand, allocate, graph-assembly, backend-compilation, and error contracts, and removed the same obsolete DAG, synthetic-texture, pass-field, and runtime-shape claims from the pipeline and effect-authoring guides. No effect or user-facing feature was added, so no new narrative guide was required. Revalidated the production dirhtml Sphinx build and static-document asset checks.

    • 2026-09-04 — caught up through 8cad384f: verified help.md file coverage for all 210 effect definitions and ran the existing parameter-table checker across the 200 definitions it recognizes. Audited the documentation-only range: its refreshed AI development contract and corrected oscillator, normalized automation, selected-device, raw-audio, and nested-automation language are already reflected in the existing shader guides. No effect or user-facing feature was added, so no new narrative guide was required. Revalidated the production dirhtml Sphinx build and static-document asset checks.

    • 2026-09-04 — caught up through 5b0b7aa7: verified help.md file coverage for all 210 effect definitions and ran the existing parameter-table checker across the 200 definitions it recognizes. Updated the existing CLI, general DSL language, and shader guides for strobe verdicts, magic-mashup input gating, normalized and nested MIDI/audio automation, selected external devices, raw audio, external-state exports, device-bound 3D volume sizing, and MRT precision fallback. The remaining dependency, ledger, localization, and tiled-text changes did not require another narrative guide. Revalidated the production dirhtml Sphinx build and static-document asset checks.

    • 2026-08-19 — caught up through 1ee891a2: verified help.md coverage for all 210 effects and audited the shader-parity attestation infrastructure, PixelSort parity repair, JS/Python parity corrections, fixture tooling, and dependency updates in the range; updated the release workflow narrative for the new CPU parity-attestation gate, while no new effect or user-facing feature required an additional feature guide.

    • 2026-08-14 — caught up through 24c7053c: verified help.md coverage for all 210 effects and audited the renderer-output guide and navigation links, i18n catalogs, AI contract, CI, and vendored tooling in the range; the existing guide covers the only Sphinx-relevant feature, so no additional narrative changes were needed in this pass.

    • 2026-08-14 — caught up through 0d34e10d: verified help.md coverage for all 210 effects, added the renderer-output guide for sinks and bounded asynchronous frame export, and audited the in-band export-kit, static-site, effect-authoring, and repository-link changes; no other Sphinx narrative was invalidated.

    • 2026-08-11 — caught up through f33b5bfb: verified help.md coverage for all 210 effects and audited the AI development contract refresh plus the prior documentation catch-up; neither shipped a shader feature nor invalidated Sphinx narrative, so no guide or narrative edits were needed.

    • 2026-08-11 — caught up through 7cc4894b: verified help.md coverage for all 210 effects, corrected the remaining stale pre-move repository URL in the CLI help snapshot, and audited the Playwright and Ruff dependency bumps; no additional shader feature guides needed.

    • 2026-08-10 — caught up through 7cdf2b6d: verified help.md coverage for all 210 effects (including 25 effects added since the checkpoint), corrected four stale pre-move noisemaker repository links, and audited the in-band coding-agents guide plus runtime fixes; no additional shader feature guides needed.

    • 2026-07-10 — initial catch-up: added help.md for mixer/channelCombine and filter/temporalAberration; added feature guides for parallax, the 3D pipeline, and mashup.

AI development contract (llms-full.txt)

  • Checkpoint: noisemaker db2173e34223a663b08815fa32c34608ec5b4b91 / shade-mcp 63e53ad89dc43526335da6e947ed31667929e43b, 2026-11-13
  • Scope: compatibility between Noisemaker and Shade MCP, recorded in this shared ledger and the hand-authored agent contract llms-full.txt — the executable-source companion served at the site root that describes current runtime behavior across nine surfaces (DSL, effect definition, parameters/globals, passes/graph, textures, compatibility/mutation, rendered output, cross-backend parity, Shade MCP tool contracts), a fully worked validated effect and the surface × capability traceability matrix. Gap records live in repository issues. The file pins its own audited SHAs in the "Source snapshots used for this contract" block at its head. For a completed pass, that block and this checkpoint contain the same two SHAs. Exact-source CI requires a published candidate. During that CI and live readback, the candidate snapshot can lead this checkpoint. Record the candidate SHAs and pending checks in the log. Do not call the pass complete. Keep the last completed checkpoint until all checks pass. Then make the checkpoint and snapshot equal again. There is no generator. Verify every hand edit against live source. llms-full.txt states current behavior and open gaps only. A closed gap's row is deleted, and every reference to it goes with it. Never add closure notes, audit or re-capture history, commit provenance, CI run IDs, or verification timestamps to it; record those in this section's log. The short public index llms.txt carries no pinned snapshot and is kept current in-band with its links, so it is not part of this pass. Work spans both repositories even when only one changed. Noisemaker owns the engine, effects, viewer, MCP configuration, and vendored integration; Shade MCP owns its parsers, analysis, knowledge, browser tools, and harness exports. Fix affected integration code in its owning repository before describing the pair as compatible. Preserve valid DSL behavior, rendered output, defaults, saved programs, step indexes, and public tool result shapes. Unrelated engine gap closure is outside this pass.
  • Delivery evidence: record four identities separately in each completed pass: Noisemaker source SHA, Shade MCP source SHA, the immutable Shade MCP commit configured in .mcp.json, and the release tag/source SHA delivered to vendor/shade-mcp/. The MCP pin and vendored harness are independent delivery paths. A local Shade build or source fix proves neither release nor installation. Use the existing release and pull-shade-mcp process; never hand-edit generated vendor bundles or replace the immutable pin with a floating branch. Commit, push, release, and deployment actions still require explicit operator authorization.
  • Gap detection:

    1. Noisemaker drift — inspect changes since this section's current Noisemaker checkpoint, including source consumed by Shade and changes to either delivery path:

      git log --oneline <noisemaker-checkpoint>..HEAD -- shaders/src/ shaders/effects/ shaders/tests/ demo/shaders/ .mcp.json vendor/shade-mcp/ pull-shade-mcp .github/workflows/pull-shade-mcp.yml test/mcp-config.test.js

      Each can invalidate a behavior statement, typed grammar, or validator message, catalog reference, viewer bridge, harness API, or gap status. Check Shade's consumers against these changes, re-audit the affected contract sections, and re-check gaps whose "Source evidence" changed.

    2. Shade MCP drift — .mcp.json runs npx -y github:noisedeck/shade-mcp#<sha> pinned to an immutable commit (GAP-013 closed). Check upstream noisefactorllc/shade-mcp for new commits since the Shade checkpoint and for newer releases. Compare tool schemas/results, harness exports, and browser behavior with Noisemaker's viewer, vendor imports, and configured MCP. Reconcile both delivery paths with the tested Shade source and record their provenance. A newer release does not reopen GAP-013: that gap tracks immutable pinning, not freshness.

    3. Audit both complete checkpoint-to-source ranges. A queue trigger range is not evidence that an earlier pass completed. Missing vendor provenance, pending source fixes, undelivered releases, and unavailable checks are explicit blockers, not grounds to advance a checkpoint.

  • Validation and completion: run Noisemaker's prescribed JS tests and lint, Shade's typecheck and tests, relevant structure/compile/render checks on WebGL2 and WebGPU through the configured MCP and vendored harness, and Shade's real viewer smoke test (NOISEMAKER set to the Noisemaker checkout, node scripts/browser-smoke.mjs). Build Shade when needed under its instructions; never build Noisemaker dist locally. Re-capture MCP initialization, tools/list, the server/protocol version triple, and the affected worked request/response from one session at the tested immutable pin. Unit tests or a tool count alone do not prove browser compatibility. Verify required CI for each changed repository's exact published commit and verify the resulting delivery state. Then update affected contract sections, the traceability matrix, and source-evidence gaps, and advance both checkpoints together with the contract's snapshot block, or advance the checkpoint to an already published candidate after its exact-source CI and live readback pass. Include all four identities and validation evidence in the log. If any part remains unverified, keep the last completed checkpoints and record remaining work here. Historical entries below do not retroactively certify these added delivery checks.
  • Log:
    • 2026-11-13 — completed delivery pair: Noisemaker db2173e34223a663b08815fa32c34608ec5b4b91 and Shade MCP 63e53ad89dc43526335da6e947ed31667929e43b. Audited the Noisemaker range 6f98d26a..db2173e3. Its runtime changes are cf95f49c, which gives each further instance of a stateful effect in a chain its own private global_ state and sizing parameter (shaders/src/runtime/expander.js), and db2173e3, which makes the pipeline's default SinkManager pass an isolated sink error to the failing sink's optional onError(error) (shaders/src/runtime/pipeline.js). The vendored vendor/shade-mcp/ tree is unchanged over the range, .mcp.json still pins the Shade commit above, and release v0.5.3 still resolves to it. Shade MCP main at c7a69c24ad85d1a722d8827c039b87f3aee72331 differs from the pin only in dependency bumps to package.json and package-lock.json, so the pin stays. The contract now states per-instance scoping of private global_ state and sizing parameters in the input, output and texture naming rules, adds the optional Sink.onError method, states the default manager's reporter, and names db2173e3 in its snapshot block. A compile of repeated reactionDiffusion and watercolor instances produced global_rd_state_chain_0, global_rd_state_chain_0_node_1, zoom_chain_0 and zoom_chain_0_node_1, and compileGraph() reproduced the worked example's pass array. A fresh npx installation of the configured pin loaded SDK 1.32.2 and initialized server shade-mcp / 0.5.3 / 2025-11-25 with 21 tools. One configured-pin session on an Apple silicon macOS host with headless Chromium ran all eight worked calls without MCP isError. Every response matched the recorded transcript except the benchmark timings, which now hold this session's values; parity returned maxDiff=0 with frames: 8 and stateReset: true. Noisemaker checks: node scripts/run-js-tests.js --skip-parity (exit 0), npm run lint (exit 0), the four shaders/tests/test_contract_* tests (exit 0), node shaders/tests/test_stateful_instances.js (exit 0) and node --test test/mcp-config.test.js (exit 0). Candidate publication 901c6d25cfcc26785366a43826a85473d4717518 passed JavaScript CI (43433811701), CodeQL (43433810352) and site dispatch (43433811698), and static site release 43433830177 deployed it. The complete served llms-full.txt matched its published bytes with SHA-256 1f3bf844cd9fd906a4a6396ea9a7a9a99631522ba3c3366dcea36f805a40c34f. The checkpoint now matches the audited runtime snapshots above.
    • 2026-11-06 — completed delivery pair: Noisemaker 6f98d26ab84f17ba7fb154bb74252d6362736f89 and Shade MCP 63e53ad89dc43526335da6e947ed31667929e43b. Audited the Noisemaker range c0aa621a..6f98d26a and the Shade MCP range 43d2b942..63e53ad8. The configured immutable MCP pin in .mcp.json is that Shade commit. Release v0.5.3 resolves to the same Shade source. Its downloaded shade-mcp-dist.tar.gz has SHA-256 66d61d3e5d4ecf6fb5c9524fb25dc1496dc523c9397d2da60a94d75757dab804. The automatic Noisemaker import at d4613d7d matches the extracted release in all four vendored directories, and the vendored tree is unchanged through 6f98d26a. A fresh npx installation of the configured pin built a dist byte-identical to the release, loaded SDK 1.32.2 (the lockfile also pins 1.32.2), and initialized server shade-mcp / 0.5.3 / 2025-11-25 with 21 tools. Separate initializations echoed 2025-11-25, 2025-06-18, 2025-03-26, 2024-11-05 and 2024-10-07, and answered 2025-11-25 to an unlisted version. One configured-pin session ran all eight worked calls without MCP isError, on an Apple silicon macOS host with headless Chromium; the WebGPU adapter identity was not captured. Both compilers succeeded, the WebGL2 render was nonblank, parity returned maxDiff=0 with frames: 8 and stateReset: true, and the benchmark measured 60.48 FPS against a 30 FPS target. In the same session runDslProgram rendered an asymmetric uvMap program on WebGL2 and WebGPU; the two contact sheets were byte-identical, with high uv.y at the top. The Noisemaker range changed the WebGPU canvas upload (a caller-passed context2d, transparent black for a canvas without a context), WebGPU device-loss recovery, demo mesh paths and effect shaders. The Shade MCP range added the program rebuild, the fixed-frame parity capture with frames and stateReset, uniform reset mode, the neq gate start value and the WebGPU runDslProgram row flip. The contract now states the canvas-upload change and every Shade MCP runtime change; the six-frame live warmup and its 30000 ms bound are removed. The device-loss, mesh-path and effect-shader changes touch no contract statement. Shade checks ran on source exported with git archive 63e53ad: npm ci (exit 0; its prepare build passed check:dist and check:drop, and the built dist is byte-identical to the release), npm run typecheck (exit 0), npm test (exit 0), NOISEMAKER=<noisemaker checkout> node scripts/browser-smoke.mjs (exit 0, every check passed, including the genuine WebGPU DSL render and the WebGPU grid orientation) and NOISEMAKER=<noisemaker checkout> PORTABLE=<portable source> node scripts/authored-dsl-smoke.mjs (exit 0, with Portable 86c638a1). Noisemaker checks: sh scripts/test-shade-mcp (exit 0), node shaders/tests/test_contract_delivery_pair.mjs (exit 0), node --test test/mcp-config.test.js (exit 0), node scripts/run-js-tests.js --skip-parity (exit 0), npm run lint (exit 0) and node shaders/tests/test_texture_upload_source_context.mjs (exit 0, both backends). Candidate publication 8902e6951c1c5915a23aa862900015c54c4d8f2d passed JavaScript CI (42299528825), main-push analysis (42299527879), site dispatch (42299528785) and downstream notification (42299528705). The live deployment reported that candidate SHA, and the complete served llms-full.txt matched its published bytes with SHA-256 231d84165fe944b7b0b6153f68a084f4a74e5f0bad7bae65c6921f04aa733ca3. The checkpoint now matches the audited runtime snapshots above.
    • 2026-11-01 — completed delivery pair: Noisemaker c0aa621a53976729d9d598295701f895a68badb7 and Shade MCP 43d2b942cdba795c99e4a7a820570987d0b5437a. The configured immutable MCP pin is that Shade commit. Release v0.5.2 resolves to the same Shade source; the Noisemaker import at c0aa621a matches its downloaded release in all vendored directories. The configured Git installation built and initialized server 0.5.2, protocol 2025-11-25, with SDK 1.32.2. The worked requests were recaptured in one session: WebGL2 and WebGPU compilation and rendering succeeded, and pixel parity returned zero maximum and mean difference. testNoPassthrough reported skipped: Not a filter effect for the generator; this is not a filter acceptance result. The immutable-pin and delivery/parity contract tests pass. The checkpoint-to-source audit corrected the stale replacement-cloning statement and removed an absent configuration-file citation. The prescribed scripts/test-shade-mcp consumer gate passed with its unchanged Linux browser driver; octaveWarp returned exact GLSL/WGSL parity. Shade's browser-smoke.mjs and authored-dsl-smoke.mjs passed against this pair, including genuine WebGPU rendering. A separate NVIDIA OpenGL override experiment failed octaveWarp parity with maximum difference 233; its WebGPU adapter identity was not captured, and this pass makes no claim of cross-driver octaveWarp parity. Candidate validation also includes node scripts/run-js-tests.js --skip-parity and the repository ESLint targets; both passed. Candidate publication f9176fa8c309d565bb58d506920ca2b8d8b92cc7 passed JavaScript CI (41274838925), main-push analysis (41274838920), site dispatch (41274838931) and downstream notification (41274838945). The live deployment reported that candidate SHA, and the complete served llms-full.txt matched its published bytes with SHA-256 77d1df94a1189aa070e294cc31bbfac21261d69f7807c5a904440df082cf4aec. The checkpoint now matches the audited runtime snapshots above.
    • 2026-10-15 — completed delivery pair for work queue item 1116. Audited the complete Noisemaker range b319e794..7e6c6130 and Shade MCP range f36a5f62..830603b5. The source identities are Noisemaker 7e6c613079b8074a7f9b0f0f7b3c7d8d87123f17 and Shade MCP 830603b5bc112460eb3cf0b22f28e7dd51f84895. The configured MCP pin in .mcp.json is 830603b5bc112460eb3cf0b22f28e7dd51f84895. Release v0.4.13 points to that Shade source. Its downloaded shade-mcp-dist.tar.gz has SHA-256 e4b3069409fd62da96a0ca74f33d0861de4d5e21b424afb9ff120a0c5f067ffe. The automatic Noisemaker import at 05793839 matches the extracted release in all four vendored directories. The Noisemaker range retired parameter aliases, corrected shader tile behavior, and updated both delivery paths. Shade corrected DSL exemplars and enabled WebGPU browser capability for testPixelParity. The contract now states this parity launch behavior and records the current configured-pin worked session. Shade build, typecheck, 345 unit tests, standalone-drop check, and real Noisemaker viewer smoke passed, including a WebGPU DSL render. Shade quality CI, release CI, and the Noisemaker vendor import passed. Noisemaker non-parity JS tests and lint passed. The delivered vendor passed the WebGL2 and WebGPU render gates. The configured-pin MCP initialization negotiated shade-mcp / 0.4.13 / 2025-11-25 and listed 18 tools. One configured-pin session ran all eight worked calls. Both compilers succeeded. The WebGL2 render was nonblank. WebGPU parity returned maxDiff=0 with two non-solid frames. The benchmark measured 60.35 FPS against a 30 FPS target. Noisemaker publication commit 13eb5fb6 adds this contract and its regression guard to the source pair. Its exact-source JavaScript, Shaders, Site, Docs site, Downstream, and code analysis workflows passed. The site, docs site, and shader library release workflows passed for that commit. noisemaker.app and docs.noisemaker.app both serve deployment metadata naming 13eb5fb6. The live /llms-full.txt response is HTTP 200 and byte-identical to the published contract (SHA-256 19152bbcf58db3be6a5cbd15267679128d004d656626877ab1919717ba43cd33). Shader library version 1.0.295 serves metadata naming 13eb5fb6. Every one of its 229 manifest files returned HTTP 200 with the listed SHA-256 from /1.0.295/, /1.0/, and /1/. The complete pair is now delivered. This ledger-only commit advances the checkpoint to the two source SHAs in the contract snapshot block.
    • 2026-10-03 — completed compatibility pass (work queue item 794): audited Noisemaker 5ff0445939d0238c51f0ab93a30baa238a6509ec..bed7da28c32063b5837d4050843fed60847b602a and the trigger range d143cb51..058d15dc; Shade MCP remained at 8e0c80a436fbe1ead57bdde3ff3b9f605d34f342. The changed browser shader is filter/octaveWarp: its GLSL and WGSL coordinate/seed hash conversion changed, and its 64x48 parity attestation was added. The other watched changes only correct Shade-harness provenance comments and regression guards. No DSL, effect definition, viewer, or tool schema changed, so the contract's behavior statements, traceability matrix, and three open gaps remain valid. The configured immutable MCP pin is github:noisedeck/shade-mcp#8e0c80a436fbe1ead57bdde3ff3b9f605d34f342. Shade release v0.3.1 points to that source and publishes shade-mcp-dist.tar.gz with SHA-256 e6a29674f356350b13c2fc009c02b5fa4e6e53d7b42aa18642028a2386cfbeb7; all 34 vendored files imported at Noisemaker 10386a1e remain unchanged. Noisemaker JS tests and lint passed, as did the 206-effect structure scan. The delivered vendor compiled and rendered octaveWarp with confirmed effect/backend identity on WebGL2 and WebGPU; its pixel parity had maxDiff=0. Shade typecheck, 257 tests, build/drop checks, and real viewer smoke (including WebGPU DSL) passed. One configured-pin MCP session negotiated shade-mcp / 0.3.1 / 2025-06-18, listed 18 tools, and repeated the eight worked synth/testPattern calls in llms-full.txt; it also checked octaveWarp structure and compiled/rendered it on both backends with nonblank frames. The worked render returned 90x90 and warned that the requested 256x256 resolution was not honored. The worked benchmark reached 60.76 FPS at the measured 90x90 resolution.
    • 2026-10-03 — completed compatibility pass (work queue item 774): audited both complete checkpoint ranges and the trigger ranges through Noisemaker source 5ff0445939d0238c51f0ab93a30baa238a6509ec and Shade MCP source 8e0c80a436fbe1ead57bdde3ff3b9f605d34f342. Noisemaker's degauss tile-coordinate and WGSL uniform changes are compatible with Shade's parser and renderer. Shade now guards in-flight index invalidation and confirms effect/backend/graph identity in browser tools, reads WebGPU pixels, reports per-uniform measurements, compares filter output with its consumed input, and annotates resolution mismatch. The immutable configured MCP pin is github:noisedeck/shade-mcp#8e0c80a436fbe1ead57bdde3ff3b9f605d34f342. Shade release v0.3.1 points to that source; its published tarball SHA-256 is e6a29674f356350b13c2fc009c02b5fa4e6e53d7b42aa18642028a2386cfbeb7. The automatic Noisemaker import at 10386a1e is byte-identical to all 34 relevant files in that tarball; the delivered vendor and configured pin therefore match Shade source. Noisemaker updated only the pin, affected vendored-source regression guards, and the current contract. Local Noisemaker non-parity JS tests, lint, 206-effect structure gate, and delivered-vendor degauss compile/render checks on WebGL2 and WebGPU passed with page-confirmed effect/backend identity. Degauss parity matched at epsilon zero. Shade typecheck, 257 tests, build/drop checks, and real viewer smoke (including WebGPU DSL) passed. One configured-pin MCP session negotiated shade-mcp / 0.3.1 / 2025-06-18, listed 18 tools, and ran the eight worked synth/testPattern calls recorded in llms-full.txt: structure, both backend compiles, render, uniform responsiveness, expected nonfilter passthrough skip, pixel parity, and benchmark. A second session compiled and rendered degauss on WebGPU and confirmed its WebGL2 input/output difference. The worked render returned 90x90 with a warning against the requested 256x256; the contract records the measured size. Shade release CI, Noisemaker vendor CI, Shaders CI, Portable CI, and the Shade deployment passed. The final Noisemaker commit's CI and noisemaker.app delivery are checked separately after publication. The contract removes resolved GAP-010, GAP-011, GAP-012, GAP-019, GAP-021, GAP-024, and GAP-029; GAP-015, GAP-017, and GAP-020 remain open.
    • 2026-10-03 — delivery-blocked compatibility pass (work queue item 744): audited noisemaker e9a357401f281405b5bf46e324238a3e160f1cd9..378c8bf80e5f61efeeaa479b0187ef92a242ff03 and shade-mcp 6a7e2540b211efd2bd2f5be40aa8c2f1c0b0476d..0d7c071a0cd03adece1e5c3b7e19e15de3db30b8. The Noisemaker changes include own-choice validation, uniform aliases, renderer registration, and reseed-only UI metadata; Shade's newer browser tools bind requested effect, backend, and resolution to the rendered page and repair WebGPU readback. The tested source pair is noisemaker 378c8bf8 and shade-mcp 0d7c071; Shade's lifecycle-test mock isolation was the only edit needed in this pass. Noisemaker JS tests (--skip-parity), lint, and both 206-effect structure gates passed; Shade typecheck, 255 tests, build/drop checks, and the real viewer smoke (including WebGPU rendering) passed. Shade CI and browser quality passed at 0d7c071. The configured MCP pin remains github:noisedeck/shade-mcp#6a7e2540b211efd2bd2f5be40aa8c2f1c0b0476d: one fresh session negotiated shade-mcp / 0.3.0 / 2025-06-18, listed 18 tools, and compiled synth/cell on WebGL2. The vendored vendor/shade-mcp/ subtrees are byte-identical to the v0.3.0 release asset (tarball SHA-256 f8c88ab90c297e1422663e527fb7ef9ec4528de2f8285dfcc0e23bafc57b89ea), whose source is 6a7e254. Shade source 0d7c071 has package version 0.3.1 but no v0.3.1 release; the newer source is therefore absent from both configured and vendored delivery. The llms-full.txt source snapshots and this section's checkpoints remain at the last completed pair pending release, vendor refresh, immutable pin update, and delivery qualification.
    • 2026-10-02 — compatibility pass auditing noisemaker e9a357401f281405b5bf46e324238a3e160f1cd9..1fd893483c83f7602416cf8e9c48c9605a8f202c and shade-mcp 6a7e2540b211efd2bd2f5be40aa8c2f1c0b0476d..bdf186638b26fad4811b7d21aec25b3571f15040 (work queue item 743; both ranges diffed from the local checkouts against this checkpoint rather than trusted). The noisemaker range's watched-roots drift is empty — its commits are this ledger's and the gap register's own documentation (f8d5940..1fd8934, including the register's rework into the "Upstream Shade MCP defects" list mapping the ten former gap ids onto shade-mcp issues 27-35). The shade range is five commits: bc4a7fa (testUniformResponsiveness now reports overall ok only when every tested uniform affected output, error/skipped otherwise, with measured per-uniform default/test values and luma/max-channel deltas against the stated 0.002 threshold — the upstream fix for issues #27, the source of the former GAP-010/011), 2d85160 (regression-test isolation only), 76d8d79 (version bump 0.3.1), 1aea08a (renderEffectFrame, testUniformResponsiveness and testNoPassthrough read pixels through the backend's async readPixels(textureId) with Float32 conversion and WebGPU row flip when there is no GL context, instead of failing — the upstream fix for issue #28, the source of the former GAP-012), and 987b14d (the three verbs drain queue.onSubmittedWorkDone() before reading the surface and missing-renderer errors now name the pipeline backend instead of unknown). One further commit, bdf1866, is the dependabot npm-minor-major dependency bump alone (package.json + package-lock.json, no source change); the battery below was re-run at bdf1866. None of this is delivered: git ls-remote --tags shows the newest tag is still v0.3.0 = 6a7e2540b211efd2bd2f5be40aa8c2f1c0b0476d (no v0.3.1 tag or release exists — corroborated by an independent unauthenticated GitHub API lookup: refs/tags/v0.3.1 returns 404, the releases list ends at v0.3.0, and releases/latest reports tag_name v0.3.0; receipt shade-tags-public-api.txt), the .mcp.json pin is unchanged, and the vendored vendor/shade-mcp/ bundle therefore still matches the v0.3.0 release — so no vendored module, tool schema, or public result shape changed for the delivered pair, and llms-full.txt stays accurate for it. Four identities of this pass: noisemaker source 1fd8934, shade source bdf1866, MCP pin github:noisedeck/shade-mcp#6a7e2540b211efd2bd2f5be40aa8c2f1c0b0476d (= release v0.3.0), vendor delivery v0.3.0 (unchanged). No integration code change was required on either side: noisemaker's harness mirrors and wrappers (uniform-status.js aggregate over the unchanged tested_uniforms strings, frame-readback.js wrapper path, and the additive backend field on error results) remain valid at the delivered pin. The audited shade refactor is reconciled in this repository's own test guard: shaders/tests/test_passthrough_input.js pinned the vendored testNoPassthrough's literal renderer.render(0)/renderer.render(1) calls, and 987b14d's verb renders inside a readFrame(t) helper; the guard now accepts both eras while still requiring two distinct compared render times, so the vendor refresh that delivers the issue-#28 fix does not turn the suite red. Both checkpoints and the contract snapshot block remain unchanged at noisemaker e9a3574 / shade-mcp 6a7e254: the v0.3.1 release (tag push), the pull-shade-mcp vendor refresh, and the .mcp.json pin bump are operator-authorized publication actions and stay recorded as the pending blocker. Validation battery on the exact pair: noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 9 suites, 213 tests, 0 failures) and lint (exit 0); Shade typecheck (0 errors), unit tests (28 files / 187 tests pass) and build (npm run build, self-contained dist advertising 0.3.1) at bdf1866; structure check via the vendored harness (npm run test:shaders:structure, 1258 documented parameters across 200 effects); WebGL2 render check (npm run test:shaders:render:webgl2, exit 0, 1/1, temporal_diff=0.216246, mesh/GLSL-orientation/overlay sub-tests; the external-texture test's WebGPU leg skipped after backend fallback — not GPU qualification); WebGPU render check (node shaders/tests/test-harness.js --effects synth/noise --backend webgpu --verbose, 1/1 pass, temporal_diff=0.216246, page-confirmed identity_check match, backend observed webgpu, generation 2→4, backend_switch match; SwiftShader selected with SHADE_SWIFTSHADER=1 and VK_ICD_FILENAMES/VK_DRIVER_FILES at the Playwright chromium_headless_shell-1243 bundle's vk_swiftshader_icd.json); and Shade's real viewer smoke test (NOISEMAKER=<this checkout> node scripts/browser-smoke.mjs from the shade-mcp checkout at bdf1866, all available checks OK including the genuine WebGPU DSL render — source- stamped re-run against the exact candidates (noisemaker 41d1ead, 0 dirty files; shade-mcp bdf1866, 0 dirty files; dist advertising 0.3.1 built at shade HEAD) with SHADE_SWIFTSHADER=1 and VK_ICD_FILENAMES/ VK_DRIVER_FILES at the chromium_headless_shell-1243 bundle's vk_swiftshader_icd.json; this is software-WebGPU (SwiftShader) evidence, not physical-GPU qualification; receipt shade-smoke-stamped.log). MCP session re-captured from one stdio session at the tested immutable pin 6a7e2540b211efd2bd2f5be40aa8c2f1c0b0476d: server shade-mcp version 0.3.0, protocol 2025-06-18 when requested, the 18-tool tools/list, and a worked timed renderEffectFrame {effect_id: "synth/noise", time: 0.5, warmup_frames: 3} returning status ok, WebGL2, 1024x1024, mean_rgb 0.457433/0.443662/0.445664 with 1001 unique sampled colors — the same values as the 2026-10-01 pin capture, so the pinned verb is deterministic across sessions. The same session's WebGPU captures at that pin (mcp-recapture-v030-webgpu.json, driver mcp-recapture-webgpu.mjs, SwiftShader env): runDslProgram {backend: "webgpu", dsl: "search synth\nnoise(10).write(o0)"} returns status ok, backend observed WebGPU, 960x540, frame 1 mean_rgb 0.416145/0.437018/0.469605 with 1000 unique sampled colors — WebGPU rendering works through the configured pin's batch verb; but the pinned renderEffectFrame with backend: "webgpu" still returns {"status": "error", "backend": "unknown", "error": "Failed to read pixels"} — the delivered pin's GAP-012 defect (issue #28) reproduced live, exactly as llms-full.txt states for the delivered pair. That failed configured-pin check is an outstanding blocker recorded here: the item stays pending on it until the delivered pin's renderEffectFrame passes a WebGPU render (which the unreleased 987b14d fix provides once released, vendored and pinned); shaders/tests/test_frame_readback.js now source-guards the vendored verb's GL-only failure path, so the suite turns red exactly when a vendor refresh delivers the upstream fix and the wrapper's WebGPU path must be re-derived. Receipts in the automation archive ledger-743: mcp-recapture-v030.json, mcp-recapture-v030-webgpu.json, mcp-recapture.mjs, mcp-recapture-webgpu.mjs, test and smoke logs).
    • 2026-10-01 — delivery reconciliation: Shade MCP v0.3.0 = 6a7e2540b211efd2bd2f5be40aa8c2f1c0b0476d was released upstream (6a7e254, one commit over the audited 687dbbd: the version bump), the .mcp.json pin was bumped to it (df62a43), and the existing vendor workflow delivered the release bundle in Noisemaker deb4b41 (recorded by the operator in f76c372). This pass re-verified the delivery itself: the freshly downloaded v0.3.0 tarball shade-mcp-dist.tar.gz has sha256 f8c88ab90c297e1422663e527fb7ef9ec4528de2f8285dfcc0e23bafc57b89ea and the vendored vendor/shade-mcp/ trees (harness/ai/formats/analysis) are byte-identical to it (receipt vendor-check-030/tarball-sha256.txt); the immutable pin in .mcp.json selects exactly the released source commit. Both checkpoints and the contract snapshot block now advance together to noisemaker e9a3574 / shade-mcp 6a7e254: GAP-014's row is deleted with every reference (the timed-capture hang and the pinned time-0 redraw are fixed in the delivered verb — verified live at the pin: renderEffectFrame {time: 0.5, warmup_frames: 3} returns status ok, WebGL2, mean_rgb 0.457433/0.443662/0.445664 with 1001 unique sampled colors, byte-identical to this viewer's deterministic per-time frame recorded in the 2026-09-29 entry, so the redraw honors the requested time); the runDslProgram section now documents the delivered batch verb (fresh blank page, SHADE_DSL_RENDERER_MODULE/SHADE_DSL_ASSETS_BASE/ SHADE_DSL_USE_BUNDLES, effects Portable package selection, per-capture full ImageMetrics, PNG contact sheet as MCP image content, live re-capture status ok on WebGL2 with surfaces ["o7"], frames [1,120,600], grid 1014x212, content types ["text","image"], and frame.image_uri always null); the timed/untimed renderEffectFrame capture semantics and the MCP image transport (toolResult() appends image blocks after the JSON text block; the isError rule unchanged from 98253a8c) are documented; GAP-020/021/024/029 rows were re-verified against the delivered release (src/tools/tool-result.ts unchanged since 98253a8c; src/knowledge/shared-instances.ts unchanged since d6b8ab0; the GL-only readback and the unconfirmed viewport-vs-canvas dimension handling unchanged in the released render.ts); the open-gap count is corrected to 10; the matrix cell and worked-instance audit were extended; and the register count sentence, snapshot block, and this checkpoint were advanced together. The released delta over the previously audited shade-mcp 687dbbd is only the version bump (git diff 687dbbd..6a7e254: package.json/package-lock.json), and Noisemaker's watched-roots drift since cb22a05 is only the inert favicon-link commit e9a3574 (demo/shaders/index.html + index.html icon link, plus the vendor/pin/docs delivery commits). Validation battery on the exact delivered pair: noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 9 suites, 213 tests, 0 failures) and lint (exit 0); Shade typecheck (0 errors), unit tests (26 files / 175 tests pass), and build (npm run build, self-contained dist advertising 0.3.0) at 6a7e254; structure check via the delivered vendored harness (npm run test:shaders:structure, 1258 documented parameters across 200 effects); WebGL2 render check (npm run test:shaders:render:webgl2, exit 0, 1/1 plus the mesh/GLSL-orientation/overlay/external-texture sub-tests); WebGPU render check (node shaders/tests/test-harness.js --effects synth/noise --backend webgpu --verbose, 1/1 pass, temporal_diff=0.216246, page-confirmed identity_check match, backend observed webgpu, generation 2→4, backend_switch match; SwiftShader selected with SHADE_SWIFTSHADER=1 and VK_DRIVER_FILES at the Playwright chromium_headless_shell-1243 bundle's vk_swiftshader_icd.json); and Shade's real viewer smoke test (NOISEMAKER=<this checkout> node scripts/browser-smoke.mjs from the shade-mcp checkout at 6a7e254, all available checks OK through the genuine WebGPU DSL render and the virtual-route cleanup). Re-captured MCP initialization (server shade-mcp version 0.3.0, protocol 2025-06-18 when requested, 2024-11-05 for the legacy client used in the session captures), the 18-tool tools/list, listEffects (210 effects), worked analyzeEffect/checkEffectStructure (synth/noise: lossy projection; status ok, passCount 1), the delivered runDslProgram batch call, and the timed renderEffectFrame call from stdio sessions at the tested immutable pin 6a7e2540b211efd2bd2f5be40aa8c2f1c0b0476d (receipts pin-capture-030.json, pin-capture-2-030.json). Archived run evidence (automation archive ledger-723): *-030.log, *-030.json, pin-capture-030.mjs, pin-capture-2-030.mjs, and vendor-check-030/.
    • 2026-10-01 — Noise Factor AI delivery: Shade MCP v0.3.0 is released from 6a7e2540b211efd2bd2f5be40aa8c2f1c0b0476d; exact-source quality CI and release passed. The immutable .mcp.json pin now selects that source. The existing vendor workflow delivered it in Noisemaker deb4b413823d1ba79332337b29d53574f1b62e18; its shader/GPU CI passed. All 34 vendored files match the release asset (SHA-256 f8c88ab90c297e1422663e527fb7ef9ec4528de2f8285dfcc0e23bafc57b89ea). Noisemaker source at df62a434 passed non-parity JS, lint, structure, WebGL2 rendering and page-confirmed WebGPU noise rendering. The separate external-texture test skipped its WebGPU leg after fallback; it is not GPU qualification. The configured Git pin negotiated shade-mcp / 0.3.0 / 2025-06-18, listed 18 tools, and completed timed captures with distinct time-0/time-0.5/live images plus WebGL2/WebGPU DSL grids. The downloaded release also passed built-in and two-effect authored browser smokes on both hardware backends. Delivery is established; the shared checkpoint remains 73c15be / 00340b1 pending reconciliation of the contract's batch schemas, MCP image transport, timed-render guidance, matrix and GAP-014/020/024/029. GAP-012's direct effect-render WebGPU readback and the other browser verbs' identity limitations remain open. This delivery does not advance the llms-full.txt snapshots or certify those remaining paths.
    • 2026-10-01 — compatibility pass auditing noisemaker e24c844f8dada85551ab084f41db8944fbc176c8..cb22a05eff9afed99fcf22a482b944c26f43e814 and shade-mcp 79c2e011a2a8088bf88462aae93ea7e2e135de10..687dbbd7d37a1220067ef08dacc350abfe62ebd5 (work queue item 723, job ledger-llms-contract; both trigger ranges were diffed from the local checkouts against this checkpoint rather than trusted. Noisemaker's forced start e24c844 is a linear ancestor: its watched-root commits dd4606e/a0e9bbf/e24c844 (GAP-007) and ed47815 (GAP-010), the ledger's own documentation commits through f5ca07c, and the Portable-registration source step cb22a05 are already audited in the entries above; since cb22a05 the watched-roots drift is empty — only the dependabot brace-expansion bump 1b088ad, outside the watched roots. shade-mcp's range extends the already-audited paired candidate c641503 by one commit, 687dbbd, which touches only scripts/authored-dsl-smoke.mjs — not a vendored module, tool schema, result shape, exported harness surface, or release artifact — so no integration-code change or new regression was required on either side; no public tool schema or result shape changed. llms-full.txt's runDslProgram pending-delivery note was refreshed to the extended unreleased range and now records c641503's optional effects input (Portable package selection, at most 16, registered live through registerPortableEffect, matching the cb22a05 renderer registration audited above). Delivery remains the blocker exactly as recorded for the prior entries: the newest Shade release is still v0.2.3 = cbcab33363851016f65391fbb9ef71d66729c07f (git ls-remote --tags, receipt vendor-check/shade-tags-remote.txt), the .mcp.json pin is unchanged, and the vendored vendor/shade-mcp/ bundle was re-verified byte-identical to the freshly downloaded v0.2.3 tarball shade-mcp-dist.tar.gz (sha256 4256ea5c695f15b3eec292f599a1ae1aada03a2628e33db8b3c9864a6b11f93f; all four module trees identical, vendor-check/tarball-sha256.txt). The batch verb, c641503's Portable effects input, and 687dbbd's smoke-script hardening all wait on an operator-authorized Shade release, the existing pull-shade-mcp vendor refresh, and the .mcp.json pin bump. Both checkpoints and the contract snapshot block remain unchanged at noisemaker 73c15be / shade-mcp 00340b1. Validation battery on the exact pair: noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 9 suites, 213 tests, 0 failures, including the cb22a05 portable-registration suite) and lint (exit 0); Shade typecheck (0 errors), unit tests (26 files / 175 tests pass), and build (npm run build, self-contained dist advertising 0.2.3); structure check via the vendored harness (npm run test:shaders:structure, 1258 documented parameters across 200 effects); WebGL2 render check (npm run test:shaders:render:webgl2, exit 0, 1/1, temporal_diff=0.216246, mesh/GLSL-orientation/overlay/external-texture sub-tests pass); WebGPU render check (node shaders/tests/test-harness.js --effects synth/noise --backend webgpu --verbose, 1/1 pass, temporal_diff=0.216246, page-confirmed identity_check match, backend observed webgpu, generation 2→4, backend_switch match; SwiftShader selected via SHADE_SWIFTSHADER=1 and VK_DRIVER_FILES at the Playwright chromium_headless_shell-1243 bundle's vk_swiftshader_icd.json — this container's Vulkan loader 1.4.360 ignores the legacy VK_ICD_FILENAMES variable, and without VK_DRIVER_FILES the SwANGLE init fails with VK_KHR_surface unsupported); and Shade's real viewer smoke test (NOISEMAKER=<this checkout> node scripts/browser-smoke.mjs from the shade-mcp checkout at 687dbbd, all 14 checks OK: compileEffect, renderEffectFrame, the DSL grid, invalid-DSL cleanup, defaults-after-failure, concurrency isolation, setContent module import, contact-sheet pixels, swallowed-renderer-error and backend-fallback detection, isolated secure loopback DSL page, a genuine WebGPU DSL render, and the virtual-route cleanup). Re-captured MCP initialization (server shade-mcp version 0.2.3, protocol 2024-11-05), the 18-tool tools/list, and the worked untimed runDslProgram (search synth/read(o0).write(o7)/render(o7)) from one stdio session at the immutable pin cbcab33363851016f65391fbb9ef71d66729c07f: status ok, the pin's old one-frame shape with frame.image_uri: null, keys backend/console_errors/frame/ metrics/status, WebGL2, mean_rgb 0.232998/0.407600/0.448649, 929 unique sampled colors — the pin's documented live-frame read, nondeterministic with the capture moment. Archived run evidence (automation archive ledger-723): nm-tests.log, nm-lint.log, shade-typecheck.log, shade-tests.log, shade-build.log, structure.log, webgl2-render.log, webgpu-render.log, browser-smoke.log, pin-capture.log, pin-capture.mjs, pin-capture.json, and vendor-check/.
    • 2026-10-01 — Noise Factor AI authored-effect source step, above Noisemaker b30d506f and Shade MCP 79c2e01. The publication carrying this entry adds CanvasRenderer.registerPortableEffect for initial registration of loaded Portable definitions in a fresh JavaScript realm. It retains Effect lifecycle identity, render metadata, aliases and enum behavior; protects bare built-in names; rejects duplicate names, malformed registration inputs, and prototype-control keys. Shade's paired candidate adds explicit package selection to its fresh DSL batch renderer. Local registration regressions, prescribed non-parity JS, shader language/runtime, lint, Shade typecheck, 175 unit tests and build passed. Real MCP browser checks covered two authored effects on WebGL2 and genuine WebGPU, filter contribution, repeatable frames, flat effect layout, isolated calls and malformed/missing shader failures; the existing built-in browser smoke also passed. These are source checks, not delivery qualification. Exact published-source CI, the released renderer, a Shade release, immutable MCP pin and vendored harness reconciliation remain pending. The last completed checkpoint and contract snapshot stay unchanged.
    • 2026-10-01 — compatibility pass auditing noisemaker 4f5e0d28bdc155700393c314e9a5aafcc4da91fd..ed478159e5a31870c318be05ff755e533c754126 and shade-mcp 153eebd3b08c849a3fbf20d43527ae48a9dff746..79c2e011a2a8088bf88462aae93ea7e2e135de10 (work queue item 699, job ledger-llms-contract; both trigger ranges were diffed from the local checkouts against this checkpoint rather than trusted — noisemaker's force-flagged deliveries resolve to the linear watched-root drift c2a19c7→dd4606e→a0e9bbf→e24c844 (GAP-007) and e105344→16c1997→ed47815 (GAP-010), plus this ledger/contract's own documentation commits through 89523ee; shade-mcp's range is the two commits 7b7f6cb, 79c2e01, linear). Noisemaker→Shade: the GAP-007 and GAP-010 statements that those commits added to llms-full.txt (the DiagnosticCollector codes and their queryable backend.diagnostics / pipeline.diagnostics records, the 'input'/'resolution' recognized-dimension exception whose absence had made dd4606e's fallback diagnostic flag filter/grade's legitimate 'input' keyword and fail the CI GPU parity battery, and the mirrored resolveUniformGateStatus() harness gate) were re-verified statement-by-statement against the live source (shaders/src/runtime/backends/diagnostics.js, webgl2.js, webgpu.js, pipeline.js, shaders/tests/uniform-status.js, test-harness.js). Shade's parsers, analysis, knowledge, and browser tools reference none of the new surfaces (grep of Shade src/ and scripts/ for backend.diagnostics/DiagnosticCollector/ resolveDimension/uniform-status: zero relevant hits — Shade's browser wrappers keep consuming the unchanged thrown legacy ShaderDiagnostic surface, whose code/detail fields dd4606e preserved), so no integration-code change or new Shade regression was required and no delivered tool schema or result shape changed. Shade→Noisemaker: upstream 7b7f6cb/79c2e01 reshape runDslProgram into a fresh-blank-page per-surface batch grid verb with three new SHADE_DSL_* config keys whose defaults resolve under SHADE_VIEWER_ROOT (verified compatible with this checkout's repository root and .mcp.json, which sets none of them), and widen tool-result.ts with optional MCP image content while the isError-marking rule is unchanged. Recorded in llms-full.txt: the runDslProgram section gained a pending-delivery note (the delivered pin's old one-frame shape re-verified live; the batch shape is unreleased), the GAP-020 row's upstream evidence was refreshed (file last logic change still 98253a8c, 2026-08-10), and the register-count sentence was corrected from 10 to the true 11 narrowed/blocked rows — the count went stale when the 2026-09-29 pass narrowed GAP-029's row without bumping it, and closing GAP-007 preserved the off-by-one; this ledger's scope sentence now matches the register. Delivery remains the blocker exactly as recorded for the prior pass: the newest Shade release is still v0.2.3 = cbcab33363851016f65391fbb9ef71d66729c07f (git ls-remote --tags, receipt vendor-source-verification.json — annotated tag a1faf90d peels to that commit), the .mcp.json pin is unchanged, and the vendored vendor/shade-mcp/ bundle was re-verified byte-identical to the freshly downloaded v0.2.3 tarball shade-mcp-dist.tar.gz (sha256 4256ea5c695f15b3eec292f599a1ae1aada03a2628e33db8b3c9864a6b11f93f; all four module trees identical, recorded per-tree in the same receipt) — the batch verb, GAP-014's upstream timed-render fixes, and the upstream GAP-012 leg all wait on an operator-authorized Shade release, the existing pull-shade-mcp vendor refresh, and the .mcp.json pin bump. Both checkpoints and the snapshot block are deliberately unchanged at noisemaker 73c15be / shade-mcp 00340b1. Validation battery on the exact pair: noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 8 suites, 206 tests, 0 failures) and lint (exit 0, re-run at this candidate commit with the receipt nm-lint-result.json recording the command, exit code, and candidate SHA); Shade MCP typecheck (0 errors) and unit tests (26 files / 174 tests pass, including the new config, dsl-batch, and tool-result regressions); structure check via the vendored harness (npm run test:shaders:structure, 1258 documented parameters across 200 effects); WebGL2 render check (npm run test:shaders:render:webgl2, exit 0, all PASS including the external-texture WebGL2 leg); WebGPU render check (node shaders/tests/test-harness.js --effects synth/noise --backend webgpu --verbose, 1/1 pass, temporal_diff=0.216246, page-confirmed identity_check match, backend observed webgpu, generation 2→4, backend_switch match; SHADE_SWIFTSHADER=1 with VK_ICD_FILENAMES/VK_DRIVER_FILES at the Playwright chromium_headless_shell-1243 bundle's vk_swiftshader_icd.json, PLAYWRIGHT_BROWSERS_PATH=<browser cache> holding both pinned builds — chromium-1243 for noisemaker, chromium-1234 for shade-mcp); and Shade's real viewer smoke test (NOISEMAKER=<this checkout> node scripts/browser-smoke.mjs from the shade-mcp checkout at 79c2e01, all 14 checks OK: compileEffect, renderEffectFrame, the DSL grid with per-surface metrics and a verified PNG, invalid-DSL cleanup, defaults-after-failure, concurrency isolation, setContent module import, contact-sheet pixel content, swallowed-renderer-error and backend-fallback failure detection, isolated secure loopback DSL page, a genuine WebGPU DSL render, and the virtual-route cleanup). Re-captured MCP initialization (server shade-mcp version 0.2.3, protocol 2024-11-05), the 18-tool tools/list, and the worked calls from one stdio session at the immutable pin cbcab33363851016f65391fbb9ef71d66729c07f: untimed renderEffectFrame (status ok, WebGL2, mean_rgb 0.456897/0.443613/0.445983, 1013 unique sampled colors — the pin's documented live-frame read) and the pin's runDslProgram (status ok, old one-frame shape, frame.image_uri: null, keys backend/console_errors/frame/metrics/status). Final-acceptance evidence additions (automation archive ledger-699): exact-source CI at the audited noisemaker source tip ed47815 observed via the public check-runs API (archived check-runs-ed47815.json) — 16 check runs, all completed with conclusion success or skipped, including shaders.yml's Shader tests and GPU tests at that SHA, release.yml's Publish release and bundle/standalone builds, docs-site.yml's Verify docs asset paths, and downstream.yml's notify (site.yml's own check at that SHA is the Dispatch static-site-release dispatch; the js.yml workflow does not run there by design — its path filter excludes commits touching only shaders/** and llms-full.txt); source-SHA mapping for the local logs (archived source-sha-mapping.md) — noisemaker's shaders//js//scripts//test// demo/ trees at capture time are byte-identical to ed47815 (git diff --stat ed478159..58953caa -- shaders/ js/ scripts/ test/ demo/ is empty; the intervening commits touch only documentation, ledgers, and dependency bumps) and the shade-mcp logs ran at the checkout tip 79c2e01 exactly; exact-commit CI at this documentation commit 58953caa0abb2071f0493d0021bfe2c6b5e029ac (archived check-runs-58953ca.json) — Dispatch static-site-release success, correctly the only check (a docs-only commit runs only the Site workflow, with the source CI carried at ed47815 above); and deployment (archived deployment-readback-58953ca.json and deployed-llms-full.txt) — https://noisemaker.app/deployment-meta.json serves git_hash: 58953caa0abb2071f0493d0021bfe2c6b5e029ac and the deployed https://noisemaker.app/llms-full.txt is byte-identical to this commit's llms-full.txt (sha256 90c5883dfddb3a51… on both), containing both of this pass's edits. Candidate-boundary clarifications from final acceptance: this job's shade-mcp candidate is unchanged published source (79c2e01 entered this item as an already-published audited input, and no shade-mcp commit belongs to this candidate) — the delivered pin keeps the old one-frame runDslProgram semantics and the upstream batch reshape is recorded above as pending delivery, so no shipped behavior changed through this job; reversing the already-published upstream commits is outside this pass, and their compatibility resolution is exactly the pending-delivery record. Likewise the Portable CDN/viewer qualification that ships inside upstream 79c2e01 (scripts/portable-browser-smoke.mjs and the README's Portable configuration) is a Portable-product concern delivered in upstream Shade before this item; this pass validated only the local Noisemaker-runtime DSL path against this checkout (verified defaults, no SHADE_DSL_* overrides) and makes no Portable claim. Archived run evidence (automation archive ledger-699): nm-tests.log, nm-lint.log, shade-typecheck.log, shade-tests.log, structure.log, webgl2-render.log, webgpu-render.log, browser-smoke.log, pin-capture.log, pin-capture.mjs, vendor-check/, check-runs-ed47815.json, check-runs-58953ca.json, source-sha-mapping.md, deployment-readback-58953ca.json, deployed-llms-full.txt, nm-lint-result.json, and vendor-source-verification.json (with shade-tags-remote.txt).
    • 2026-09-29 — compatibility pass auditing noisemaker 73c15be00d6888f4b5d2835d8e242ee9e840df45..682739066d3b74962febbdcdae85b5aa4d2e19f3 and shade-mcp 00340b148e109464b1a87d89ff29fc7622d384c9..07af4ffa392197748216f3a0beaebee365d44292 (work queue item 662, job ledger-llms-contract; both triggers were force-pushed/non-contiguous, so both ranges were diffed from the local checkouts against this checkpoint rather than trusting the queue ranges — noisemaker's observed deliveries 3e21906..a505910, a505910..6827390 plus this ledger's own bff453e/8fec3d0/4284359 documentation commits are linear inside the checkpoint-to-HEAD audit; shade-mcp's 00340b1..ed73367..ca94cfa..07af4ffa is linear). Noisemaker→Shade: the watched-roots diff is exactly shaders/src/runtime/external-input.js and shaders/tests/test_external_input.js — the GAP-032 fix commits a505910/6827390 (whose own llms-full.txt prose updates were re-verified statement-by-statement against the source: registration of the browser-selected device and default channels, per-requirement enumerateDevices() resolution with exact-id authority and unique-name matching re-checked at enable and every 60 ticks, per-channel splitter analysis, aggregate and per-channel rawReady from the bipolar time-domain mean, uncapturable-binding warning with min fallback, and the register shrinking to 11 open gaps; that re-verification itself then surfaced one gap in the shipped warning claim — an already-captured device lacking the requested channel (or a capture freshly opened by the sync whose track delivered fewer channels than the graph selects) returned null from the channel lookup and silently evaluated to min with no diagnostic. Fixed in this pass's follow-up commit: _syncCaptures() now runs a post-open validation pass (AudioInputManager._channelShortfall()) that warns for every selected requirement — default-device, id-selected, and name-selected alike — whose captured device does not expose the requested channel, while requirements with no capture at all keep their existing specific warnings. Final acceptance surfaced one more hole in that pass: the browser-selected capture is stored under the null key when the track reports no deviceId, and the default-device branch's this._deviceId guard skipped exactly that capture — fixed by dropping the guard (Map.get(null) resolves the deviceless capture), with a deviceless-fixture regression (default channel 2 against a 1-channel deviceless capture warns, in-range channel 1 stays silent and resolves); red-before holds (audio-deviceless-red.log 65/1 against the unpatched source, audio-channel-green.log 66/0 after). Focused regression in shaders/tests/test_external_input.js (default/id/name channel-3 requirements against 2-channel fixture captures warn, in-range channel-2 requirements do not, the in-range channel resolves and the beyond-range channel does not); red-before holds (the suite fails 64/1 against the unpatched source, 65/0 after) — see the archived audio-channel-red.log and audio-channel-green.log. The affected llms-full.txt prose now includes the channel shortfall in the uncapturable-binding enumeration in both the parameters/globals surface and the Diagnose section. Shade's parsers, analysis, knowledge, and browser tools reference none of the new surfaces (grep of Shade src/, scripts/, and the vendored vendor/shade-mcp/ for AudioInputManager/getAudioInputRequirements/registerDefaultChannels/ rawReady/enumerateDevices: zero hits outside the noisemaker demo page itself), so no integration-code change or new Shade regression was required and no tool schema or result shape changed. Shade→Noisemaker: upstream ed73367/ca94cfa fix the GAP-014 hang class in the upstream verb itself (warmup while the loop is live, warmup waits bounded at session.timeoutMs with a diagnostic, unpause in a finally) and 07af4ff adds a pre-readback redraw — the audit found that redraw issued unconditionally and as a hardcoded renderer.render(0), while noisemaker's viewer draws at the render() argument (CanvasRenderer.render(normalizedTime) → Pipeline.render(time) → updateGlobalUniforms(time, …)), so every timed capture drew the time-0 frame instead of the requested paused time and every untimed capture (historically a live-frame readback) was pinned to time 0 as well; corrected in the owning Shade checkout to a timed-only redraw at the requested time, with the regression extended to pin the redraw's time argument and the untimed no-redraw. Validation battery on the pair: Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 8 suites, 200 tests, 0 failures) and lint (exit 0), Shade MCP typecheck (0 errors) and unit tests (25 files / 161 tests pass, including the extended timed-render regression), structure check via the vendored harness (npm run test:shaders:structure, 1258 documented parameters across 200 effects), WebGL2 render check (npm run test:shaders:render:webgl2, all pass), WebGPU render check (node shaders/tests/test-harness.js --effects synth/noise --backend webgpu --verbose, 1/1 pass, temporal_diff=0.216246, identity match, generation 2→4, backend_switch match; run with SHADE_SWIFTSHADER=1 and VK_ICD_FILENAMES/VK_DRIVER_FILES at <browser cache>/chromium_headless_shell-1243/chrome-headless-shell-linux64/vk_swiftshader_icd.json — Playwright browsers reinstalled to PLAYWRIGHT_BROWSERS_PATH=<browser cache> (chromium-1243 for noisemaker, chromium-1234 for shade-mcp) and npm cache at <npm cache>), and the live browser smoke test against the corrected Shade build (NOISEMAKER=<noisemaker checkout> node scripts/browser-smoke.mjs from the Shade checkout, all 3 checks OK). Re-captured MCP initialization (server shade-mcp version 0.2.3, protocol 2024-11-05), 18 tools via tools/list, and the worked timed renderEffectFrame call from one stdio session at the immutable pin cbcab33363851016f65391fbb9ef71d66729c07f (time: 0, warmup_frames: 0 → status ok, WebGL2, 1024×1024; the pinned verb performs no pre-readback redraw, so this reads the loop's last live frame and is nondeterministic with the capture moment — two archived runs of the identical request returned mean_rgb 0.237010/0.408313/0.446757 with 917 sampled colors and 0.245578/0.416137/0.453570 with 931; an earlier run returned 0.241492/0.412125/0.449711 with 928, and one of the discarded pre-review captures returned values byte-identical to the corrected build's controlled time-0 redraw frame — an uncontrolled live-frame read can coincide with a time-0 render when the pause lands right after the pipeline restart, which is exactly why pixel values alone cannot attribute provenance and the archived per-session logs and drivers are the provenance record). The timed request time: 0.5, warmup_frames: 3 at that pin produced no response within 240000 ms (Error: NO RESPONSE within 240000 ms waiting for tools/call) — the GAP-014 hang is live at the pin. Through the corrected Shade build the same request completes, and a three-way probe returned three distinct frames: time 0 mean_rgb 0.445390/0.433124/0.469914 and time 0.5 0.457433/0.443662/0.445664 (both reproduce byte-identically across runs, deterministic per requested time), and untimed 0.446032/0.432419/0.475967 (the live loop frame; nondeterministic with the capture moment — an earlier probe run observed 0.286384/0.461296/0.503775), confirming the redraw honors the requested time. The pre-review capture log superseded here was discarded: leaked environment made it run the corrected build at time 0.5, so its values were mislabeled as a pin capture; every capture in this entry was re-run with the explicit-provenance drivers archived beside the logs. Archived run evidence (automation archive ledger-662): pin-capture-time0-run1.log, pin-capture-time0-run2.log, pin-hang-time0.5.log, corrected-build-probe.log, audio-channel-red.log, audio-channel-green.log, nm-tests.log, nm-lint.log, shade-typecheck.log, shade-tests.log, structure.log, webgl2-render.log, webgpu-render.log, browser-smoke.log, vendor-diff.txt, and the pin-capture.mjs/corrected-build-probe.mjs drivers. Publication and delivery-state record: the Shade redraw correction is shade-mcp 153eebd3b08c849a3fbf20d43527ae48a9dff746, published to main together with noisemaker c4606d1194c5fb0c6a3b27bdd542d54780da3919; exact-commit CI at those SHAs all succeeded (shade-mcp quality.yml run 36538519097, downstream.yml run 36538519252, ci.yml run 36538519002; noisemaker site.yml run 36538515092). Required CI at this publication's own commit (4d47b3fd826288077a58a95da3597dba59007fd9) likewise all succeeded — js.yml (JS tests, JS lint), shaders.yml (Shader tests, GPU tests, Bundle shaders, release dispatches), site.yml (Verify docs asset paths, notify, Dispatch static-site-release), docs-site.yml, downstream.yml, and release.yml (Publish release at tag v1.0.203, all bundle/standalone builds) — and the deployed artifact readback (archived as deployment-readback-4d47b3f.json) shows https://noisemaker.app/deployment-meta.json serving git_hash: 4d47b3fd826288077a58a95da3597dba59007fd9 (date 1790672256). Earlier readbacks in this entry's history: the deployment served git_hash: c4606d1194c5fb0c6a3b27bdd542d54780da3919 between that commit's publication and this one. Required CI at the noisemaker implementation commit 682739066d3b74962febbdcdae85b5aa4d2e19f3 (observed via the public check-runs API in this follow-up): 16 check runs, all success — Shaders (Shader tests, GPU tests, Bundle shaders, Dispatch static-site-release ×3, Dispatch library-release), Release (Publish release, Build JS bundles, Build shader bundle, Build standalone Linux/Windows/macOS), Site (Verify docs asset paths, notify), plus the auto-merge skip; the js.yml workflow did not run there by design, its path filter (js/, scripts/, package files) excludes a commit touching only shaders/** and llms-full.txt. Delivery is the open blocker: the newest Shade release remains v0.2.3 = cbcab33363851016f65391fbb9ef71d66729c07f (git ls-remote --tags), so the .mcp.json pin, the vendored vendor/shade-mcp/ bundle (re-verified this pass against the freshly downloaded v0.2.3 tarball shade-mcp-dist.tar.gz, sha256 4256ea5c695f15b3eec292f599a1ae1aada03a2628e33db8b3c9864a6b11f93f: the four vendored module trees ai/analysis/formats/harness are byte-identical; last vendor change fa83eea, 2026-09-25), and GAP-014's upstream-fixed status all wait on an operator-authorized Shade release carrying ed73367/ca94cfa/07af4ffa plus the redraw correction, followed by the existing pull-shade-mcp vendor refresh and pin bump. llms-full.txt GAP-014 row and its Act prose were updated to this verified source state with the pending delivery stated; both checkpoints are deliberately unchanged (delivery pending), so the snapshot block still pins noisemaker 73c15be / shade-mcp 00340b1. Recorded observation left for the gap owners, not changed here: the repository harness wrapper's own timed capture (GAP-014's path, pre-existing) redraws renderer.render(0) twice before readback after setPausedTime(time), so a nonzero requested time draws the time-0 frame there too; no caller currently requests a nonzero time through it. Pass completion follow-up (work queue item 673): the entry's described ranges were re-diffed locally and extend exactly through noisemaker 4f5e0d28bdc155700393c314e9a5aafcc4da91fd (c4606d1→4d47b3f→4f5e0d2, linear) and shade-mcp 153eebd3b08c849a3fbf20d43527ae48a9dff746; both checkouts are in sync with their published mains. The validation battery was completed on the exact pair on a real-GPU macOS host (Automated job a88a5288, evidence ledger-673, host-run-note.md): the full WebGL2 render chain exits 0 (synth/noise 1/1 with page-confirmed identity, temporal_diff=0.216193, plus the mesh first-frame, GLSL y-orientation directMean=0.000/yFlipRatio=1.000, overlay-recompile, and external-texture-upload sub-tests), the WebGPU chain exits 0 (synth/noise backend=webgpu 1/1, same temporal_diff=0.216193; the chain's external-texture sub-test self-reports its pre-existing "SKIP WebGPU: backend resolved to WebGL2"), and Shade's real viewer smoke test passes all 3 checks (compileEffect, renderEffectFrame, module import from a setContent page) against the corrected 153eebd build with NOISEMAKER at the noisemaker checkout. Re-captured MCP initialization (server shade-mcp version 0.2.3, protocol 2024-11-05), the 18-tool tools/list, and the worked UNTIMED renderEffectFrame call from one stdio session at the immutable pin cbcab33363851016f65391fbb9ef71d66729c07f (status ok, WebGL2, 1024×1024, mean_rgb 0.450318/0.438138/0.453507, 1001 unique sampled colors — the pin's documented live-frame read, nondeterministic with the capture moment; the pin's timed-path GAP-014 hang stands as recorded above and was not re-run). The affected TIMED path was then re-captured from two further one-session stdio captures on the same GPU host (same evidence): through the corrected 153eebd build the timed request renderEffectFrame {time: 0.5, warmup_frames: 3} completes — status ok, WebGL2, 1024×1024, mean_rgb 0.457276/0.443470/0.445504, 1001 unique sampled colors (deterministic per requested time, matching the redraw-honors-requested-time behavior recorded above) — while the identical timed request at the immutable pin cbcab33363851016f65391fbb9ef71d66729c07f again produced no response within 240000 ms (tools/call NO RESPONSE within 240000 ms), re-running the recorded GAP-014 hang live at the pin this pass. Environment adaptations used on the host (all temporary run scaffolding, uncommitted and reverted; detailed in the evidence note): the sandbox binds only loopback ports 43117–43126, so SHADE_VIEWER_PORT was pinned and four scripts' hardcoded acquireServer(0/undefined, …) calls were port-pinned for the run; sibling checkouts are EPERM-isolated per run, so the smoke used a git archive copy of noisemaker 4f5e0d2; and the host proxy blocks handfish.noisefactor.io, so that copy's demo page imported a locally vendored byte-copy of the CDN bundle (sha256 003a8ab98a2367b852bce423a1edc4e93829676a87b47b3f1c4a268dbe717fea) — no repository file was changed for any of this. Exact-commit CI observed via the public check-runs API, with the full API responses archived beside the run logs for independent re-verification (evidence ledger-673): check-runs-4f5e0d2.json — 22 check runs at head_sha 4f5e0d28bdc155700393c314e9a5aafcc4da91fd, every one completed with conclusion success or skipped (Shaders, JS tests/lint, Release builds, Site, notify); check-runs-153eebd.json — 6 check runs at 153eebd3b08c849a3fbf20d43527ae48a9dff746 (quality/ci/browser/ notify success, two auto-merge skips). The re-diffed source ranges and their exact commit lists are archived in range-audit.txt (the noisemaker watched-roots drift is exactly the GAP-032 commits a505910/6827390/4d47b3f/4f5e0d2 and their external-input.js + test_external_input.js files; shade-mcp 00340b1..153eebd is the four warmup/redraw commits). The live deployment readback is archived in deployment-readback-4f5e0d2.json: https://noisemaker.app/deployment-meta.json serving git_hash: 4f5e0d28bdc155700393c314e9a5aafcc4da91fd (date 1790673091). Final-record verification at this documentation commit itself: the Noisemaker battery was re-run green at 72b07e3 (receipts nm-tests-final.log 8 suites / 200 tests / 0 failures, nm-lint-final.log exit 0, structure-final.log 1258 parameters across 200 effects); 72b07e3 is a LEDGER.md-only commit and correctly has zero check runs — no workflow's path filter watches LEDGER.md (the same path-filtered behavior recorded for prior documentation-only commits, which ran only the Site workflow because they touched llms-full.txt), and the published source commits it documents carry the CI evidence above. Deployment remains at the last source commit, git_hash: 4f5e0d28bdc155700393c314e9a5aafcc4da91fd (a records-only commit ships no site artifact; the deployment readback for the published source commit stands as recorded). Deployment: https://noisemaker.app/deployment-meta.json serves git_hash: 4f5e0d28bdc155700393c314e9a5aafcc4da91fd. Delivery re-check: the newest Shade release is still v0.2.3 = cbcab33363851016f65391fbb9ef71d66729c07f, so the .mcp.json pin, the vendored bundle, and GAP-014's delivered state all remain on the operator-authorized-release blocker recorded above; both checkpoints stay deliberately unchanged at noisemaker 73c15be / shade-mcp 00340b1 (delivery pending), and the snapshot block is unchanged with them. The container/host receipts for every battery leg are archived with the job (evidence ledger-673: nm-tests-green.log 8 suites / 200 tests / 0 failures re-run at this documentation commit, nm-lint-green.log, structure-green.log, shade-typecheck-host.json, shade-tests.log 25 files / 161 tests, the host-*.json run outputs, and the probe receipts for the host adaptations); an earlier partial nm-tests.log capture taken before that container's npm ci finished is superseded by the green rerun and kept only as the record of the failed attempt. 73c15be00d6888f4b5d2835d8e242ee9e840df45 / shade-mcp 00340b148e109464b1a87d89ff29fc7622d384c9 (Automated job 67ecc03a, selected existing GAP-029): the watched-root drift range 73c15be..HEAD is empty (the range contains only this contract's own documentation advances and eslint/ruff dependency bumps outside the watched roots), so no behavior statement, typed grammar, validator message, catalog reference, viewer bridge, harness API, or delivery path changed. The pass narrowed the GAP-029 register row in llms-full.txt: the defect and its fix live in the Shade repository's src/knowledge/shared-instances.ts (invalidateSharedEffectIndex() leaves building intact and an in-flight pre-change build repopulates the cache after invalidation, so a list/search immediately after a library change or generateManifest can retain a pre-change index for the next five-second TTL), which noisemaker cannot write; nothing of the knowledge module is vendored here (vendor/shade-mcp/ ships only ai/analysis/formats/harness, and invalidateSharedEffectIndex appears nowhere in js/, scripts/, shaders/, test/, demo/), so there is no repository-side surface and the row stays open, narrowed/blocked, with the existing retry-after-TTL workaround. Shade-side audit by unauthenticated remote fetch: git ls-remote puts both noisefactorllc/shade-mcp and noisedeck/shade-mcp tips at 00340b1; the content hash of src/knowledge/shared-instances.ts — git blob c3081989adfbf4e568e31043c248d35d0ab74fbe, unchanged since d6b8ab017bbf682055a1872e3fd86c273c28d5fc (2026-08-12) — is byte-identical at that tip, at the .mcp.json pin cbcab33, and at the newest release tag v0.2.3 (which points at that pin). Recorded limitation: no Shade-side execution of the race itself was run in this pass — fixing it is a Shade-repository commit, and Shade's own 157 passing unit tests do not cover invalidation during an in-flight build — so the upstream persistence claim rests on the content-addressed blob identity above, re-derivable from any unauthenticated clone. Verified four identities: Noisemaker source 73c15be00d6888f4b5d2835d8e242ee9e840df45, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ unchanged in this pass's range (last vendor change 63349a7, 2026-09-25; release v0.2.3 byte-verified against the shade-mcp-dist.tar.gz in the 2026-09-28 pass). Validation battery on the exact pair: noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 0 failures) and lint (exit 0), Shade MCP typecheck (0 errors) and unit tests (24 files / 157 tests pass on the 00340b1 checkout), structure check via the vendored harness (npm run test:shaders:structure, 1258 documented parameters across 200 effects), WebGL2 render check (npm run test:shaders:render:webgl2, all pass) and WebGPU render check (node shaders/tests/test-harness.js --effects synth/noise --backend webgpu --verbose, 1/1 pass, temporal_diff=0.216246, page-confirmed identity_check match: backend observed webgpu, generation 2→4, backend_switch match; SHADE_SWIFTSHADER=1 with VK_ICD_FILENAMES/VK_DRIVER_FILES at the Playwright chromium_headless_shell-1243 bundle's vk_swiftshader_icd.json and PLAYWRIGHT_BROWSERS_PATH=<browser cache> holding both pinned Playwright browser builds — chromium-1243 for noisemaker, chromium-1234 for shade-mcp), and the live browser smoke test (NOISEMAKER=<noisemaker checkout> node scripts/browser-smoke.mjs from the Shade checkout, all 3 checks OK: compileEffect, renderEffectFrame, module import from a setContent page). Re-captured MCP initialization (server shade-mcp version 0.2.3, protocol 2024-11-05), 18 tools via tools/list, and the GAP-029 verbs' worked calls from one stdio session at the tested immutable pin cbcab33363851016f65391fbb9ef71d66729c07f: listEffects returns 210 effects and searchEffects query:"noise", limit:3 returns scored hits (classicNoisedeck/cellNoise 96, classicNoisedeck/noise 96, classicNoisedeck/noise3d 96). Gap count stands at 12 (GAP-029 narrowed/ blocked — a Shade-repository item; no newer Shade release exists). Updated llms-full.txt (GAP-029 row narrowed with the upstream evidence and the recorded limitation) and advanced this checkpoint together.
    • 2026-09-28 — caught up through noisemaker 73c15be00d6888f4b5d2835d8e242ee9e840df45 / shade-mcp 00340b148e109464b1a87d89ff29fc7622d384c9 (work queue item 658, job ledger-llms-contract): audited watched source roots across noisemaker range 296e0138c4744ed485b2e95de3eeb466c17629ee..73c15be00d6888f4b5d2835d8e242ee9e840df45. The trigger's forced start a912749 is the prior entry's audited range start (its delivery ranges 04e8582..c28e8fd, c28e8fdb..7aff843, 7aff843..73c15be are linear inside this checkpoint-to-HEAD audit, which also contains the prior entry's own checkpoint-advance commit 04e8582), so every delivered commit is covered with no gap. The range is four commits: 04e8582 (the prior entry's documentation-only checkpoint advance), c28e8fd (engine fix — webgl2 ensureDepthBuffer() leaves the FBO unbound after initial depth allocation; the mesh render path now re-binds gl.bindFramebuffer(gl.FRAMEBUFFER, fbo), with a focused mesh first-frame regression in shaders/tests/test_mesh_first_frame.mjs), 7aff843 (GAP-024 — every browser result the repository harness produces carries the additive identity_check record built by shaders/tests/session-identity.js, the wrapper's readiness poll is identity-bound, backend switches are confirmed against the page, and demo/shaders/index.html publishes __noisemakerPipelineGeneration; mismatches gate only behind --strict-identity), and 73c15be (GAP-026 — the production pipeline invokes onInit/onUpdate/onDestroy via Pipeline.initLifecycleEffects()/_invokeUpdateHooks()/ _withRuntimeUniforms() with fallback uniform semantics, regressions in shaders/tests/test_lifecycle_hooks.js; no effects, catalog, .mcp.json, or vendor changes). Shade-side audit: upstream noisefactorllc/shade-mcp and the local checkout are unchanged at 00340b1 (git ls-remote origin main == 00340b1; newest release still v0.2.3 / cbcab33363851016f65391fbb9ef71d66729c07f). Shade's parsers, analysis, knowledge, and browser tools reference none of the new surfaces — grep of Shade src//scripts/ and the vendored vendor/shade-mcp/ tree for onInit/onUpdate/onDestroy/ensureDepthBuffer/ initLifecycleEffects/session-identity/identity_check/ __noisemakerPipelineGeneration finds zero vendor hits (Shade's own src/__tests__/session-lifecycle.test.ts is its browser-session lifecycle, unrelated) — so no integration-code change or new Shade regression was required, and the public tool result shapes are unchanged. Contract re-audit found and fixed three stale statements the range's own GAP-026 row update left behind (llms-full.txt lines 722/826-828 still claimed no production caller exists, and this section's own Scope register still listed GAP-026 among 13 open gaps): all now describe the verified state, and the worked-instance audit was extended across 296e013..73c15be (the range does touch engine runtime, but synth/testPattern is hook-less and byte-unchanged, and the webgl2 depth fix only affects mesh/depth paths it never enters). Verified four identities: Noisemaker source 73c15be00d6888f4b5d2835d8e242ee9e840df45, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ byte-identical to the release v0.2.3 tarball (shade-mcp-dist.tar.gz, re-downloaded in this pass, sha256 4256ea5c695f15b3eec292f599a1ae1aada03a2628e33db8b3c9864a6b11f93f, diffed for harness/, ai/, formats/, analysis/; the vendor tree last changed at 13a8a04, an ancestor of this checkpoint). Validation battery on the exact pair: Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 8 suites, 200 tests, 0 failures, including the GAP-024 test_session_identity.js and GAP-026 test_lifecycle_hooks.js suites) and lint (npm run lint, exit 0), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via the vendored harness (npm run test:shaders:structure, exit 0, 1258 documented parameters across 200 effects), WebGL2 render check (npm run test:shaders:render:webgl2, 1/1 pass, temporal_diff=0.216246, is_no_animation=false, is_low_variety=false), WebGPU render check (node shaders/tests/test-harness.js --effects synth/noise --backend webgpu --verbose, 1/1 pass, temporal_diff=0.216246, page-confirmed identity_check match: backend observed webgpu, generation 2→4, backend_switch match — run with SHADE_SWIFTSHADER=1 and VK_ICD_FILENAMES/VK_DRIVER_FILES at the Playwright Chromium bundle's vk_swiftshader_icd.json; Playwright browsers reinstalled to PLAYWRIGHT_BROWSERS_PATH=<browser cache> — chromium-1243 for noisemaker, chromium-1234 for shade-mcp — and npm used npm_config_cache=<npm cache>), and the live browser smoke test (NOISEMAKER=<noisemaker checkout> node scripts/browser-smoke.mjs from the Shade checkout, all 3 checks OK: compileEffect, renderEffectFrame, module import from a setContent page). Re-captured MCP initialization (server shade-mcp version 0.2.3, protocol 2024-11-05), 18 tools via tools/list, and the worked analyzeEffect and checkEffectStructure calls with effect_id: "synth/noise" from one stdio session at the tested immutable pin cbcab33363851016f65391fbb9ef71d66729c07f (analyzeEffect returns the lossy regex projection without error; checkEffectStructure returns status: "ok", passCount 1, no issues). CI evidence for the published source commits (observed via the public GitHub check-runs API in this pass): each of c28e8fd, 7aff843, and 73c15be ran Shaders (Shader tests, GPU tests, Bundle shaders, Dispatch static-site-release and library-release), JS tests (JS tests, JS lint), Verify docs asset paths, notify, and Release (Publish release, Build JS bundles, Build shader bundle, Build standalone Linux/Windows/macOS) — all success; the documentation-only 04e8582 correctly ran only the path-filtered Site workflow (Dispatch static-site-release, success). Deployment: https://noisemaker.app/deployment-meta.json serves git_hash: 73c15be00d6888f4b5d2835d8e242ee9e840df45; the published contract at that deployment still pins 296e0138 in its snapshot block because this checkpoint-advance commit is documentation-only and not yet published. Open gap count stands at 12 (GAP-026 closed at 73c15be; GAP-024 narrowed repository-side, the upstream Shade verbs remain Shade-repository items; GAP-013 stays closed on immutable pinning — no newer Shade release exists). Updated llms-full.txt (snapshot block advanced to Noisemaker 73c15be00d6888f4b5d2835d8e242ee9e840df45, the stale GAP-026 narrative statements corrected against verified source, the worked-instance audit extended across 296e013..73c15be, and the 2026-09-28 re-capture evidence) and advanced this checkpoint together.
    • 2026-09-27 — caught up through noisemaker 296e0138c4744ed485b2e95de3eeb466c17629ee / shade-mcp 00340b1 (work queue item 644, job ledger-llms-contract): audited watched source roots across noisemaker range 93229933b102ba82e713402be19db57207698850..296e0138c4744ed485b2e95de3eeb466c17629ee. The trigger's forced start 12b4d74 is the prior entry's audited range start and both observed delivery ranges (7c5f176..a912749, 11d7c69..296e013) are linear inside this checkpoint-to-HEAD audit, so every delivered commit is covered with no gap. The range is four commits, all test-harness or documentation layer — no engine, effects, viewer, catalog, .mcp.json, or vendor changes: 7c5f176 (the prior entry's own checkpoint-advance documentation commit), a912749 (GAP-019 — shaders/tests/passthrough-input.js probes TRUE input passthrough in-browser with a live write-blit positive control, consumed by shaders/tests/test-harness.js only behind the explicit --passthrough-input opt-in; regressions in shaders/tests/test_passthrough_input.js), 11d7c69 (GAP-020 — one llms-full.txt line narrowing the row to the Shade-repository error envelope; nothing of the tool-result wrapper is vendored and no noisemaker module consumes the MCP envelope), and 296e013 (GAP-021 — the harness renderEffectFrame wrapper annotates every resolution-bearing result with additive requested_resolution/resolution_check fields, pinned by shaders/tests/frame-resolution.js with regressions in shaders/tests/test_frame_resolution.js; mismatch gates only behind the explicit --strict-resolution opt-in; no status, frame, or metric field changes). Shade-side audit: upstream noisefactorllc/shade-mcp and the local checkout are unchanged at 00340b1 (git ls-remote origin main == 00340b1; newest release still v0.2.3 / cbcab33363851016f65391fbb9ef71d66729c07f). Shade's parsers, analysis, knowledge, and browser tools reference none of the new modules — grep of Shade src//scripts/ and the vendored vendor/shade-mcp/ tree for passthrough-input/frame-resolution finds zero hits; the new surfaces are repository-internal test/harness code, and the GAP-019/021 upstream verbs are explicitly left as Shade-repository items per the rows — so no integration-code change or new Shade regression was required, and the public tool result shapes are unchanged (verified live: analyzeEffect still returns the lossy regex projection, checkEffectStructure still status: "ok"). Verified four identities: Noisemaker source 296e0138c4744ed485b2e95de3eeb466c17629ee, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ byte-identical to the release v0.2.3 tarball (shade-mcp-dist.tar.gz, re-downloaded in this pass, sha256 4256ea5c695f15b3eec292f599a1ae1aada03a2628e33db8b3c9864a6b11f93f, diffed for harness/, ai/, formats/, analysis/; the vendor tree last changed at 9d3474d, an ancestor of this checkpoint). Validation battery on the exact pair: Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 8 suites, 200 tests, 0 failures, including the GAP-019 test_passthrough_input.js and GAP-021 test_frame_resolution.js suites) and lint (npm run lint, exit 0), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via the vendored harness (npm run test:shaders:structure, exit 0, 1258 documented parameters across 200 effects), WebGL2 render check (npm run test:shaders:render:webgl2, exit 0, temporal_diff=0.216246, is_no_animation=false, is_low_variety=false), WebGPU render check (npm run test:shaders:render:webgpu, 1/1 pass, temporal_diff=0.216246, is_no_animation=false, is_low_variety=false — run with SHADE_SWIFTSHADER=1 and VK_ICD_FILENAMES/VK_DRIVER_FILES at the Playwright Chromium bundle's vk_swiftshader_icd.json; the browser cache had been pruned, so Playwright browsers were reinstalled to PLAYWRIGHT_BROWSERS_PATH=<browser cache> — chromium-1243 for noisemaker, chromium-1234 for shade-mcp), and the live browser smoke test (NOISEMAKER=<noisemaker checkout> node scripts/browser-smoke.mjs from the Shade checkout, all 3 checks OK: compileEffect, renderEffectFrame, module import from a setContent page). Re-captured MCP initialization (server shade-mcp version 0.2.3, protocol 2024-11-05), 18 tools via tools/list, and the worked analyzeEffect and checkEffectStructure calls with effect_id: "synth/noise" from one stdio session at the tested immutable pin cbcab33363851016f65391fbb9ef71d66729c07f (analyzeEffect returns the lossy regex projection without error; checkEffectStructure returns status: "ok", passCount 1, no issues). CI evidence for the published commits (observed via the public GitHub check-runs API in this pass): each of a912749 and 296e013 ran Shaders (Shader tests, GPU tests, Bundle shaders, Dispatch static-site-release and library-release), JS tests (JS tests, JS lint), Verify docs asset paths, notify, and Release (Publish release, Build JS bundles, Build shader bundle, Build standalone Linux/Windows/macOS) — all success; the documentation-only 7c5f176 and 11d7c69 correctly ran only the path-filtered Site workflow (Dispatch static-site-release, success). Updated llms-full.txt (snapshot block advanced to Noisemaker 296e0138c4744ed485b2e95de3eeb466c17629ee, worked-instance audit extended across 9322993..296e013, and the GAP-019/GAP-021 re-capture evidence) and advanced this checkpoint together. Open gap count stands at 13 (GAP-019 and GAP-021 narrowed repository-side; the upstream Shade verbs remain Shade-repository items, and GAP-013 stays closed on immutable pinning — no newer Shade release exists).
    • 2026-09-27 — caught up through noisemaker 93229933b102ba82e713402be19db57207698850 / shade-mcp 00340b148e109464b1a87d89ff29fc7622d384c9 (work queue item 631, job ledger-llms-contract): audited watched source roots across noisemaker range 12b4d74fb4f28d5f00bb1dde107fa8673814d8b9..93229933b102ba82e713402be19db57207698850 — linear, two commits: 8fe3cca (documentation-only: post-publication deployed-contract evidence and the open-gap enumeration corrected to 13) and 9322993 (GAP-017 — shaders/tests/definition-schema.js imports each effect's definition live and pins with auditSchemaLoss()/auditDefinitionLoss() what the real vendored regex parser drops; the harness gains the informational, no-browser --describe <patterns> mode in shaders/tests/test-harness.js printing each matched effect's complete schema, the upstream projection, and the loss audit; regressions in shaders/tests/test_definition_schema.js registered in scripts/run-js-tests.js; no engine, effects, viewer, catalog, .mcp.json, or vendor changes). Shade-side audit: upstream noisefactorllc/shade-mcp and the local checkout are unchanged at 00340b1 (git ls-remote origin main == 00340b1; newest release still v0.2.3 / cbcab33363851016f65391fbb9ef71d66729c07f). Shade's parsers, analysis, knowledge, and browser tools reference none of the new modules — grep of Shade src/ and the vendored vendor/shade-mcp/ tree for definition-schema/runDescribeMode/auditSchemaLoss finds zero hits; the new surfaces are repository-internal test/harness code — so no integration-code change or new Shade regression was required, and the public tool result shapes are unchanged (verified live: analyzeEffect still returns the lossy regex projection, checkEffectStructure still status: "ok"). Verified four identities: Noisemaker source 93229933b102ba82e713402be19db57207698850, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ byte-identical to the release v0.2.3 tarball (shade-mcp-dist.tar.gz, re-downloaded in this pass, sha256 4256ea5c695f15b3eec292f599a1ae1aada03a2628e33db8b3c9864a6b11f93f, diffed for harness/, ai/, formats/, analysis/; the vendor tree last changed at 286ccf6, an ancestor of the prior checkpoint). Validation battery on the exact pair: Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 200 tests, 0 failures, including the GAP-017 test_definition_schema.js suite) and lint (npm run lint, exit 0), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via the vendored harness (npm run test:shaders:structure, exit 0, 1258 documented parameters across 200 effects), WebGL2 render check (npm run test:shaders:render:webgl2, exit 0), WebGPU render check (npm run test:shaders:render:webgpu, 1/1 pass, temporal_diff=0.216246, is_no_animation=false, is_low_variety=false — run with SHADE_SWIFTSHADER=1 and VK_ICD_FILENAMES/VK_DRIVER_FILES at the Playwright Chromium bundle's vk_swiftshader_icd.json; the browser cache had been pruned, so Playwright browsers were reinstalled to PLAYWRIGHT_BROWSERS_PATH=<browser cache> — chromium-1243 for noisemaker, chromium-1234 for shade-mcp — and the MCP re-capture used npm_config_cache=<npm cache>), and the live browser smoke test (NOISEMAKER=<noisemaker checkout> node scripts/browser-smoke.mjs from the Shade checkout, all 3 checks OK: compileEffect, renderEffectFrame, module import from a setContent page). Re-captured MCP initialization (server shade-mcp version 0.2.3, protocol 2024-11-05) and 18 tools via tools/list in one stdio session, and the worked analyzeEffect and checkEffectStructure calls with effect_id: "synth/noise" in a second stdio session, both at the tested immutable pin cbcab33363851016f65391fbb9ef71d66729c07f (analyzeEffect returns the lossy regex projection without error; checkEffectStructure returns status: "ok", passCount 1, no issues). CI evidence for the published source commit 9322993 (observed via the GitHub check-runs API in this pass): Shaders (Shader tests, GPU tests, Bundle shaders, Dispatch static-site-release and library-release), JS tests (JS tests, JS lint), Verify docs asset paths, notify, and Release (Publish release, Build JS bundles, Build shader bundle, Build standalone Linux/Windows/macOS) — all success; the documentation-only commit 8fe3cca correctly triggered only the Site workflow (push, success) per the repository's path-filtered workflow triggers. Updated llms-full.txt (snapshot block advanced to Noisemaker 93229933b102ba82e713402be19db57207698850, worked-instance audit extended across 12b4d74..9322993, and the GAP-017 re-capture evidence recording the analyzeEffect and checkEffectStructure worked calls) and advanced this checkpoint together. Open gap count stands at 13 (GAP-017 narrowed repository-side at 9322993; the upstream Shade definition parser remains a Shade-repository item).
    • 2026-09-27 — caught up through noisemaker 12b4d74fb4f28d5f00bb1dde107fa8673814d8b9 / shade-mcp 00340b1 (work queue item 619, job ledger-llms-contract): audited watched source roots across noisemaker range 7443f6e6180300a45c5b97608459e5094504659d..12b4d74fb4f28d5f00bb1dde107fa8673814d8b9. The trigger's three observed delivery ranges (132d1bf..c2252f0, c2252f0..e73a44a, e73a44a..12b4d74) are linear; 132d1bf is the prior entry's documentation-only checkpoint advance, so the checkpoint-to-HEAD audit plus the prior entry cover every delivered commit with no gap. The range is three commits: c2252f0 (GAP-015 — shaders/tests/image-metrics.js mirrors the vendored metric implementations and adds auditMetricImplementationEquivalence(), with regressions in shaders/tests/test_image_metrics.js), e73a44a (GAP-014 — the harness renderEffectFrame wrapper (shaders/tests/test-harness.js) warms up its frames before pausing on explicit-time requests, pinned by the mirror shaders/tests/frame-warmup.js with regressions in shaders/tests/test_frame_warmup.js), and 12b4d74 (GAP-016 engine change — preflightEffect(definition, capabilities, shaders) in shaders/src/runtime/preflight.js reports per-backend authorability, predicted MRT format demotions, and predicted maxTextureSize clamps before compilation; Pipeline.preflight(capabilities?) exposes it over the pipeline's own graph and Pipeline.mrtFormatBytes() now delegates to the shared mrtFormatBytes() so prediction and runtime stay in lockstep; regressions in shaders/tests/test_preflight.js registered in npm run test:shaders:lang). Shade-side audit: upstream noisefactorllc/shade-mcp and the local checkout are unchanged at 00340b1 (git ls-remote origin main == 00340b1; no release newer than v0.2.3 / cbcab33363851016f65391fbb9ef71d66729c07f). Shade's parsers, analysis, knowledge, and browser tools consume none of the new surfaces — grep of Shade src/ and the vendored vendor/shade-mcp/ tree finds no reference to preflightEffect, mrtFormatBytes, or the new test modules; the browser tools compute their own metrics from viewer canvas globals and never import the repository's harness wrappers — so no integration-code change or new Shade regression was required, and the public tool result shapes are unchanged. The mrtFormatBytes() delegation is behavior-preserving (the shared implementation is byte-for-byte the former switch). Verified four identities: Noisemaker source 12b4d74fb4f28d5f00bb1dde107fa8673814d8b9, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ byte-identical to the release v0.2.3 tarball (shade-mcp-dist.tar.gz, re-downloaded in this pass, sha256 4256ea5c695f15b3eec292f599a1ae1aada03a2628e33db8b3c9864a6b11f93f, diffed for harness/, ai/, formats/, analysis/; the vendor tree is unchanged across the audited range). Validation battery on the exact pair: Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 8 suites, 200 tests, 0 failures, including the GAP-014 test_frame_warmup.js, GAP-015 test_image_metrics.js, and GAP-016 test_preflight.js suites) and lint (npm run lint, exit 0), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via the vendored harness (npm run test:shaders:structure, exit 0, 1258 documented parameters across 200 effects), WebGL2 render check (npm run test:shaders:render:webgl2, exit 0), WebGPU render check (npm run test:shaders:render:webgpu, 1/1 pass, temporal_diff=0.216246, is_no_animation=false, is_low_variety=false — run with SHADE_SWIFTSHADER=1 and VK_ICD_FILENAMES/VK_DRIVER_FILES at the Playwright Chromium bundle's vk_swiftshader_icd.json), and the live browser smoke test (NOISEMAKER=<noisemaker checkout> node scripts/browser-smoke.mjs from the Shade checkout, all 3 checks OK: compileEffect, renderEffectFrame, module import from a setContent page). Playwright browsers were installed to PLAYWRIGHT_BROWSERS_PATH=<browser cache> (noexec $HOME tmpfs; chromium-1243 for noisemaker, chromium-1234 for shade-mcp) and the MCP re-capture used npm_config_cache=<npm cache>. Re-captured MCP initialization (server shade-mcp version 0.2.3, protocol 2024-11-05), 18 tools via tools/list, and a worked checkEffectStructure call with effect_id: "synth/noise" (status ok, passCount 1, no issues) from one stdio session at the tested immutable pin cbcab33363851016f65391fbb9ef71d66729c07f. CI evidence for the published source commits (observed via the GitHub check-runs API in this pass): each of c2252f0, e73a44a, and 12b4d74 ran Shaders (Shader tests, GPU tests, Bundle shaders, Dispatch static-site-release and library-release), JS tests (JS tests, JS lint), Verify docs asset paths, notify, and Release (Publish release, Build JS bundles, Build shader bundle, Build standalone Linux/Windows/macOS) — all success. Deployment: https://noisemaker.app/deployment-meta.json serves git_hash: 12b4d74fb4f28d5f00bb1dde107fa8673814d8b9 at the time of the audit (the deployment then still served this checkpoint-advance commit's documentation predecessor, since the checkpoint commit itself was not yet published). Post-publication delivery state, verified after the checkpoint-advance commit was pushed: https://noisemaker.app/deployment-meta.json serves git_hash: ec457c2eec695427fdbf026bd5ce5db04456a151 and https://noisemaker.app/llms-full.txt serves the advanced contract whose snapshot block pins Noisemaker 12b4d74fb4f28d5f00bb1dde107fa8673814d8b9. Open gap count stands at 13 (GAP-009 closed at b35361e and GAP-016 closed repository-side; GAP-014 and GAP-015 narrowed repository-side, the upstream Shade verbs remain Shade-repository items). Updated llms-full.txt (snapshot block, worked-instance note extended across 7443f6e..12b4d74, re-capture date, open-gap enumeration corrected to the 13 register rows still open — GAP-009's closed row was wrongly still enumerated — and the post-publication deployed-contract evidence) and advanced this checkpoint together.
    • 2026-09-27 — caught up through noisemaker 7443f6e6 / shade-mcp 00340b1 (work queue item 610, job ledger-llms-contract): audited watched source roots across noisemaker range 9f85687d1bafc445dcd38e28cf5f0c6dfba562f8..7443f6e6180300a45c5b97608459e5094504659d. The trigger's forced start 403c2a4 is the prior entry's source tip and 407eb7a is its documentation-only checkpoint advance, so the checkpoint-to-HEAD audit plus the prior entry cover every delivered commit with no gap. The range is two commits, both confined to the repository's test-harness layer: 7dc0f56 (GAP-011 — shaders/tests/uniform-deltas.js exports UNIFORM_RESPONSE_THRESHOLD (0.002) and re-measures every testable uniform in the live pipeline with per-uniform {name, default_value, test_value, luma_diff, max_channel_diff, responds} deltas; the harness --uniforms run reports them and cross-checks the upstream verdicts via auditUniformResponsiveness(), informationally only, keeping the upstream outer-status semantics and the GAP-010 aggregate; regressions in shaders/tests/test_uniform_deltas.js registered in scripts/run-js-tests.js) and 7443f6e (GAP-012 — the harness renderEffectFrame WebGPU path reads the render surface through the backend's async readPixels(textureId), returning the full ImageMetrics shape plus the GAP-009 additive metrics, pinned by the auditable mirror shaders/tests/frame-readback.js (renderSurfaceCandidateIds()/isValidPixelRead()/readRenderSurfacePixels()) with regressions in shaders/tests/test_frame_readback.js). No engine, effects, viewer, catalog, .mcp.json, or vendor changes: git log over the watched roots shows exactly these two commits. Shade-side audit: upstream noisefactorllc/shade-mcp and the local checkout are unchanged at 00340b1 (git ls-remote origin main == 00340b1; no release newer than v0.2.3 / cbcab33363851016f65391fbb9ef71d66729c07f). Shade's parsers, analysis, knowledge, and browser tools consume none of the new modules — they are repository-internal (shaders/tests/), referenced only by shaders/tests/test-harness.js and scripts/run-js-tests.js, and neither the vendored bundle nor the Shade src/ tree references them — so no integration-code change or new Shade regression was required, and the public tool result shapes are unchanged. Verified four identities: Noisemaker source 7443f6e6180300a45c5b97608459e5094504659d, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ byte-identical to the release v0.2.3 tarball (shade-mcp-dist.tar.gz, re-downloaded in this pass and diffed for harness/, ai/, formats/, analysis/; the vendor tree is unchanged across the audited range). Validation battery on the exact pair: Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 200 tests, 0 failures, including the GAP-011 test_uniform_deltas.js and GAP-012 test_frame_readback.js suites) and lint (npm run lint, exit 0), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via the vendored harness (npm run test:shaders:structure, exit 0, 1258 documented parameters across 200 effects), WebGL2 render check (npm run test:shaders:render:webgl2, 1/1 pass, temporal_diff=0.216246), WebGPU render check (npm run test:shaders:render:webgpu, 1/1 pass, temporal_diff=0.216246 — run with SHADE_SWIFTSHADER=1 and VK_ICD_FILENAMES/VK_DRIVER_FILES at the Playwright Chromium bundle's vk_swiftshader_icd.json), and the live browser smoke test (NOISEMAKER=<noisemaker checkout> node scripts/browser-smoke.mjs from the Shade checkout, all 3 checks OK: compileEffect, renderEffectFrame, module import from a setContent page). Playwright browsers were installed to PLAYWRIGHT_BROWSERS_PATH=<browser cache> (noexec $HOME tmpfs; chromium-1243 for noisemaker, chromium-1234 for shade-mcp) and the MCP re-capture used npm_config_cache=<npm cache>. Re-captured MCP initialization (server shade-mcp version 0.2.3, protocol 2024-11-05), 18 tools via tools/list, and a worked checkEffectStructure call with effect_id: "synth/noise" (status ok, passCount 1, no issues) from one stdio session at the tested immutable pin cbcab33363851016f65391fbb9ef71d66729c07f. CI evidence for the published source commits: 7dc0f56 ran Shaders (36281900855: Shader tests, GPU tests, Bundle shaders, Dispatch static-site-release and library-release), JS tests (36281900856: JS tests, JS lint), Verify docs asset paths (36281900804), notify (36281900825), and Release (36282108985: Publish release, Build JS bundles, Build shader bundle, Build standalone Linux/Windows/macOS) — all success; 7443f6e ran Shaders (36286368540: Shader tests, GPU tests, Bundle shaders, Dispatch static-site-release and library-release), JS tests (36286368505: JS tests, JS lint), Verify docs asset paths (36286368503), notify (36286368515), and Release (36286557153: Publish release, Build JS bundles, Build shader bundle, Build standalone Linux/Windows/macOS) — all success. Deployment: https://noisemaker.app/deployment-meta.json serves git_hash: 7443f6e6180300a45c5b97608459e5094504659d; the published contract at that deployment still pins 9f85687d in its snapshot block because this checkpoint-advance commit is documentation-only and not yet published. Open gap count stands at 15 (GAP-011 and GAP-012 narrowed repository-side; the upstream Shade tools' returned objects remain Shade-repository items). Updated llms-full.txt (snapshot block, worked-instance note extended across 9f85687d..7443f6e) and advanced this checkpoint together.
    • 2026-09-26 — caught up through noisemaker 9f85687d / shade-mcp 00340b1 (work queue item 605, job ledger-llms-contract): audited watched source roots across noisemaker range 0ac5250052e2b2e683f086959451c3bad8369a68..9f85687d1bafc445dcd38e28cf5f0c6dfba562f8, which covers the trigger's forced start 403c2a4bf2cb56307448ea2fc1d6fa3cd74b7d6e — 403c2a4 is the prior entry's source tip and 0ac5250 is the documentation-only checkpoint-advance commit above it, so the checkpoint-to-HEAD audit plus the prior entry cover every delivered commit with no gap. The range is two commits: 9f85687 (GAP-010 aggregateUniformResponsiveness() in shaders/tests/uniform-status.js, the --strict-uniforms opt-in gate in shaders/tests/test-harness.js, and regressions in shaders/tests/test_uniform_status.js registered in scripts/run-js-tests.js) and 7730ea4 (docs-only llms-full.txt prose). Shade-side audit: upstream noisefactorllc/shade-mcp and the local checkout are unchanged at 00340b1 (upstream main == 00340b1; no release newer than v0.2.3 / cbcab33363851016f65391fbb9ef71d66729c07f). The new noisemaker module consumes none of Shade's surfaces — it reclassifies the vendored testUniformResponsiveness() result shape ({status, tested_uniforms, details} with name:pass/:fail/:error entries), verified against vendor/shade-mcp/harness/index.js, and Shade's parsers, analysis, knowledge, and browser tools consume none of the new module — so no integration-code change or new Shade regression was required, and the public tool result shapes are unchanged. Verified four identities: Noisemaker source 9f85687d1bafc445dcd38e28cf5f0c6dfba562f8, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ byte-identical to the release v0.2.3 tarball (shade-mcp-dist.tar.gz, re-extracted in this pass and diffed for harness/, ai/, formats/, analysis/; the vendor tree is unchanged across the audited range). Validation battery on the exact pair: Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, exit 0, zero failures, including the GAP-010 test_uniform_status.js suite) and lint (npm run lint), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via the vendored harness (npm run test:shaders:structure, all checks pass, 1258 documented parameters across 200 effects), WebGL2 render check (npm run test:shaders:render:webgl2, exit 0), WebGPU render check (npm run test:shaders:render:webgpu, 1/1 pass — run with SHADE_SWIFTSHADER=1 and VK_ICD_FILENAMES/VK_DRIVER_FILES at the Playwright Chromium bundle's vk_swiftshader_icd.json; synth/noise → temporal_diff=0.216246, is_no_animation=false, is_low_variety=false), and the live browser smoke test (NOISEMAKER=<noisemaker checkout> node scripts/browser-smoke.mjs from the Shade checkout, all 3 checks OK: compileEffect, renderEffectFrame, module import from a setContent page). Playwright browsers were installed to PLAYWRIGHT_BROWSERS_PATH=<browser cache> (noexec $HOME tmpfs; chromium-1243 for noisemaker, chromium-1234 for shade-mcp) and the MCP re-capture used npm_config_cache=<npm cache>. Re-captured MCP initialization (server shade-mcp version 0.2.3, protocol 2024-11-05), 18 tools via tools/list, and a worked checkEffectStructure call with effect_id: "synth/noise" (status ok, passCount 1, no issues) from one stdio session at the tested immutable pin cbcab33363851016f65391fbb9ef71d66729c07f. CI evidence for the published source commits: 9f85687 ran Release (36275489676), Shaders (36275251110), JavaScript (36275251057), Site (36275251035), Downstream (36275251021), and Docs site (36275251053) — all success; 7730ea4 ran Site (36276274476, success) and is docs-only. Deployment: https://noisemaker.app/deployment-meta.json serves git_hash: 7730ea4aba56ab3101b8eac125a21361c2ff0ce8; the published contract at that deployment still pins 403c2a4 in its snapshot block because this checkpoint-advance commit is documentation-only and not yet published. Open gap count stands at 15 (GAP-010 narrowed repository-side; the upstream Shade tool's outer status remains a Shade-repository item). Updated llms-full.txt (snapshot block, worked-instance note extended across 403c2a4..9f85687) and advanced this checkpoint together.
    • 2026-09-26 — caught up through noisemaker 403c2a4 / shade-mcp 00340b1 (work queue item 601, job ledger-llms-contract): audited watched source roots across noisemaker range a651c075..403c2a4 (covering the trigger's 8eeb7b5a..403c2a4 delivery; 8eeb7b5 is a linear ancestor of the prior checkpoint a651c075, verified with git merge-base --is-ancestor, so the checkpoint-to-HEAD audit plus the prior entry covers every delivered commit with no gap). The range is four commits: 66b8ce7 (prior checkpoint advance), b35361e (GAP-009 engine change — additive temporal_diff, is_no_animation, is_low_variety metrics in the repository's test-harness renderEffectFrame wrapper plus the --low-variety opt-in gate), 19fdcb5 (GAP-009 closure docs), and 403c2a4 (GAP-008 engine change — predictReplacement()/preflight in shaders/src/lang/transform.js with the read-only getParamAliases() in shaders/src/lang/paramAliases.js). Shade-side audit: upstream noisefactorllc/shade-mcp and the local checkout are unchanged at 00340b1 (upstream main == 00340b1; no release newer than v0.2.3 / cbcab33363851016f65391fbb9ef71d66729c07f); Shade's parsers, analysis, knowledge, and browser tools consume none of the new surfaces — src/tools/browser/render.ts computes its own metrics from viewer canvas globals and never imports the repository's harness wrapper, and no Shade tool calls getCompatibleReplacements()/replaceEffect() — so no integration-code change or new Shade regression was required, and the public tool result shapes are unchanged. Contract re-audit: updated the Compatibility-and-mutation section's typed grammar for the additive predictions/blocked fields and preflight/manifest options, the enumerated rejection strings (Replacement preflight failed: ...), the two stale GAP-008 references in the Diagnose/Validate narrative, the Compatibility-and-mutation matrix cell (now cites GAP-024 only), the GAP-008 register row (now cites the machine verification at 403c2a4), and the worked-instance note (extended to a651c075..403c2a4 with GAP-008/009 added to its range-change list); the GAP-009 register row and Rendered-output matrix cell were closed in-band by 19fdcb5 and verified against source. Verified four identities: Noisemaker source 403c2a4bf2cb56307448ea2fc1d6fa3cd74b7d6e, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ byte-identical to the release v0.2.3 tarball (shade-mcp-dist.tar.gz, re-extracted and diffed for harness/, ai/, formats/, analysis/ in this pass; the vendor tree is unchanged across the audited range). Validation battery on the exact pair, with the environment gaps closed as recorded: Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 8 suites, 200 tests, 0 failures, including the GAP-008 preflight and GAP-009 frame-metrics regressions) and lint (npm run lint), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via the vendored harness (npm run test:shaders:structure, all checks pass, 1258 documented parameters across 200 effects), WebGL2 render check (npm run test:shaders:render:webgl2, pass, including the known WebGPU-skip in test_external_texture_upload), WebGPU render check (npm run test:shaders:render:webgpu, 1/1 pass — run with SHADE_SWIFTSHADER=1 and VK_ICD_FILENAMES/VK_DRIVER_FILES at the Playwright Chromium bundle's vk_swiftshader_icd.json; synth/noise → temporal_diff=0.216246, is_no_animation=false, is_low_variety=false, matching the GAP-009 row's in-browser observation), and the live browser smoke test (NOISEMAKER=<noisemaker checkout> node scripts/browser-smoke.mjs, all 3 checks OK: compileEffect, renderEffectFrame, module import from a setContent page). Both repositories' Playwright browsers were installed to PLAYWRIGHT_BROWSERS_PATH=<browser cache> (noexec $HOME tmpfs; chromium-1243 for noisemaker, chromium-1234 for shade-mcp) and the MCP re-capture used npm_config_cache=<npm cache>. Re-captured MCP initialization (server shade-mcp version 0.2.3, protocol 2024-11-05), 18 tools via tools/list, and a worked checkEffectStructure call with effect_id: "synth/noise" (status ok, passCount 1, no issues) from one stdio session at the tested immutable pin cbcab33363851016f65391fbb9ef71d66729c07f. CI evidence for the published source commits: the range tip 403c2a4 ran Release (36263379194), Shaders (36263149990: Shader tests, GPU tests, Bundle shaders, Dispatch static-site-release and library-release), Downstream (36263149985), Site (36263150004), and Docs site (36263149987) — all success; b35361e has no exact-commit runs (pushed inside the burst whose tip 19fdcb5 ran Release 36261105261, Downstream 36260869750, Docs site 36260869768, Site 36260869766, JavaScript 36260869765, and Shaders 36260869790 — all success), and 19fdcb5 is that tip. Deployment: https://noisemaker.app/deployment-meta.json serves git_hash: 403c2a4bf2cb56307448ea2fc1d6fa3cd74b7d6e; the published contract at that deployment still pins a651c075 in its snapshot block because this checkpoint-advance commit is documentation-only and not yet published. Open gap count stands at 15 (GAP-008 resolved; a Shade MCP preflight verb remains a Shade-repository item, not a register entry). Updated llms-full.txt (snapshot block, Compatibility-and-mutation section, Diagnose rejection strings, matrix cell, GAP-008 row, worked-instance note) and advanced this checkpoint together.
    • 2026-09-26 — caught up through noisemaker a651c075 / shade-mcp 00340b1 (work queue item #570, job ledger-llms-contract): audited watched source roots across noisemaker range 5e52a2a2..a651c075, which closes work queue item #570's three observed delivery ranges (27590caa..8eeb7b5a, 428ea29b..9574362, 9574362..6a0af04) — the trigger's forced start 4891b995 is a linear ancestor of 27590caa, so trigger range 4891b995..8eeb7b5a and this checkpoint-to-HEAD audit together cover every delivered commit with no gap. Engine drift in the range: fa83eeab (GAP-005 expanded pass-field propagation and Pipeline.resolvePassViewport()), 6113da00 (GAP-006 texturePooling opt-in plus queryable Pipeline.getResourcePlan()), 95743621 (viewport-write pooling guard), and f83a427e (GAP-007 structured ShaderDiagnostic union); the remaining commits are documentation, ledger, and record updates only. Shade-side audit: upstream noisefactorllc/shade-mcp and the local checkout are unchanged at 00340b1 (upstream main == 00340b1; no release newer than v0.2.3 / cbcab33363851016f65391fbb9ef71d66729c07f), Shade's parsers, analysis, knowledge, and browser tools consume none of the new Noisemaker surfaces, and ShaderDiagnostic passes the legacy raw detail string to super(detail) so err.message and the err.detail || err.message fallbacks reach the Shade browser tool wrappers byte-identical — no integration-code change or new Shade regression was required, and the public tool result shapes are unchanged. Verified four identities: Noisemaker source a651c075bb2848b584b2bf2484f5f8a0db754b0c, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ delivered from release v0.2.3 (cbcab33363851016f65391fbb9ef71d66729c07f). Validation battery on the exact pair: Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, all suites pass) and lint (npm run lint), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via the vendored harness (npm run test:shaders:structure, all checks pass, 1258 documented parameters across 200 effects), WebGL2 render check (npm run test:shaders:render:webgl2, pass) and WebGPU render check (npm run test:shaders:render:webgpu, 1/1 pass — run with Chromium's bundled SwiftShader Vulkan ICD selected via SHADE_SWIFTSHADER=1 and VK_ICD_FILENAMES/VK_DRIVER_FILES pointing at the Playwright Chromium bundle's vk_swiftshader_icd.json; test_external_texture_upload still skips its WebGPU assertion in headless), and the live browser smoke test (NOISEMAKER=... node scripts/browser-smoke.mjs, all 3 checks OK: compileEffect, renderEffectFrame, module import from a setContent page). Re-captured MCP initialization, server info (shade-mcp 0.2.3), protocol (2024-11-05), 18 tools via tools/list, and a worked checkEffectStructure call with effect_id: "synth/noise" (status ok, passCount 1, no issues) from one session at the tested immutable pin cbcab33363851016f65391fbb9ef71d66729c07f. CI evidence: exact-commit check runs do not exist for fa83eeab, 6113da00, or f83a427e (they were pushed in one burst and CI ran at the tip); the published range tip 6a0af04d3c4f345ffab5e9f8e54e532216b4cdaa ran the full battery — Publish release, Build JS bundles, Build shader bundle, Bundle shaders, Build standalone (Linux/Windows/macOS), GPU tests, Shader tests, JS tests, JS lint, Dispatch static-site-release, Verify docs asset paths, notify — all success; the docs pass 01e9d620 passed Dispatch static-site-release, Verify docs asset paths, and notify; a651c075 is a LEDGER-only commit (auto-merge only, outside the site workflow's push paths). Deployment: https://noisemaker.app/llms-full.txt serves the contract including the GAP-005/006/007 closures. Updated llms-full.txt: snapshot block advanced to a651c075/00340b1, the GAP-006 and GAP-007 "pending publication" wording replaced with the 6a0af04d CI evidence, the worked-instance note replaced with the live pin re-capture, and the open-gap count corrected to 16; advanced this checkpoint together.
    • 2026-09-25 — caught up through noisemaker 5e52a2a2 / shade-mcp 00340b1: audited watched source roots across noisemaker range 69d83b80..5e52a2a2 (covering work queue item #551 trigger 4891b995..2f47612c through HEAD). Commits a021a283, 62eb56fa, and 2f47612c implemented GAP-004 authorable mipmaps/persistent/3D filter texture policies across WebGL2 and WebGPU backends, compiler, and pipeline. Commit 85ded3a6 closed GAP-004 with verified publication evidence for 2f47612c (review 55cfba6d-bd00-477d-9ee1-f19ddf6fade9; Release v1.0.182 run 36173647198, Shaders 36172741063, Site 36172741094, Downstream 36172741064, Docs site 36172741147, JavaScript 36172741121). Commit 63349a7d retired per-push snapshot releases. Commits c6bc8e17 and 5e52a2a2 advanced Documentation checkpoints and recorded docs deployment. Shade MCP source unchanged at 00340b1. Verified four identities: Noisemaker source 5e52a2a2fcaeb3cd540e7bdc0328a2a009116ff5, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ delivered from release v0.2.3 (cbcab33363851016f65391fbb9ef71d66729c07f). Open gap count stands at 19 entries (GAP-004 closed). Verified Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, all suites pass) and lint (npm run lint), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via vendored harness (npm run test:shaders:structure, 206/206 pass, 1258 parameters), WebGL2 and WebGPU render checks (npm run test:shaders:render:webgl2 and npm run test:shaders:render:webgpu), and live browser smoke test against Noisemaker viewer (NOISEMAKER=... node scripts/browser-smoke.mjs, all 3 checks OK: compileEffect, renderEffectFrame, module import from setContent page). Re-captured MCP initialization, server info (shade-mcp 0.2.3), protocol (2024-11-05), 18 tools via tools/list, and worked tool call checkEffectStructure with effect_id: "synth/noise". Updated llms-full.txt snapshot block and advanced LEDGER.md checkpoint together.
    • 2026-09-25 — caught up through noisemaker 69d83b80 / shade-mcp 00340b1: audited watched source roots across noisemaker range 5f4b7c43..69d83b80 (covering work queue item #535 trigger 4891b995..9d3474df through HEAD). Commits ba87ffae and 9d3474df implemented GAP-003 runtime effect definition validation contract (validateEffectDefinition in shaders/src/runtime/effect-validator.js) with an explicit-denominator 210/210 corpus gate wired into test suites. Commit 0bd09d00 recorded GAP-027 verified publication closure evidence. Commits 7a643033 and fa4b2f02 closed GAP-003 with verified publication evidence and aligned llms-full.txt validator cross-references. Commits 286ccf68 and 69d83b80 advanced Documentation checkpoints through fa4b2f02. Shade MCP source unchanged at 00340b1. Verified four identities: Noisemaker source 69d83b80e09227195fa9810e37b96dd594713ac3, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ delivered from release v0.2.3 (cbcab33363851016f65391fbb9ef71d66729c07f). Open gap count stands at 20 entries (GAP-003 and GAP-027 closed). Verified Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, all suites pass) and lint (npm run lint), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via vendored harness (npm run test:shaders:structure, 206/206 pass, 1258 parameters), WebGL2 and WebGPU render checks (npm run test:shaders:render:webgl2 and npm run test:shaders:render:webgpu), and live browser smoke test against Noisemaker viewer (NOISEMAKER=... node scripts/browser-smoke.mjs, all 3 checks OK: compileEffect, renderEffectFrame, module import from setContent page). Re-captured MCP initialization, server info (shade-mcp 0.2.3), protocol (2024-11-05), 18 tools via tools/list, and worked tool call checkEffectStructure with effect_id: "synth/noise". Updated llms-full.txt snapshot block and advanced LEDGER.md checkpoint together.
    • 2026-09-25 — caught up through noisemaker 5f4b7c43 / shade-mcp 00340b1: audited watched source roots across noisemaker range 60b90af3..5f4b7c43 (covering work queue item #516 trigger 4891b995..240740dd through HEAD). Commit 66b2c721 implemented GAP-027 subchain-argument validation contract (P008 unknown/discarded key, P009 duplicate key, P010 missing comma separator, and subchainArguments: 'strict' opt-in rejection). Commit 240740dd (v1.0.180) registered differential gate baseline against 3886ecfa (0 regressions). Commits 5a58a325 and c85db3f0 recorded published contract and gap status. Commit 5f4b7c43 advanced the Documentation checkpoint through 240740dd. Shade MCP source unchanged at 00340b1. Verified four identities: Noisemaker source 5f4b7c43a0201ab197cd1b27a66e6217b776bf57, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ delivered from release v0.2.3 (cbcab33363851016f65391fbb9ef71d66729c07f). Open gap count stands at 22 entries (GAP-027 documented and active). Verified Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, all suites pass) and lint (npm run lint), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via vendored harness (npm run test:shaders:structure, 206/206 pass, 1258 parameters), and live browser smoke test against Noisemaker viewer (NOISEMAKER=... node scripts/browser-smoke.mjs, all 3 checks OK: compileEffect, renderEffectFrame, module import from setContent page). Re-captured MCP initialization, server info (shade-mcp 0.2.3), protocol (2024-11-05), 18 tools via tools/list, and worked tool call checkEffectStructure with effect_id: "synth/noise". Updated llms-full.txt snapshot block and advanced LEDGER.md checkpoint together.
    • 2026-09-24 — caught up through noisemaker 60b90af3 / shade-mcp 00340b1: audited watched source roots across noisemaker range 13853dff..60b90af3 (covering work queue item #497 triggers 5b81e04f..4891b995 through HEAD). Commit 4891b995 exposed structured call-form and expectation diagnostics (P007/P001), closing GAP-002 parser throw sites. Commit 9fa1a221 derived parser diagnostic coordinates from source positions. Commit fca611fd derived numeric-coercion coordinates from array positions. Commits 8a21c9ca and bbdeb56c documented verified GAP-002 closure provenance and publication evidence in docs/plans/active-framework-gap.md and llms-full.txt. Commit 60b90af3 advanced the Documentation checkpoint through bbdeb56c. Shade MCP source unchanged at 00340b1. Verified four identities: Noisemaker source 60b90af373bceaf62060346d873d4f66232c7173, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ delivered from release v0.2.3 (cbcab33363851016f65391fbb9ef71d66729c07f). Open gap count stands at 22 entries (GAP-002 closed, GAP-027 active). Verified Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 122/122 pass) and lint (npm run lint), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via vendored harness (npm run test:shaders:structure, 206/206 pass, 1258 parameters), WebGL2/WebGPU render checks (npm run test:shaders:render:webgl2 and npm run test:shaders:render:webgpu), and live browser smoke test against Noisemaker viewer (NOISEMAKER=... node scripts/browser-smoke.mjs, all 3 checks OK: compileEffect, renderEffectFrame, module import from setContent page). Re-captured MCP initialization, server info (shade-mcp 0.2.3), protocol (2025-11-25), 18 tools via tools/list, and worked tool call checkEffectStructure with effect_id: "synth/noise". Updated llms-full.txt snapshot block and advanced LEDGER.md checkpoint together.
    • 2026-09-24 — caught up through noisemaker 13853dff / shade-mcp 00340b1: audited watched source roots across noisemaker range 823bbff1..13853dff (covering work queue item #475 triggers 5b81e04f..13fa8b54 through HEAD). Commit 13fa8b54 narrowed GAP-002 by exposing structured parser diagnostics (P006) for explicit subchain validation sites (non-string arguments, missing body dot, empty body), preserving permissive argument keys/separators/duplicates and shared-expectation precedence. Commits c7e9e09d (AI development contract checkpoint advance), 9928905a (Documentation checkpoint advance), 741333cb (GAP-002 subchain validation evidence docs), 30c47030 (site: white ink on magenta surfaces), and 13853dff (Documentation checkpoint advance) contained site script, documentation, and contract checkpoint updates with zero engine drift. Shade MCP source unchanged at 00340b1. Verified four identities: Noisemaker source 13853dff08f583c14edd3536774b19e41eda8364, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ delivered from release v0.2.3 (cbcab33363851016f65391fbb9ef71d66729c07f). Open gap count stands at 23 entries (GAP-002 narrowed). Verified Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 79 parser tests + 91 diagnostic location tests + all integration tests pass) and lint (npm run lint), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via vendored harness (npm run test:shaders:structure, 206/206 pass, 1258 parameters), WebGL2/WebGPU render checks (npm run test:shaders:render:webgl2 and npm run test:shaders:render:webgpu), and live browser smoke test against Noisemaker viewer (NOISEMAKER=... node scripts/browser-smoke.mjs, all 3 checks OK: compileEffect, renderEffectFrame, module import from setContent page). Re-captured MCP initialization, server info (shade-mcp 0.2.3), protocol (2025-11-25), 18 tools via tools/list, and worked tool call checkEffectStructure with effect_id: "synth/noise". Updated llms-full.txt snapshot block, DSL diagnostic description, and diagnostic code table, and advanced LEDGER.md checkpoint together.
    • 2026-09-24 — caught up through noisemaker 823bbff1 / shade-mcp 00340b1: audited watched source roots across noisemaker range 7a54ab38..823bbff1 (covering work queue item #453 triggers 5b81e04f..c9ee8a04 through HEAD). Commit c9ee8a04 advanced AI development contract checkpoint for GAP-002 (P005). Commits 285e50f5 (site header strip program pixel density/AA ink), 65ff358f (active-framework-gap docs reconciliation for GAP-002), and 823bbff1 (Documentation checkpoint advance) contained site script and documentation updates with zero engine drift across watched source roots (shaders/, demo/shaders/, .mcp.json, vendor/shade-mcp/). Shade MCP source unchanged at 00340b1. Verified four identities: Noisemaker source 823bbff1d17061d231cb0c7f5bf4527b3344abab, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ delivered from release v0.2.3 (cbcab33363851016f65391fbb9ef71d66729c07f). Open gap count stands at 23 entries. Verified Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 79 parser tests + all integration tests pass) and lint (npm run lint), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via vendored harness (npm run test:shaders:structure, 206/206 pass, 1258 parameters), WebGL2/WebGPU render checks (npm run test:shaders:render:webgl2 and npm run test:shaders:render:webgpu), and live browser smoke test against Noisemaker viewer (NOISEMAKER=... node scripts/browser-smoke.mjs, all 3 checks OK: compileEffect, renderEffectFrame, module import from setContent page). Re-captured MCP initialization, server info (shade-mcp 0.2.3), protocol (2025-11-25), and 18 tools via tools/list. Updated llms-full.txt snapshot block and advanced LEDGER.md checkpoint together.
    • 2026-09-23 — caught up through noisemaker 7a54ab38 / shade-mcp 00340b1: audited watched source roots across noisemaker range e11f0767..7a54ab38 (covering work queue item #434 triggers e32a5a4a..5b81e04f through HEAD). Commit 9b88e567 optimized classicNoisedeck/glitch by guarding glitch, scanline, and snow work when amounts are zero and attested WebGL2/WebGPU pixel parity. Commit 5b81e04f implemented output sink deferral via optional deferRender() method on sinks, tracked in renderer.deferredFrameCount, with error isolation incrementing stats.failed. Commits 7d0be45c and cc1ba268 advanced documentation and ledger checkpoints. Commit 7a54ab38 narrowed GAP-002 by exposing structured parser diagnostics (P005) for reachable output validation sites (invalid render targets, expression-context writes, invalid write surfaces, invalid write3d texture/geo arguments). Shade MCP source unchanged at 00340b1. Verified four identities: Noisemaker source 7a54ab3856d71ce037f11215572e8137ad4f53c0, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ delivered from release v0.2.3 (cbcab33363851016f65391fbb9ef71d66729c07f). Open gap count stands at 23 entries (GAP-002 narrowed). Verified Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 79 parser tests + all integration tests pass) and lint (npm run lint), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via vendored harness (npm run test:shaders:structure, 206/206 pass), WebGL2/WebGPU render checks (npm run test:shaders:render:webgl2 and npm run test:shaders:render:webgpu), and live browser smoke test against Noisemaker viewer (NOISEMAKER=... node scripts/browser-smoke.mjs, all 3 checks OK: compileEffect, renderEffectFrame, module import from setContent page). Re-captured MCP initialization, server info (shade-mcp 0.2.3), protocol (2024-11-05), and 18 tools via tools/list. Updated llms-full.txt snapshot block, Sink interface, and Act/Validate subsections for sink deferral, and advanced LEDGER.md checkpoint together.
    • 2026-09-23 — caught up through noisemaker e11f0767 / shade-mcp 00340b1: audited watched source roots across noisemaker range 9e188535..e11f0767 (covering work queue item #415 trigger 44bc4ed4..e11f0767). Commits e32a5a4a, 731b76e4, 0766743e, and 32f818e7 narrowed GAP-002 by exposing structured parser diagnostics for automation arguments (P003) and search directives (P004) with token coordinates and null spans while retaining legacy errors, updating llms-full.txt and docs/plans/active-framework-gap.md. Commits e2874c85 and 3a32b198 preserved chain- and node-scoped texture sizes and pass-level precedence during pipeline.setUniform(). Commits fde2ea40 and e11f0767 optimized classicNoisedeck/noise to skip refraction when zero and attested GLSL/WGSL pixel parity with refraction active. Shade MCP source unchanged at 00340b1. Verified four identities: Noisemaker source e11f0767993ae77eab6f3e353c942cf721aa9c90, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ delivered from release v0.2.3 (cbcab33363851016f65391fbb9ef71d66729c07f). Open gap count stands at 23 entries. Verified Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 63 tests pass) and lint (npm run lint), Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via vendored harness (206/206 pass), and live browser smoke test against Noisemaker viewer (NOISEMAKER=... node scripts/browser-smoke.mjs, all 3 checks OK). Advanced llms-full.txt snapshot block and LEDGER.md checkpoint together.
    • 2026-09-22 — caught up through noisemaker 9e188535 / shade-mcp 00340b1: audited watched source roots across noisemaker range ad60cc9b..9e188535 (covering work queue item #395 trigger e5bd2013..44bc4ed4 through HEAD). Commits 36a519a2, 31ab014f, and ae4e3302 added pruned isosurface mode to renderLandscape3d with coordinate fixes. Commits 44bc4ed4 and dd38fdd2 narrowed GAP-002 by exposing structured parser expectation diagnostics (P001/P002 in parser expect()) with token coordinates and null spans while retaining legacy errors, updating llms-full.txt and docs/plans/active-framework-gap.md. Commits 1d581ffa, eb46047b, and 9e188535 advanced ledger and docs checkpoints. Shade MCP source unchanged at 00340b1. Verified four identities: Noisemaker source 9e188535a1736780f5318d5412f0154f4209e8b1, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ delivered from release v0.2.3 (cbcab33363851016f65391fbb9ef71d66729c07f). Open gap count stands at 23 entries. Verified Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 65 tests pass) and lint, Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure check via vendored harness (206/206 pass), and live browser smoke test against Noisemaker viewer (NOISEMAKER=... node scripts/browser-smoke.mjs, all 3 checks OK). Advanced llms-full.txt snapshot block and LEDGER.md checkpoint together.
    • 2026-09-22 — caught up through noisemaker ad60cc9b / shade-mcp 00340b1: audited watched source roots across noisemaker range e5bd2013..ad60cc9b (covering work queue item #375 trigger e5bd2013..643b2be1 through HEAD). Commit 643b2be1 narrowed GAP-002 by attaching structured JSON-safe error.diagnostic (L001-L004) to thrown SyntaxError on DSL lexer failures in shaders/src/lang/lexer.js and diagnostics.js, preserving legacy error messages, types, and successful token output. Commits 52ac841b, b0364da0, a0ff705a, 2779b409, and ad60cc9b recorded gap verifications, docs and contract ledger advances. Shade MCP source unchanged at 00340b1. Verified four identities: Noisemaker source ad60cc9b62aa3b5f0674c590880898694ede5ddf (source through 2779b409fc4cfcb923cfc67c771d80ad398cd2a3), Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ delivered from release v0.2.3 (cbcab33363851016f65391fbb9ef71d66729c07f). Open gap count stands at 23 entries. Verified Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 64 tests pass) and lint, Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), structure-only check via vendored harness (206/206 pass), and live browser smoke test against Noisemaker viewer (NOISEMAKER=... node scripts/browser-smoke.mjs, all 3 checks OK). Advanced llms-full.txt snapshot block and LEDGER.md checkpoint together.
    • 2026-09-22 — caught up through noisemaker e5bd2013 / shade-mcp 00340b1: audited watched source roots across noisemaker range 50b8f909..e5bd2013. Commit 68d37721 closed GAP-022 by excluding compiled nodes marked builtin: true from mutation introspection (listSteps, replaceEffect, getCompatibleReplacements) in shaders/src/lang/transform.js. Commit e5bd2013 narrowed GAP-002 by preserving source columns from parser loc.col into location.column in shaders/src/lang/validator.js. In shade-mcp across range cbcab33..00340b1, commit 00340b1 documented paired Noisemaker compatibility and delivery checks in CLAUDE.md, with zero runtime, tool, or schema changes. Verified four identities: Noisemaker source e5bd2013087e54d53841db8c45a54f973aaa5174, Shade MCP source 00340b148e109464b1a87d89ff29fc7622d384c9, .mcp.json pinned to cbcab33363851016f65391fbb9ef71d66729c07f, and vendor/shade-mcp/ delivered from release v0.2.3 (cbcab33363851016f65391fbb9ef71d66729c07f). Open gap count stands at 23 entries. Verified Noisemaker non-parity JS tests (node scripts/run-js-tests.js --skip-parity, 64 tests pass) and lint, Shade MCP typecheck (npm run typecheck, 0 errors) and unit tests (npm test, 24 files / 157 tests pass), Shade MCP build and drop check, structure-only check via vendored harness, and live browser smoke test against Noisemaker viewer (NOISEMAKER=... node scripts/browser-smoke.mjs, all 3 checks OK). Advanced llms-full.txt snapshot block and LEDGER.md checkpoint together.
    • 2026-09-21 — caught up through noisemaker 50b8f909 / shade-mcp cbcab33: audited watched source roots across noisemaker range ea113f97..50b8f909. Commit 50b8f909 closed GAP-001 by enforcing DSL output surface range o0..o7 in shaders/src/lang/lexer.js, throwing located SyntaxError before parsing, while preserving member segment access and other reference families. Added 6 unit tests in shaders/tests/test_output_surface_range.js. Open gap count stands at 24 entries. Shade MCP unchanged at cbcab33 (18 tools, v0.2.3, protocol 2025-11-25). Verified non-parity JS test suite (node scripts/run-js-tests.js --skip-parity, 64 tests pass). Advanced llms-full.txt snapshot block and LEDGER.md checkpoint together.
    • 2026-09-21 — caught up through noisemaker ea113f97 / shade-mcp cbcab33: audited watched source roots across noisemaker range beabda38..ea113f97. Commit 0139e958 closed GAP-028 by accounting canceled pending frame-export queue frames as dropped. Commit 7706a715 / 782f0726 closed GAP-013 by pinning the Shade MCP package reference in .mcp.json to immutable commit cbcab33363851016f65391fbb9ef71d66729c07f. Commit 2f855c9c removed expired effects bc, colorspace, and hs. Shade MCP cbcab33 bumped version to 0.2.3 and migrated exemplar/catalog references to adjust (18 tools, v0.2.3, protocol 2025-11-25). Open gap count stands at 25 entries. Verified non-parity JS tests pass cleanly (57 tests, 15 nested automation tests, 4 docs tests, 1 mcp config test). Advanced llms-full.txt snapshot block and LEDGER.md checkpoint together.
    • 2026-09-20 — caught up through noisemaker beabda38 / shade-mcp 088e1ef: audited watched source roots across noisemaker range 2df19feb..beabda38. Commit 6e0166ce closed GAP-030 by enforcing backend format parity when reusing textures in pipeline.js. Commit beabda38 closed GAP-031 by requiring static integers 1..16 for legacy MIDI note mode channels in validator.js. Updated traceability matrix in llms-full.txt to remove references to closed gaps (GAP-018, GAP-023, GAP-030, GAP-031). Shade MCP unchanged at 088e1ef (18 tools, v0.2.2, protocol 2025-11-25). Verified non-parity JS tests pass cleanly (57 MIDI/audio parser tests, 15 nested automation tests, 4 docs tests). Advanced llms-full.txt snapshot block and LEDGER.md checkpoint together.
    • 2026-09-19 — caught up through noisemaker 2df19feb (shade-mcp unchanged at 088e1ef): audited watched source roots across noisemaker range 0209609e..2df19feb. Commit 2df19feb added borrowed VideoFrame support in updateTextureFromSource on both WebGL2 and WebGPU backends (synchronous display dimension extraction, immediate queue submission without ImageBitmap conversion, synchronous caller release, and anamorphic scaling rejection on WebGL2). Updated texture surface contract in llms-full.txt. Shade MCP unchanged at 088e1ef (18 tools, v0.2.2, protocol 2025-11-25). Verified Playwright VideoFrame upload tests (2/2 pass) and non-parity JS test suite pass cleanly. Advanced llms-full.txt snapshot block and LEDGER.md checkpoint together.
    • 2026-09-19 — caught up through noisemaker 0209609e / shade-mcp 088e1ef: audited watched source roots across noisemaker range ead42a5d..0209609e (zero commits touched watched source roots; GAP-023 already documented as closed at f1d2b46a). In shade-mcp across range 0a92bd83..088e1ef, single commit 088e1ef modified CI workflow files only (.github/workflows/), with zero changes to runtime tools, schemas, dependencies, or server version (18 tools, v0.2.2, protocol 2025-11-25 unchanged). Re-audited worked effect transcript note. Verified non-parity JS tests pass cleanly. Advanced llms-full.txt snapshot block and LEDGER.md checkpoint together.
    • 2026-09-18 — caught up through noisemaker ead42a5d (shade-mcp unchanged): audited watched source roots across range 688c5146..ead42a5d. Single code commit f2506d21 modified shaders/src/lang/transform.js, expanding starter-position detection in replaceEffect(), listSteps(), and getCompatibleReplacements() to recognize inline surface producers (registered starter effects with no pipeline predecessor flattened as a dependency of a surface parameter like heightTex: noise()). Updated compatibility definitions and narrowed GAP-008 in the contract. Shade MCP remains at 0a92bd83. Verified non-parity JS tests pass cleanly. Updated contract snapshot block and LEDGER.md checkpoint together.
    • 2026-09-17 — caught up through noisemaker 688c5146 (shade-mcp unchanged): audited WebGPU frame export row-orientation fix (5ceb97ba and 688c5146). Documented that WebGPU frame export resolve inverts vertical orientation in the resolve shader (textureDimensions(sourceTexture).y - 1 - i32(position.y)) to match canvas presentation and WebGL2 parity. Verified non-parity JS tests and WebGPU frame export unit tests all pass cleanly.
    • 2026-09-15 — caught up through noisemaker 9a31314c (shade-mcp unchanged): re-audited the perspective/depth-sort/defocus/3D-landscape round's 7 commits touching the watched source roots. Narrowed GAP-005 — conditions used to be on its uncopied-pass-field list; expander.js now copies it and Pipeline.shouldSkipPass() consumes it at execution time, the mechanism this round's per-viewMode/blendMode pass-cloning pattern relies on to select which cloned pass variant runs (name, viewport, clear, samplerTypes, type remain uncopied). Added PassSpec.defines to the typed grammar (undocumented before this pass): it does not land on the expanded pass object directly, but derives a __KEY_value-suffixed compiled-program variant merging effect- and pass-level compile-time defines. Documented that a numeric PassSpec. uniforms value bypasses global lookup as a literal compile-time constant, that evaluateAutomation() now rounds int-typed globals' resolved automation value, that expander.js follows volume/geometry surface dimensions through write3d()/read3d() handoffs (exports, forward reads, repeated same-surface filters) instead of resetting to the 64 default, and that a custom vertex shader now receives the same #define injection as the fragment stage. Re-resolved the unpinned Shade MCP package to its current commit: still 0a92bd83, no drift, so the MCP-side sections and worked-example transcript needed no re-capture — confirmed by source inspection that the worked synth/testPattern effect's own files are unchanged in this range and none of the round's expander/pipeline changes are reachable by an effect using none of those mechanisms. The full non-parity JS test suite passed.
    • 2026-09-07 — caught up through noisemaker 246ff57f / shade-mcp 0a92bd83: documented lossless vec4 formatting, CC/CC14, NRPN, pressure, bend, MPE selection, port inventories/lifecycle, default audio channels, and aggregate/raw cleanup. Narrowed GAP-031 to legacy note modes, extended GAP-032 to default channels, and closed GAP-018 after the tool descriptions stopped promising glob support. Rechecked changed-source engine and MCP gaps. Resolved the runtime mirror to the pinned Shade commit, built its locked dependencies, and captured the 0.2.2/2025-11-25 handshake, all 18 schemas, and eight worked calls in one session. Tool descriptions and error strings changed; argument types/defaults did not. The exact two-pass source graph reproduced, while the compile reports still demonstrate GAP-024. Noisemaker language/runtime and focused MIDI suites passed; Shade's 157 tests, typecheck, build, and standalone-drop checks passed.
    • 2026-09-04 — caught up through noisemaker 691eea16 / shade-mcp 23258ca3: re-audited selected MIDI ports, selected audio device/channels, bipolar raw audio, recursive capture requirements, and eight-level nested automation across the changed parser, validator, unparser, external-input, and pipeline sources. Added the exact mixed-argument exception, selector validation/round-trip contract, runtime state resolution, nested-field ranges and deterministic phase integration, and host capture requirements. Opened GAP-031 for the unenforced MIDI channel range and GAP-032 because the built-in audio manager cannot populate selected-device or raw-ready state; rechecked the other changed-source gaps without closing them. Re-resolved the configured GitHub runtime to 23258ca3, whose drift is README-only, and recaptured the shade-mcp/0.2.2/2025-11-25 handshake and all 18 unchanged tool schemas in one fresh session. Recaptured all eight worked calls and reproduced the exact two-pass source graph. The returned responsiveness fields do not belong to synth/testPattern, strengthening GAP-024's stale-graph finding rather than validating that response.
    • 2026-08-31 — caught up through noisemaker a4701daf / shade-mcp ba407982: verified that Noisemaker's watched source roots are unchanged from the prior checkpoint, re-resolved the unpinned GitHub package to the exact Shade MCP commit, and recaptured the shade-mcp/0.2.2/2025-06-18 handshake and all 18 tool signatures from the configured runtime. The tool names, input schemas, absent output schemas, and forbidden task support are unchanged. Audited the Shade MCP dependency/release changes and documented its build-time version injection plus standalone-drop handshake gate; its tests, typecheck, build, external dependency gate, and bare-drop check all passed. Rechecked every MCP-side gap; none closed. Recaptured the full eight-call worked transcript in the same 0.2.2 session; the seven deterministic responses were byte-for-byte unchanged, the FPS sample still passed its threshold, and the results continue to expose GAP-024 rather than proving end-to-end backend identity. The source-level worked program also reproduced its exact two-pass graph.
    • 2026-08-29 — caught up through noisemaker c767e481 / shade-mcp ac4c6ba1: re-audited the two watched device-limit commits and documented the expanded capability object, numeric volume-atlas clamping across graph, host, and UI paths, WebGL MRT budget probing/error cleanup, and ordered full-to-half-float MRT adaptation across WebGL2/WebGPU. Rechecked the changed-source evidence for GAP-001/004/005/006/007/016/026; none closed. Opened GAP-030 because MRT adaptation mutates graph formats without invalidating same-size reused backend textures. Reproduced the worked source graph and device-limit regression gates. Shade MCP did not move in this work queue range, so its handshake, 18 tool schemas, historical browser transcript, and MCP-side gaps remain pinned unchanged.
    • 2026-08-14 — caught up through noisemaker f6b22ab3 / shade-mcp ac4c6ba1: verified that no Noisemaker watched source root changed, re-resolved the unpinned MCP package, and recaptured the shade-mcp/0.2.1/2025-06-18 handshake, all 18 tool schemas, the exact two-pass source graph, and the eight-call worked transcript. Re-audited the affected MCP contract sources and every MCP-side gap; documented the configurable browser/AI timeouts, headless/session lifecycle, restricted loopback/opaque-origin CORS, opt-in description image, refreshable effect index, and manifest invalidation, narrowed GAP-020 now that pixel parity also uses the shared isError wrapper, and opened GAP-029 for invalidation racing an already in-flight index rebuild.
    • 2026-08-11 — caught up through noisemaker e1feefa0 / shade-mcp 346ee022: documented the renderer sink and bounded asynchronous frame export contracts across WebGL2/WebGPU, rechecked every changed-source gap, and opened GAP-028 for pending accepted frames canceled without a terminal queue result. Re-resolved the unpinned MCP package; recaptured the shade-mcp/0.1.4/2025-06-18 handshake, all 18 tool schemas, and the eight-call worked transcript after the Node 22/SDK/Zod upgrades.
    • 2026-08-10 — caught up through noisemaker 7cc4894b / shade-mcp 3e531fc6: re-audited Noisemaker runtime drift; re-resolved the unpinned MCP package; recaptured the shade-mcp/0.1.4/2025-06-18 handshake, all 18 tool signatures, and the worked transcript; narrowed GAP-020 for the new whole-call isError marker; closed GAP-025.
    • 2026-07-14 — contract instrumented (dc67827b) at snapshot noisemaker 75507112 / shade-mcp 7fd0d975: nine surface sections, worked validated effect, 9×4 traceability matrix, 26-gap register. Audited clean through 478989b6; checkpoint established at the contract's own snapshot.

AI contract checkpoint

Last good source revision: 61e80cc4cc4fc40fa6c6396bd2f2da2e2ed0dc32.

The tracked AI contracts were independently audited against that source revision.

  • llms-full.txt: ec20487bac1532dda4b017328df86f2380fc43b0
  • llms.txt: 4571eb66007cf606c868b79e6d7103b3b39388f7
{
  "version": 1,
  "repository": "noisefactorllc/noisemaker",
  "branch": "main",
  "last_good": {
    "revision": "61e80cc4cc4fc40fa6c6396bd2f2da2e2ed0dc32",
    "contracts": [
      {
        "path": "llms-full.txt",
        "sha": "ec20487bac1532dda4b017328df86f2380fc43b0"
      },
      {
        "path": "llms.txt",
        "sha": "4571eb66007cf606c868b79e6d7103b3b39388f7"
      }
    ],
    "references": [
      {
        "repository": "noisefactorllc/shade-mcp",
        "branch": "main",
        "revision": "e67a3b474a750c1f187c9b60d9da7a8964f060f9"
      }
    ],
    "verified_at": "2026-10-25T21:02:48.946Z",
    "audit": {
      "job_id": "11d484a0-9f4e-4e6e-b1ac-6e8b9f978bd9",
      "review_id": "44e0a078-6136-448e-a0fc-c8e3cb4897e9",
      "reviewer_session_id": "20261025_210119_0a5d5b"
    }
  }
}