docs: propose local artifact core certification

.github/workflows/ci.yml

@@ -39,3 +39,33 @@ jobs:
             release-evidence/
             opencode-artifacts-*.tgz
           retention-days: 14
+
+  portable-platforms:
+    name: portable-platform-${{ matrix.os }}
+    needs: verify
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, macos-latest, windows-latest]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-node@v4
+        with:
+          node-version: "24"
+          cache: npm
+      - run: npm install -g npm@latest
+      - run: npm ci
+      - run: npm run build
+      - uses: actions/download-artifact@v4
+        with:
+          name: exact-candidate-evidence
+          path: candidate-input
+      - name: Clean packed install and portable-output smoke
+        run: node scripts/candidate-platform-smoke.ts --output platform-evidence/${{ runner.os }}.json --candidate-dir candidate-input
+      - uses: actions/upload-artifact@v4
+        with:
+          name: portable-platform-${{ runner.os }}-${{ runner.arch }}
+          path: platform-evidence/*.json
+          if-no-files-found: error
+          retention-days: 14

.gitignore

@@ -1,4 +1,5 @@
 node_modules/
+.goal5-private/
 docs/evidence/journeys/raw/
 dist/
 *.tgz

benchmarks/page-quality/v1/benchmark-manifest.json

@@ -0,0 +1,45 @@
+{
+  "schemaVersion": 1,
+  "corpusId": "page-quality-v1",
+  "systems": ["opencode", "claude"],
+  "taskIds": [
+    "dashboard",
+    "incident",
+    "pr-walkthrough",
+    "system-explainer",
+    "compare",
+    "plan-checklist",
+    "findings-table",
+    "interactive-decision"
+  ],
+  "minimumRunsPerSystemTask": 3,
+  "captureModes": [
+    { "id": "desktop-light", "width": 1440, "height": 900, "colorScheme": "light" },
+    { "id": "mobile-dark", "width": 390, "height": 844, "colorScheme": "dark" }
+  ],
+  "hardGates": [
+    "facts",
+    "interactions",
+    "layout",
+    "composition",
+    "accessibility",
+    "dataHonesty",
+    "offlineBudget"
+  ],
+  "dimensions": [
+    "taskOrientation",
+    "informationHierarchy",
+    "compositionDensity",
+    "typographyReadability",
+    "visualEncoding",
+    "coherenceCraft",
+    "interactionQuality",
+    "responsiveAdaptation"
+  ],
+  "thresholds": {
+    "openCodeBetterOrEquivalent": 0.8,
+    "openCodeDimensionMedian": 4,
+    "minimumReviewers": 3,
+    "requiredReviewerRoles": ["design-ux", "technical"]
+  }
+}

benchmarks/page-quality/v1/benchmark.template.json

@@ -0,0 +1,17 @@
+{
+  "schemaVersion": 1,
+  "runId": "pending-authorized-run",
+  "corpusId": "page-quality-v1",
+  "manifestSha256": "2badb79c0f21649b8ed6cb72a25fd8ca2b0a9436faf29d7ce8296384cf5a8897",
+  "authorization": {
+    "status": "pending",
+    "authorizedBy": null,
+    "authorizedAt": null,
+    "claudeAccountScope": null,
+    "modelProtocol": null,
+    "retentionDisposition": null
+  },
+  "generations": [],
+  "reviewers": [],
+  "pairs": []
+}

benchmarks/page-quality/v1/corpus.json

@@ -0,0 +1,124 @@
+{
+  "schemaVersion": 1,
+  "corpusId": "page-quality-v1",
+  "license": "CC0-1.0",
+  "provenance": "Project-authored synthetic fixtures; no private repository, account, or Anthropic output is included.",
+  "captures": [
+    { "id": "desktop", "width": 1440, "height": 900 },
+    { "id": "mobile", "width": 390, "height": 844 }
+  ],
+  "bundles": [
+    {
+      "id": "dashboard",
+      "task": "Operational KPI dashboard",
+      "fixture": "examples/patterns/dashboard.md",
+      "fixtureSha256": "06081c9519780f9f74c22886026552d6f49deb3243de134c619b5a00b8c6b658",
+      "primaryReader": "Release engineering lead",
+      "readerDecision": "Prioritize the cache TTL correction before investigating lower-volume services.",
+      "prompt": "Create an offline operational dashboard from only the supplied synthetic facts. Lead with the weekly deploy-failure finding, show the daily trend and service breakdown, preserve provenance, and do not invent causes or dates.",
+      "requiredFacts": ["There were 61 failures across 1,284 deploys.", "Wednesday and Thursday account for 26 failures tied to platform#412.", "svc-payments has 19 failures and 8 rollbacks."],
+      "forbiddenClaims": ["The cache TTL caused failures outside Wednesday and Thursday.", "Any service not present in the source bundle was measured."],
+      "sourceBundle": { "license": "CC0-1.0", "provenance": "Synthetic deployment telemetry", "missingValues": ["No owner is supplied for svc-web."], "longLabel": "Mean recovery time after a failed multi-service production deployment" },
+      "interactions": [{ "id": "inspect-chart", "action": "Focus the primary chart and read its equivalent.", "expectedState": "The equivalent names the Wednesday peak of 14 and Saturday low of 5." }],
+      "stressCases": ["long-label", "missing-value", "dense-data", "narrow-viewport"]
+    },
+    {
+      "id": "incident",
+      "task": "Investigation timeline and postmortem",
+      "fixture": "examples/patterns/incident.md",
+      "fixtureSha256": "d4235196c8543314107206ce1e8fed61b4a5bb2f758fe24ae5d2cdfcb77704ce",
+      "primaryReader": "Incident commander",
+      "readerDecision": "Keep the rollback and implement the fraud check asynchronously with a 200 ms budget.",
+      "prompt": "Turn the supplied incident facts into an offline postmortem with status, chronology, trend, root cause, tested evidence, and next action. Preserve exact versions and times; do not infer customer loss.",
+      "requiredFacts": ["Peak checkout p99 was 2.6 seconds during a 38-minute incident.", "svc-payments 1.88.0 added a synchronous fraud check.", "Rollback to 1.87.2 restored p99 below 400 ms."],
+      "forbiddenClaims": ["Orders were lost.", "The fraud provider had an outage."],
+      "sourceBundle": { "license": "CC0-1.0", "provenance": "Synthetic incident log", "missingValues": ["No affected-user count was collected."], "longLabel": "Time from synchronous fraud-check deployment until three-minute recovery confirmation" },
+      "interactions": [{ "id": "scan-timeline", "action": "Read the timeline in document order.", "expectedState": "Alert, suspect, rollback, and mitigation remain chronologically ordered." }],
+      "stressCases": ["long-label", "missing-value", "dense-data", "narrow-viewport"]
+    },
+    {
+      "id": "pr-walkthrough",
+      "task": "Annotated change review",
+      "fixture": "examples/patterns/pr-walkthrough.md",
+      "fixtureSha256": "dc1b092883ff432949dad08476cf01262f2f6cf0d0710aeae8a0ac4c8ecb54d5",
+      "primaryReader": "Pull-request reviewer",
+      "readerDecision": "Request changes until the default returns to 300 seconds and invalid TTL values are rejected.",
+      "prompt": "Create an offline PR walkthrough from the supplied diff and findings. Order findings by severity, annotate exact locations, show the relevant diff, tests, and a clear merge verdict without inventing code.",
+      "requiredFacts": ["The default TTL changes from 300 seconds to 5 seconds.", "ttlSeconds lacks validation.", "The feature flag itself is acceptable after the two fixes."],
+      "forbiddenClaims": ["The PR is safe to merge unchanged.", "Validation exists elsewhere in the repository."],
+      "sourceBundle": { "license": "CC0-1.0", "provenance": "Synthetic PR diff", "missingValues": ["No benchmark result accompanies the PR."], "longLabel": "Default cache time-to-live behavior when callers omit an explicit ttlSeconds value" },
+      "interactions": [{ "id": "copy-verdict", "action": "Select the verdict for return to the coding session.", "expectedState": "The exact two blocking changes remain visible." }],
+      "stressCases": ["long-label", "missing-value", "dense-data", "narrow-viewport"]
+    },
+    {
+      "id": "system-explainer",
+      "task": "Architecture and data-flow explanation",
+      "fixture": "examples/patterns/system-explainer.md",
+      "fixtureSha256": "80fda196b54484e4902eb25bbf1e4a6248e7dc772ea99fff061c0cd9a53f0739",
+      "primaryReader": "Security-conscious developer",
+      "readerDecision": "Use portable HTML offline unless local collaboration or a registered datasource is explicitly needed.",
+      "prompt": "Explain the supplied artifact publication data flow as an offline page. Label validation, permission, storage, viewing, and datasource boundaries; include a sequence or relationship diagram and a clear reader decision.",
+      "requiredFacts": ["Validation happens before permission and writes.", "A denied operation writes nothing.", "Portable HTML uses connect-src none and adds no view-time authority."],
+      "forbiddenClaims": ["Portable HTML can execute a datasource.", "The viewer needs OpenCode installed."],
+      "sourceBundle": { "license": "CC0-1.0", "provenance": "Synthetic architecture contract", "missingValues": ["No hosted identity boundary is in this local-core diagram."], "longLabel": "Registered datasource command execution under a separately approved loopback-service permission" },
+      "interactions": [{ "id": "read-diagram", "action": "Read the diagram equivalent without graphics.", "expectedState": "The same validation, permission, commit, and offline-view order is conveyed." }],
+      "stressCases": ["long-label", "missing-value", "dense-data", "narrow-viewport"]
+    },
+    {
+      "id": "compare",
+      "task": "Alternative comparison",
+      "fixture": "examples/patterns/compare-layouts.md",
+      "fixtureSha256": "7f2e8571019c4572c915427c412cbf5e1edb2ea4c14c10d04a58b58b89c00070",
+      "primaryReader": "Settings product owner",
+      "readerDecision": "Ship two-column tabs by default and keep the dense table behind a toggle.",
+      "prompt": "Compare the four supplied settings layouts in a consistent offline decision page. Make each variant genuinely distinct, preserve every tradeoff, and recommend only from the supplied hallway-test result.",
+      "requiredFacts": ["Two-column tabs are best for known-item visits.", "The dense table is fastest for power users.", "Card-grid modals break keyboard flow."],
+      "forbiddenClaims": ["A production A/B test was run.", "The card grid is accessible with no changes."],
+      "sourceBundle": { "license": "CC0-1.0", "provenance": "Synthetic design workshop", "missingValues": ["No numeric task-completion times are available."], "longLabel": "Two-column tabs with deep links for returning users who already know the target setting" },
+      "interactions": [{ "id": "compare-variants", "action": "Read every variant and tradeoff in the same order.", "expectedState": "Four complete alternatives remain distinguishable at narrow width." }],
+      "stressCases": ["long-label", "missing-value", "dense-data", "narrow-viewport"]
+    },
+    {
+      "id": "plan-checklist",
+      "task": "Implementation and release plan",
+      "fixture": "examples/patterns/release-checklist.md",
+      "fixtureSha256": "c94a67977c27fdb1c06d0b085a8fb2033f8c1d787c2a7e050d2e42c2b28fc288",
+      "primaryReader": "Release owner",
+      "readerDecision": "Restore the performance benchmark cluster and obtain the security sign-off before tagging.",
+      "prompt": "Create an offline release checklist from the supplied state. Show progress, completed and open work, dependencies, risks, and the next blocking action. Do not treat post-tag work as a pre-tag gate.",
+      "requiredFacts": ["Seven of eleven items are complete.", "The performance baseline is blocked because the benchmark cluster is down.", "Security sign-off on the token endpoint is still open."],
+      "forbiddenClaims": ["The release is ready to tag.", "Post-tag dashboard watch blocks the tag."],
+      "sourceBundle": { "license": "CC0-1.0", "provenance": "Synthetic release plan", "missingValues": ["The benchmark cluster recovery owner is not assigned."], "longLabel": "Performance baseline re-run against the restored two-core four-gibibyte benchmark environment" },
+      "interactions": [{ "id": "scan-checklist", "action": "Navigate the task list by keyboard.", "expectedState": "Checked state and the two pre-tag blockers are announced." }],
+      "stressCases": ["long-label", "missing-value", "dense-data", "narrow-viewport"]
+    },
+    {
+      "id": "findings-table",
+      "task": "Prioritized findings table",
+      "fixture": "examples/patterns/license-audit.md",
+      "fixtureSha256": "7703cfec91a82278ea9f7d222fdd247eddb3632733a4ff4bad0310079a26e8d5",
+      "primaryReader": "Release and legal reviewer",
+      "readerDecision": "Replace or isolate imgopt and ship the required termui notice before release.",
+      "prompt": "Create an offline dependency-license findings page from the supplied synthetic inventory. Prioritize copyleft issues, provide sortable detail and provenance, and do not generalize beyond the listed packages.",
+      "requiredFacts": ["Twelve production dependencies are listed.", "vendor/imgopt is GPL-3.0 and linked into the CLI.", "vendor/termui is LGPL-2.1 and requires a notice file."],
+      "forbiddenClaims": ["Legal counsel approved the distribution.", "All transitive dependencies were audited beyond the supplied inventory."],
+      "sourceBundle": { "license": "CC0-1.0", "provenance": "Synthetic dependency inventory", "missingValues": ["No legal-review ticket is supplied."], "longLabel": "Redistribution disposition for dynamically linked native terminal user-interface bindings" },
+      "interactions": [{ "id": "sort-license-table", "action": "Sort the table by package and numeric size using keyboard controls.", "expectedState": "Sort direction is announced and all twelve rows remain available." }],
+      "stressCases": ["long-label", "missing-value", "dense-data", "narrow-viewport"]
+    },
+    {
+      "id": "interactive-decision",
+      "task": "Tune values and return a decision",
+      "fixture": "examples/patterns/tune-controls.md",
+      "fixtureSha256": "7646b761c15b76b9e249aa9827db07f557ec574030df1db3502bc73f4996063c",
+      "primaryReader": "Developer tuning a generated transition",
+      "readerDecision": "Choose frequency and amplitude, inspect the noisy-series window, then copy the chosen values back to the session.",
+      "prompt": "Build an offline tuning page from the supplied ranges and series. Controls must be discoverable, keyboard operable, immediate, reversible, and include a copy-back path; do not add network-backed data.",
+      "requiredFacts": ["Frequency ranges from 0.5 to 8 in 0.5 steps.", "Amplitude ranges from 0.2 to 3 in 0.2 steps.", "The twelve-week series rises overall from 12 to 40."],
+      "forbiddenClaims": ["The selected values are persisted remotely.", "The noisy series is live data."],
+      "sourceBundle": { "license": "CC0-1.0", "provenance": "Synthetic tuning inputs", "missingValues": ["No preferred final frequency or amplitude is supplied."], "longLabel": "Amplitude applied to the generated sine-wave transition preview before copying values back" },
+      "interactions": [{ "id": "tune-and-copy", "action": "Change both sliders by keyboard, reset to supplied defaults, and activate copy.", "expectedState": "The chart updates immediately and copy confirmation names the return path." }],
+      "stressCases": ["long-label", "missing-value", "dense-data", "narrow-viewport"]
+    }
+  ]
+}

docs/component-spec.md

@@ -87,10 +87,15 @@ Component JSON schemas have exactly one home: `skills/artifact-pages/reference/c
 rationale; never copy schema tables into it.
 
 Available fences: `stats`, `timeline`, `findings`, `compare`, `callout`, `progress`, `diff`,
-`copy`, `mermaid`, `decisions`, `table`; chart fences `vega-lite` / `vega` / `echarts`;
+`copy`, `mermaid`, `decisions`, `table`, `frame`; chart fences `vega-lite` / `vega` / `echarts`;
 interactive controls via vega-lite `params.bind` and echarts `dataZoom` (verified live in
 browser QA); free-form interactivity stays in raw-HTML mode (`format: "html"`).
 
+Frontmatter `composition:` selects one bounded layout grammar: `standard`, `narrative`,
+`dashboard`, `split`, `dense`, `quiet`, or `full`. These modes change visual scale, section
+grouping, and density while retaining source/DOM order. Invalid values refuse publication.
+At narrow widths every multi-column mode becomes the same logical single-column sequence.
+
 ### Data honesty rules (from Claude Code's dashboard/dataviz skills)
 
 - Format numbers for scanning: unit + 2–3 significant figures, thousands separators; at most

docs/evidence/journeys/goal-5-status-2026-08-18.md

@@ -0,0 +1,19 @@
+# Goal 5 first-use and comprehension status — 2026-08-18
+
+- Study owner: **unassigned; recruitment is not authorized**.
+- Candidate: commit `84f866ed742eeb08ba668115342ef0e5896dead7`, package `0.15.0`, SHA-256
+  `f6d5d5dcdd74ac31522ec9a98acfb38a04c1a4a038c09e65d85d7f4813342bf2`.
+- Corpus: journey corpus version 1 and participant materials version 1 are frozen.
+- Target install-capable policy cells: Ubuntu LTS desktop, current macOS desktop, previous macOS
+  desktop, Windows 11 native desktop, and Windows 11 WSL desktop. Exact release-time versions
+  remain unresolved. Android Chrome and iOS Safari are viewer/support cells, not README install
+  tasks, and remain unresolved in the certification support matrix.
+- Consented representative primary participants: 0.
+- Eligible README-only first-use runs: 0; every target install cell is missing.
+- Eligible comprehension participants: 0 of the minimum 10.
+- Withdrawals, failures, and exclusions: 0 because collection has not started.
+- Raw participant evidence: none collected.
+
+The checked-in validator, redacted aggregate, and boundary tests are ready. This record is
+`incomplete`, not a participant failure and not a pass. It blocks `OUT-02`, `OUT-03`, supported
+platform, equal-or-better, and Local artifact core certification claims.

docs/evidence/journeys/goal-5-status-2026-08-19.md

@@ -0,0 +1,16 @@
+# Goal 5 first-use and comprehension status — 2026-08-19
+
+- Study owner: **unassigned; recruitment is not authorized**.
+- Candidate: commit `2908d52ef94bb0f8cf74d7f2be021ada78a0d661`, package `0.15.0`,
+  SHA-256 `6d5d4df63bb2300f438a572fc0af4741b793489bbd630b55070c04987c67badd`.
+- Corpus: journey corpus version 1 and participant materials version 1 are frozen.
+- Consented representative primary participants: 0.
+- Eligible README-only first-use runs: 0; every target install cell is missing.
+- Eligible comprehension participants: 0 of the minimum 10.
+- Withdrawals, failures, and exclusions: 0 because collection has not started.
+- Raw participant evidence: none collected.
+
+Automated exact-tarball technical observations now pass on one Ubuntu runner, one macOS
+runner, and one Windows Server runner. They are not participants and do not cover Windows 11,
+WSL, previous macOS, browsers, or mobile devices. This record remains `incomplete`, blocks
+`OUT-02`/`OUT-03`, and cannot support usability, support, comparison, or certification claims.

docs/evidence/page-quality/2026-08-18-benchmark-status.md

@@ -0,0 +1,19 @@
+# Goal 5 comparative benchmark status — 2026-08-18
+
+Status: **incomplete; equal-or-better is unverified**
+
+- Corpus: `page-quality-v1`, eight hash-bound permission-safe tasks.
+- Manifest SHA-256: `2badb79c0f21649b8ed6cb72a25fd8ca2b0a9436faf29d7ce8296384cf5a8897`.
+- Required unselected end-to-end generations: 48 (three per system/task); collected: 0.
+- Required randomized pairs: 24; scored: 0.
+- Required eligible independent reviewers: at least 3 including design/UX and technical;
+  recruited: 0.
+- Claude execution authority, comparable model/settings protocol, and retention disposition:
+  absent.
+- Renderer-only browser cells: 16/16 pass in the separately scoped
+  [`local composition report`](2026-08-18-local-composition.md).
+
+`npm run quality:benchmark -- benchmarks/page-quality/v1/benchmark.template.json` validates
+the frozen input shape and returns `incomplete`. Synthetic unit records exercise pass/fail,
+no-cherry-pick, blinding mappings, confidence, all rubric medians, and hard-gate boundaries;
+they are diagnostics only and are never benchmark evidence.