Keyed chart comparison: host and export qualification

Original qualification: 2026-10-08. Task 4 baseline: 6e2afbd, clean managed keyed-chart-comparison worktree.

Latest source checkpoint: f0705f2. Independent final re-review confirms all seven source findings addressed, with no new Important/Critical regressions. The single final source correction wave from cac26d1 addresses I1-I3 and M1-M4, with 529 workspace tests, check/typecheck, both app builds, Storybook and a new standalone install/typecheck/build passing. The Task 4/5 sections below are historical evidence. See the final correction section for current source fixes and /private/tmp/comparison-final-preview. Native/browser acceptance, including visual rechecks of I2/I3, remains pending exclusive Firefox access. No browser was operated during this wave.

Release preparation was authorised on 2026-10-09. The release branch integrates main’s dataset explorer and deployment support without claiming the browser checklist is complete. Statements below about missing merge/push authority and preview processes are historical, not the current release status. Current release behaviour and compatibility changes are summarised in the release notes.

Scope

The /comparison Astro route contains one client:load island under the existing Layout, reached from Charts. Its literal lazyApp loader and the Charts/Comparison Workbench Story use the same maintained application exports. Host initialisation ignores Astro props and delegates to the state initializer with deterministic initialSettings. Plain comparison.css is imported by the shared view; no additional CSS compiler or runtime was introduced.

Automated evidence

Focused RED: bun run --filter @opsydyn/web test src/apps/comparison/host.test.ts exited 1 before host implementation: four failing assertions for the absent application configuration, Story, and route. Log: /tmp/comparison-task-4-red.log.
Focused GREEN: the same command exited 0, four tests passing. Final static-import test log: /tmp/comparison-task-4-green-final.log.
Full web suite: bun run --filter @opsydyn/web test exited 0, 202 tests in 36 files. Log: /tmp/comparison-task-4-web-tests.log.
Web production build: bun run --filter @opsydyn/web build exited 0, no warnings observed. Log: /tmp/comparison-task-4-web-build.log.
Storybook production build: bun run --filter @opsydyn/web build-storybook exited 0. Vite warns that some minified chunks exceed 500 kB; this is not a failed build or browser evidence. Log: /tmp/comparison-task-4-storybook-build.log.
Final workspace gates ran sequentially: bun run check exited 0; bun typecheck exited 0 (zero errors, warnings or hints); NPM_CONFIG_CACHE=/private/tmp/foldkit-npm-cache bun run test exited 0 with 484 tests passing: Astro integration 77, Viz 168, promo 37, web 202. Logs: /tmp/comparison-task-4-check.log, /tmp/comparison-task-4-typecheck-final.log, /tmp/comparison-task-4-tests.log.
git diff --check exited 0. Storybook’s generated index includes charts-comparison--workbench with ./src/stories/comparison.stories.ts.

The happy-dom host test exercises actual application exports and runtime: ordered accessible panel names; add both types to four and disabled fifth; keyboard histogram inspection selecting two data points with four highlighted marks across two scatters; reorder retaining keyed nodes and inspection; source removal clearing overlays; remove-all and re-add as Scatter 5; no further update calls or DOM changes after disposal. The Story test mounts via the existing helper and adds a real panel. Source assertions cover single-island wiring and Charts navigation, not live hydration or runtime multiplicity.

Browser acceptance: explicitly pending

The controller owns live browser QA using native CUA in a dedicated Firefox Developer Edition tab. No connected browser provider was available. The controller initially reported that CUA getApp found the Mac locked and auto-unlock failed. The Mac is now unlocked. The controller performed the standalone observations below, then stopped after unrelated navigation and an invalidated CUA binding. Further UI QA is pending exclusive Firefox access. The lock is no longer a blocker. No bypass or unlock attempt was made by this implementer. Task 4 host observations are still awaited; the supplied standalone observations are recorded below. Controller QA remains separate from the HTTP smoke below.

All Task 4 browser gates remain pending:

Use .superpowers/sdd/2026-10-08-keyed-chart-comparison/browser-checklist.md. Record controller-supplied observations separately before final task review. Automated DOM tests do not qualify real layout, native input, VoiceOver, physical devices, or browser hydration. Promo and standalone live acceptance are in progress under the controller, currently paused pending exclusive browser access. No push, merge, publication or deployment occurred.

Controller preview handoff

From this worktree, choose an unused port (replace 4337 if occupied):

bun run --filter @opsydyn/web preview --host 127.0.0.1 --port 4337

Open http://127.0.0.1:4337/comparison and verify the Charts link at http://127.0.0.1:4337/charts. The production build is already present. The configured Cloudflare adapter uses a local preview; it may require localhost permissions outside the sandbox. Stop only this temporary server after QA. This command was verified against the installed preview script/adapter, not executed; no server is being left running.

Baseline: 70b7c7b, clean managed worktree. The promo wrapper owns export/source state and folds the maintained workbench with foldChild and foldChildInit. It does not duplicate the reducer, panel matching, chart math or runtime. Settings are captured when Commands are created. The export replaces only src/web/src/apps/comparison/initial-settings.ts; schema/validation stays in the unchanged captured settings.ts.

The TypeScript-parser collector follows static imports and re-exports under the two approved app roots, preserves apps-relative paths, includes plain CSS, and rejects missing, escaping, excluded and symlink-escaping sources. Collection also runs after Astro relocates its prerender chunk. The existing Viz vendor walker includes consumed subpaths and compiled dependencies. No package API, runtime, lockfile or dependency-version changes were made.

RED/GREEN and automated gates

RED: bun run --filter @opsydyn/promo test test/comparison-source.test.ts test/comparison-export.test.ts test/comparison-state.test.ts exited 1: 37 existing tests passed and four new files failed to load the absent comparison contracts. The workspace script also discovers the standalone test. Log: /tmp/comparison-task-5-red.log.
GREEN: the same command with NPM_CONFIG_CACHE=/private/tmp/foldkit-npm-cache exited 0: 49 tests, zero failures. Log: /tmp/comparison-task-5-green.log. Twelve new tests cover source identity/path safety, relocated collection, exact [2,3,1] snapshots with linking off and counter 4, transient-state exclusion, independent exports, duplicate-ID rejection, pending guards, actual clipboard delivery, a delayed template fetch, failure/retry, delayed keyed child completion, shared view and suppressed standalone export controls.
Relocation regression RED/GREEN: bun test test/comparison-source.test.ts from apps/promo, exits 1 then 0. Logs: /tmp/comparison-task-5-relocated-{red,green}.log. The initial Astro build reproduced the wrong source root; the final build includes both the promo route and download JSON successfully.
Final workspace lint/format: bun run check, exit 0, no lint warnings. Log: /tmp/comparison-task-5-check.log.
Final typecheck: bun typecheck, exit 0, zero errors/warnings/hints. Log: /tmp/comparison-task-5-typecheck.log.
Final tests: NPM_CONFIG_CACHE=/private/tmp/foldkit-npm-cache bun run test, exit 0, 496 passing (Astro 77, Viz 168, promo 49, web 202). Log: /tmp/comparison-task-5-tests.log.
Final app builds: env -u FOLDKIT_BUILD_ID bun run build, exit 0, no warnings observed. Log: /tmp/comparison-task-5-build.log.
Final Storybook: bun run --filter @opsydyn/web build-storybook, exit 0. Existing >500 kB warning remains (axe 579.43 kB, iframe 812.55 kB). Log: /tmp/comparison-task-5-storybook.log. No chunk-warning remediation or Task 4 fixed-50ms disposal-wait changes belong to Task 5.

The final tests and builds were awaited sequentially. An earlier invalid run overlapped tests with rebuilding package output and caused ten web import-resolution failures. It is retained as /tmp/comparison-task-5-tests-invalid-overlap.log, not used as qualification evidence. The complete isolated rerun passed.

Standalone project and controller preview

The comparison test and all three existing line/histogram/scatter standalone tests each generate an external temporary project, run npm install --no-audit --no-fund, npm run typecheck and npm run build, and remove their fixture. No workspace: dependency or dangling app alias remains.

A separate fixture is preserved at /private/tmp/comparison-task-5-preview (/tmp/comparison-task-5-preview resolves to the same location on this Mac). It contains 64 generated files plus installed dependencies and a production build. Its exact initial settings are:

{
  "panels": [
    { "id": 2, "kind": "histogram" },
    { "id": 3, "kind": "scatter" },
    { "id": 1, "kind": "scatter" }
  ],
  "linkInspections": false,
  "nextPanelId": 4
}

Preserved-fixture install, typecheck and build all exited 0. Logs are /tmp/comparison-task-5-preview-{install,typecheck,build}.log. A brief implementer preview on port 4389 returned HTTP 200 for HTML, JavaScript and CSS; log: /tmp/comparison-task-5-preview-http.log. That server was stopped. This is transport/asset evidence only, not browser execution or layout qualification.

The controller used native Firefox for the observed subset, then stopped its preview session (31404) after unrelated navigation invalidated the CUA binding. The following command restarts the unchanged fixture once exclusive access is available:

cd /private/tmp/comparison-task-5-preview
npm exec -- vite preview --host 127.0.0.1 --port 4349 --strictPort

Open http://127.0.0.1:4349/ after restarting the preview. The fixture is frozen for this QA session. Do not regenerate it during QA. Afterwards, /private/tmp/comparison-task-5-generate.ts reproduces it from maintained sources, followed by the three npm commands above.

Promo preview, using an unused port, from the managed worktree:

bun run --filter @opsydyn/promo preview --host 127.0.0.1 --port 4350

Visit /examples/comparison/ and /downloads/comparison-template.json. The reference Astro preview instructions above remain applicable.

Controller-observed standalone browser results

The controller supplied these native Firefox observations on 2026-10-08. These were not observed independently by the implementer:

Captured order Histogram 2, Scatter 3, Scatter 1; linking off, counter 4.
Enabling linking and histogram keyboard inspection produces six matches, with dashed outlines on both scatter panels.
Reorder twice using Return retains the same focused move control at the endpoint.
Adding Histogram allocates ID 4 and disables both add controls.
At 390px the panels use a single column and controls fit.
Visual correction: the maintained scatter’s vertical SALARY($) label overlaps the 100000 y tick at desktop and 390px. The controller assigns this to the final fix/review wave, not Task 5; shared chart code and the preserved preview remain unchanged.

Remaining acceptance

Further controller native Firefox observations for reference and promo.
Complete removal/focus, pointer hit testing after movement/resize/scroll, reduced motion, teardown and runtime/console checks.
Real ZIP download and StackBlitz delivery in the hosted promo page.
Controller independent final source review and label-spacing correction.
Post-fix native visual recheck of label spacing and shared-chart focus.

These gates are pending exclusive Firefox access, not blocked by a locked Mac. The controller stopped UI actions and requested a brief exclusive session after its tab was navigated to unrelated Rosterium activity. The observed standalone subset remains valid; full 390px, host/promo, console and reduced-motion checks are not claimed. The generated fixture remains unchanged. No UI tools were used by the Task 5 implementer. Browser, native accessibility and device acceptance remain separate. No push, merge, publication or deployment occurred.

Single final source correction wave

Baseline: cac26d161a538edd45dacd914e5981970712904b, initially clean. The exact commit and per-finding commands/results are in .superpowers/sdd/2026-10-08-keyed-chart-comparison/final-fix-report.md. No subagents, new worktree, renderer/public-package API changes, dependency changes, push, merge, publication or preview servers were used.

Corrections and regression evidence

Finding	Maintained-source correction	Automated evidence
I1	Scoped Mount measures each pointer event and synchronously forwards an element-local, schema-checked custom event to the normal FoldKit DOM dispatcher. Leave/keyboard remain synchronous. No delayed clear, parent workaround, external ordering state or DOM references in Model/view/update.	Four actual `Runtime.embed` same-task tests failed before correction: move/leave, move/leave/histogram entry, move/scatter key, move/histogram key. All pass without yielding between input events. Separate-task control, fresh geometry after reorder/scroll/resize, and listener disposal remain covered.
I2	Default scatter left gutter is 104px, budgeting separate title, gap and six-digit tick bands; existing readable font sizes and scale/client-coordinate contracts are retained.	Two salary-fixture render/layout regressions failed with tick x=44 and now pass at chart widths 380 and 342. Existing actual-runtime pointer/geometry tests pass with the changed margin. These are not measured native glyph bounds or visual rechecks.
I3	Shared SVG root no longer suppresses the native outline. Comparison-specific focus-visible styling remains unchanged; no tab stop was added.	Three mounted-runtime tests failed on `outline: none` and now pass for standalone histogram, standalone scatter and both diagnostics charts outside comparison. Native focus appearance and assistive technology remain unqualified.
M1	Unselected reverse navigation selects `n - 1`; forward starts at zero.	One/two/three-item tests cover all four arrow keys, first selection and wraparound; histogram output agrees with pointer inspection. RED included the one-item negative-zero result.
M2	Scatter resize rejects a resulting `pw <= 0`, retaining the last usable layout and inspection.	Below/equal-to-margin-sum rejection and positive recovery failed before correction and now pass.
M3	Legacy linked charts and comparison matching use the same small pure app-owned `containsValue` predicate.	New shared-callable contract was RED before export; endpoint truth table and existing consumer tests pass. This removes equivalent duplication, not a previously observed endpoint divergence.
M4	Host test waits for every chart observer’s disconnect and the scatter listener’s removal before exercising disposed targets; drains one zero-delay event-loop task afterwards.	Removing observer finalisation temporarily made the strengthened test fail (expected one disconnect, received zero). Restoring it passes. This mutation check qualifies the test improvement, not a claim of a pre-existing production leak. Both 50ms sleeps are removed.

Focused logs are /private/tmp/comparison-final-i1-{red,green}.log, comparison-final-chart-{red,green}.log, comparison-final-i3-red.log, comparison-final-m3-red.log, comparison-final-m3-m4-green.log, and comparison-final-m4-{red,green}.log under the same directory. Final focused groups passed 25, 62 and 38 tests; the dedicated disposal suite passed all four tests. The final complete workspace suite subsumes these groups.

Final sequential gates

All commands below exited zero; builds and tests did not overlap:

bun run check: no lint/format errors or warnings.
bun typecheck: zero errors, warnings or hints, including 538 web files.
NPM_CONFIG_CACHE=/private/tmp/foldkit-npm-cache bun run test: 529 passing, zero failures (Astro 77, Viz 168, promo 55, web 229 in 38 files).
env -u FOLDKIT_BUILD_ID bun run --filter @opsydyn/web build: no warnings.
env -u FOLDKIT_BUILD_ID bun run --filter @opsydyn/promo build: no warnings.
env -u FOLDKIT_BUILD_ID bun run --filter @opsydyn/web build-storybook: completed with the retained greater-than-500kB advisory, axe 579.43kB and iframe 812.55kB. No bundle attribution/performance claim or optimisation.
git diff --check: no whitespace errors.

Logs: /private/tmp/comparison-final-{check,typecheck,tests,web-build,promo-build,storybook}.log.

Distinct corrected standalone fixture

bun /private/tmp/comparison-final-generate.ts generated 64 files from current maintained source into /private/tmp/comparison-final-preview. Captured settings are Histogram 2, Scatter 3, Scatter 1; linking off; nextPanelId: 4. Only the data-only initial settings are replaced; validation remains present. From that directory, all three commands exited zero:

NPM_CONFIG_CACHE=/private/tmp/foldkit-npm-cache npm install --no-audit --no-fund
npm run typecheck
npm run build

Logs: /private/tmp/comparison-final-preview-{generate,install,typecheck,build}.log. Install added 41 packages. The standalone build produced a 470.06kB JS asset without warnings. No preview server was started and no browser inspected it.

The frozen /private/tmp/comparison-task-5-preview remains unchanged. Read-only before/after fingerprints cover all 7,468 files (122,624,897 bytes), paths, metadata and symlink targets, including dependencies and build output. Both SHA-256 fingerprints are 4e15c802a05d0e01d4eb2cae7da279b285c0f0ab4cbaea70a0d84efada0e1d17. cmp of /private/tmp/comparison-final-old-preview-{before,after}.json exited zero. Earlier screenshots belong to that old fixture, not to the corrected source.

Acceptance boundary

All seven source corrections are implemented and automatically qualified; no unresolved source finding was identified in self-review. This does not close the controller’s observed axis overlap: its post-fix desktop and 390px visual recheck, native shared-chart focus (including diagnostics), the complete reference/promo journeys, pointer checks, reduced motion, teardown, console, runtime multiplicity, real ZIP/StackBlitz delivery and assistive-technology/device checks remain pending. Exclusive Firefox permission has been requested but not approved. Merge readiness remains a separate controller decision; no merge, push or publication is authorised or claimed.

Controller verification and current preview

At f0705f2, the controller independently ran these sequential gates, all with exit zero: bun run check, bun typecheck, NPM_CONFIG_CACHE=/private/tmp/foldkit-npm-cache bun run test, env -u FOLDKIT_BUILD_ID bun run build, and bun run --filter @opsydyn/web build-storybook. All 529 tests passed (77 integration, 168 Viz, 55 promo, 229 web). The Storybook chunk advisory remains; lint and typecheck are clean. Logs: /tmp/comparison-controller-final-{check,typecheck,tests,build,storybook}.log.

Independent scoped re-review checked I1-I3 and M1-M4 and found all addressed in source, with no unresolved source findings. The pointer bridge guarantees synchronous ordered enqueue into FoldKit’s FIFO queue, not unconditional reducer completion before a subsequent event when the runtime yields. Reports and the execution ledger remain in this plan’s ignored SDD directory while browser acceptance is outstanding.

A local development server is now available at http://127.0.0.1:4347/comparison (Astro PID 63378), started from this managed worktree with bun run --filter @opsydyn/web dev --host 127.0.0.1 --port 4347. The route returned HTTP 200. This is transport evidence, not browser acceptance. Use bun run --filter @opsydyn/web astro dev stop in this worktree to stop it. No unrelated server or browser tab was changed. No merge, push or publication.