Retain resolved Auto appearance in comparison observations

docs/IMPLEMENTATION_MASTERPLAN.md

@@ -2,6 +2,8 @@
 
 Last Updated: 2026-09-10
 
+Comparison attribution follow-through (#2808): retain the resolved Auto appearance in the existing portable theme label and exclude known non-shipped test inputs from the frontend fingerprint. Preserve legacy checksums/file semantics and prove live light/night switching, export/import and HTTP LAN compatibility together.
+
 Original-source portability continuation (#2808): hold a store-owned read snapshot across the five source-storage export sections, with deferred SQLite transactions and scoped disposal before later export writes. Prove both account-export routes against a concurrent committed upload and written version. Whole-account point-in-time consistency and restore acceptance remain separate.
 
 Studio/Classic planning continuity (#2808): share existing Focus resume on Classic Home and use the existing calendar-day utilities for plan date controls and card deadlines. Browser acceptance includes a western timezone, Today filtering, both Classic renderers and unchanged board-card data. Failed-plan metadata and accepted-navigation focus timestamps remain separate follow-through.

docs/STATUS.md

@@ -2,6 +2,8 @@
 
 Last Updated: 2026-09-10
 
+Comparison appearance continuation (#2808): Auto observations retain both the selected mode and the resolved light/night appearance at submission. Existing version-2/3 files keep their original attribution; no historical appearance is invented. Frontend input fingerprints exclude repository test-only sources and build proof scripts, while runtime sources, public assets, build plugins and resolved build options remain attributed. This keeps manual comparisons coherent across proof-only edits without claiming randomized or statistical results.
+
 Source-storage export continuation (#2808): buffered and streamed account exports now hold one deferred SQLite read snapshot across blob objects, references, chunks, representations and audio-answer rows. Concurrent WAL uploads can commit without reserving the writer for the duration of the export; all five sections retain the earlier view until disposal. This is source-storage consistency, not a claim of a single snapshot across every account-export section or a tested restore workflow.
 
 Planning continuity follow-through (#2808): Classic Home now offers the same private last-worked Focus resume as the other experiences, in both Paper/Grove and Legacy rendering. Personal-plan due dates use canonical calendar-day formatting; Today and tomorrow use the local calendar date without changing card deadlines. This reduces the work needed to resume a thought after switching versions.

docs/product/WORKSPACE_OVERHAUL_VALIDATION.md

@@ -369,3 +369,10 @@ PR #2866 now also contains the reviewed source interaction, board overlay, libra
 Sixteen selected Chromium journeys have passing evidence. The initial combined batch passed fifteen and found one ambiguous memory-title locator; using the checkbox role fixed that test, and both complete contextual/source specs then passed. Coverage includes four experiences, both board renderers, Los Angeles calendar dates, Classic resume, 375 px accessibility/overflow, source permissions, exact original audio, explicit Review/Apply and three production audio-policy cases. This is not a claim that the initial batch was entirely green.
 
 Two comparison identity checks, two audio-policy checks, eleven failure-ledger projection tests, documentation links and GitHub governance pass. A bounded fresh-context integration review found no HIGH/CRITICAL interaction defect. Temporary browser services are stopped; all databases are synthetic. Hosted exact-head qualification remains the merge gate. The wider #2808 scope, physical microphones/devices, live processors, restoration and the existing human decisions remain open.
+
+## Comparison appearance attribution (2026-09-10)
+
+Auto observations use the existing version-3 theme string to retain both selected mode and resolved appearance at submission: auto (paper) or auto (paper-night). Old auto labels remain unchanged and separate; no version-2 checksum input, file schema or historic identity changes. The UI explains the submission-time scope and asks the observer to note any appearance changes during the task.
+
+The HTTP LAN compatibility Chromium journey records both live color-scheme settings, confirms the body theme, exports those labels alongside a legacy auto observation, checks three separate result rows, reloads and imports all three without SubtleCrypto or randomUUID. It passes in 14.6 seconds. Node fingerprint checks verify that src/tests and build/*.node-check.mjs changes leave product attribution stable, while runtime source, public assets, build plugin code and build configuration change it. Full frontend/build results belong to the continuation PR; there is no telemetry, randomized assignment or statistical conclusion.
+Final local comparison-attribution gate: 6,377 frontend tests passed with three existing skips across 412 files. Production build/typecheck, two Node identity checks, the 14.6-second Chromium journey, documentation links and GitHub governance pass. Independent bounded Luna review is CLEAN. The test-only fingerprint exclusions are exact repository paths; this is still an input fingerprint, not a byte-for-byte bundle hash or proof of the remote server version. Temporary services are stopped; hosted checks remain separate.

frontend/taskdeck-web/build/frontendIdentity.node-check.mjs

@@ -22,12 +22,21 @@ test('frontend fingerprint is repeatable and changes with source, public assets
   const baseline = fingerprintFrontend(root, 'production', { VITE_API_BASE_URL: '/api' })
   assert.match(baseline, /^sha256:[a-f0-9]{64}$/)
   assert.equal(fingerprintFrontend(root, 'production', { VITE_API_BASE_URL: '/api' }), baseline)
+  mkdirSync(join(root, 'src/tests'))
+  writeFileSync(join(root, 'src/tests/new.spec.ts'), 'new proof, same product')
+  writeFileSync(join(root, 'build/example.node-check.mjs'), 'build proof')
+  assert.equal(fingerprintFrontend(root, 'production', { VITE_API_BASE_URL: '/api' }), baseline)
+  writeFileSync(join(root, 'src/tests/new.spec.ts'), 'updated proof')
+  assert.equal(fingerprintFrontend(root, 'production', { VITE_API_BASE_URL: '/api' }), baseline)
   assert.notEqual(fingerprintFrontend(root, 'production', { VITE_API_BASE_URL: '/other-api' }), baseline)
   assert.notEqual(fingerprintFrontend(root, 'production', { VITE_API_BASE_URL: '/api' }, { base: '/Taskdeck/' }), baseline)
   assert.notEqual(fingerprintFrontend(root, 'production', { VITE_API_BASE_URL: '/api' }, { minify: false }), baseline)
   writeFileSync(join(root, 'src/main.ts'), 'changed content')
   const changedSource = fingerprintFrontend(root, 'production', { VITE_API_BASE_URL: '/api' })
   assert.notEqual(changedSource, baseline)
   writeFileSync(join(root, 'public/theme.css'), 'changed theme')
-  assert.notEqual(fingerprintFrontend(root, 'production', { VITE_API_BASE_URL: '/api' }), changedSource)
+  const changedAssets = fingerprintFrontend(root, 'production', { VITE_API_BASE_URL: '/api' })
+  assert.notEqual(changedAssets, changedSource)
+  writeFileSync(join(root, 'build/runtime-plugin.ts'), 'plugin affecting output')
+  assert.notEqual(fingerprintFrontend(root, 'production', { VITE_API_BASE_URL: '/api' }), changedAssets)
 })

frontend/taskdeck-web/build/frontendIdentity.ts

@@ -13,6 +13,8 @@ export function fingerprintFrontend(root: string, mode: string, publicEnvironmen
   function directory(relative: string) {
     for (const entry of readdirSync(join(root, relative), { withFileTypes: true }).sort((a, b) => a.name.localeCompare(b.name, 'en'))) {
       const name = `${relative}/${entry.name}`
+      // These repository-owned proof inputs are not shipped in the frontend bundle.
+      if (name === 'src/tests' || (relative === 'build' && entry.isFile() && entry.name.endsWith('.node-check.mjs'))) continue
       if (entry.isDirectory()) directory(name)
       else if (entry.isFile()) add(name, readFileSync(join(root, name)))
       else throw new Error(`Unsupported frontend input: ${name}`)

frontend/taskdeck-web/src/views/overhaul/WorkspaceExperiencesView.vue

@@ -1,7 +1,7 @@
 <script setup lang="ts">
 import { onMounted, ref } from 'vue'
 import { useWorkspaceLayoutStore } from '../../store/workspaceLayoutStore'
-import { usePaperThemeStore } from '../../store/paperThemeStore'
+import { resolveBodyClass, usePaperThemeStore } from '../../store/paperThemeStore'
 import {
   useWorkspaceExperimentStore,
   WORKSPACE_COMPARISON_SCENARIOS,
@@ -38,7 +38,7 @@ function record() {
   saved.value = experiment.record({
     experience: layout.experience,
     presentation: layout.presentation,
-    theme: theme.mode,
+    theme: theme.mode === 'auto' ? `auto (${resolveBodyClass('auto')})` : theme.mode,
     build: version.value,
     scenario: scenario.value,
     completionOutcome: completionOutcome.value,
@@ -77,6 +77,7 @@ function download() {
     <form class="experience-lab__notes" @submit.prevent="record">
       <h2>Keep an observation</h2><p>Recording {{ layout.experience }} / {{ layout.presentation }} / {{ theme.mode }} · backend version {{ displayVersion || 'unavailable' }}. Notes stay in this session; export them before reloading or signing out. Import the file later to compare releases.</p>
       <details><summary>Frontend build attribution</summary><p>{{ frontendBuildIdentity || 'Development or unstamped build: frontend attribution unavailable.' }}</p></details>
+      <p v-if="theme.mode === 'auto'">Auto records the light or night appearance when you submit. If it changed during your task, mention that in your note.</p>
       <label for="experience-scenario">What scenario did you try?</label><select id="experience-scenario" v-model="scenario" required><option :value="null" disabled>Select a scenario</option><option v-for="item in WORKSPACE_COMPARISON_SCENARIOS" :key="item.id" :value="item.id">{{ item.label }}</option></select>
       <label for="experience-outcome">What happened?</label><select id="experience-outcome" v-model="completionOutcome" required><option :value="null" disabled>Select an outcome</option><option v-for="item in WORKSPACE_COMPLETION_OUTCOMES" :key="item.id" :value="item.id">{{ item.label }}</option></select>
       <label for="experience-ease">How easy was it to continue your work? (optional)</label><select id="experience-ease" v-model="ease"><option :value="null">Leave unrated</option><option :value="1">1 — Difficult</option><option :value="2">2 — Some friction</option><option :value="3">3 — Reasonable</option><option :value="4">4 — Easy</option><option :value="5">5 — Effortless</option></select>

frontend/taskdeck-web/tests/e2e/comparison-compatibility.spec.ts

@@ -6,14 +6,19 @@ test('comparison files work with the crypto APIs available on HTTP LAN', async (
   await page.addInitScript(() => {
     Object.defineProperty(globalThis.crypto, 'randomUUID', { value: undefined, configurable: true })
     Object.defineProperty(globalThis.crypto, 'subtle', { value: undefined, configurable: true })
+    localStorage.setItem('td.paper.mode.v2', 'auto')
   })
   await page.goto('/workspace/experiences')
-  await page.getByLabel('What scenario did you try?').selectOption('resume-thinking')
-  await page.getByLabel('What happened?').selectOption('completed')
-  await page.getByRole('button', { name: 'Record observation', exact: true }).click()
-  await expect(page.getByText('Observation recorded.', { exact: true })).toBeVisible()
+  for (const colorScheme of ['light', 'dark'] as const) {
+    await page.emulateMedia({ colorScheme })
+    await expect(page.locator('body')).toHaveClass(colorScheme === 'light' ? /\bpaper\b/ : /\bpaper-night\b/)
+    await page.getByLabel('What scenario did you try?').selectOption('resume-thinking')
+    await page.getByLabel('What happened?').selectOption('completed')
+    await page.getByRole('button', { name: 'Record observation', exact: true }).click()
+    await expect(page.getByText('Observation recorded.', { exact: true })).toBeVisible()
+  }
   const legacy = JSON.stringify({ kind: 'taskdeck-workspace-comparison', version: 2, trials: [{
-    experience: 'classic', presentation: 'studio', theme: 'grove', build: null, scenario: 'resume-thinking',
+    experience: 'classic', presentation: 'studio', theme: 'auto', build: null, scenario: 'resume-thinking',
     completionOutcome: 'completed', recordedAt: '2026-09-01T00:00:00Z', ease: null, note: 'Retained LAN observation',
   }] })
   const input = page.getByLabel('Import saved observations', { exact: true })
@@ -25,8 +30,9 @@ test('comparison files work with the crypto APIs available on HTTP LAN', async (
   await page.getByRole('button', { name: 'Export observations', exact: true }).click()
   const download = await downloading; const stream = await download.createReadStream()
   let retained = ''; for await (const chunk of stream!) retained += chunk.toString()
-  expect(JSON.parse(retained).trials).toHaveLength(2)
+  expect(JSON.parse(retained).trials.map((trial: { theme: string }) => trial.theme)).toEqual(['auto (paper)', 'auto (paper-night)', 'auto'])
+  await expect(page.getByRole('region', { name: 'Compare observations across releases' }).locator('tbody tr')).toHaveCount(3)
   await page.reload()
   await input.setInputFiles({ name: 'retained.json', mimeType: 'application/json', buffer: Buffer.from(retained) })
-  await expect(page.getByText('Imported 2 observations. Exact duplicates were skipped.', { exact: true })).toBeVisible()
+  await expect(page.getByText('Imported 3 observations. Exact duplicates were skipped.', { exact: true })).toBeVisible()
 })