The two prose labels on v2-surfaces-companion-contract now read Auth
contract and Mobile app. The full OAuth value, planned-iOS/Android
qualification, notes, speech, source citations and guide order are unchanged.
The authoring review records the pinned
source recheck and refreshed teaching fingerprint. No additional teaching unit
is claimed complete.
Verified scope#
- All 180 Python and 268 Node presentation tests pass. Historical
whole-library Agentic preservation assertions read immutable
00b6863ee7against7167e08ce5; current Agentic content/order and the V2 speech/content assertions remain live. - The focused word audit covers 11 captures: desktop, compact playing, 390px and 320px reading views, 125% row-text enlargement, table ends, both embedded sizes and a scrolled embedded-phone label view. It checks actual word line geometry and the unchanged 16px floor. Five rejected/restored injections cover both original labels, forced splitting, hidden text and reduced text size. Exact API identifiers elsewhere are not subject to this scoped prose-word rule.
- The complete standalone guide passes all twelve slides: desktop/compact/mobile, accessibility, source/transcript/dialog/navigation, PDF sequence and actual audio including HTTP failures and playback races.
- The focused center run passes twelve slides in both embedded sizes (24 captures), both actual guide audio boundaries, navigation, duration/performance checks and the twelve-page PDF. All 37 unselected guide HTML files match the previous full baseline. This is not a new full 391-slide bank or 38-guide embedded run.
- The existing table regression passes 34 cases and its broken-sticky-identity control, including numeric/missing values, narrow columns and synthetic identifiers. This checks the shared audit's unchanged behavior outside the new prose-label scope.
- All 391 recordings pass fresh fingerprint, byte/hash, duration, decode and encoded-pause verification. The manifest, speech script and every MP3 retain their baseline bytes; unique duration remains 34,841.439 seconds. No recording was regenerated and no normal-speed listening is claimed.
- Ruff, scoped ESLint/Prettier, builder/preview freshness, repository/workflow inventory freshness and teaching-assignment freshness pass. Teaching remains 26 reviewed / 103,300 unassigned documentation units and 3,718 unassigned workflow units. These incomplete totals are not release passes.
Visual review and limitations#
The state record binds thirteen retained images to the tested HTML/PDF, plus the actual input/evidence/audio hashes. Model review covers the eleven focused captures and the affected PDF page at 1440px page-fit and 2880px full raster sizes. Start/end pairs with no horizontal overflow have identical bytes; they are not independent optical observations. The eleven unaffected PDF pages have exact raster equality with the prior published review images. Their existing legacy-numbering and small-diagram-label findings remain open.
The 320px API route values retain exact characters but still wrap long technical identifiers; the scoped repair does not claim a complete code-label treatment. Tall standalone captures show full reading content, not first-viewport phone fit. The scrolled embedded image proves that both repaired labels can be read inside the actual phone frame. Row-only CSS enlargement is explicitly not a full-page browser-zoom or font-fallback test.
The initial logs/label.log failure was a new capture-harness mistake: expanded
height was requested on desktop. logs/label-v2.log then found the real enlarged
“application” split. Both are retained as failed attempts, not passing evidence.
The final label log, reports and five controlled failures cover the corrected
implementation.
The previous Agentic receipt and canonical full-bank/center reports are left
unchanged at 00b6863ee73ec25f5c4680e1ae380c4389c18149. Their historical input
hashes are not rebound to this change. This receipt proves a narrow typography
repair, not whole-guide acceptance, full composition migration, reader
comprehension, listening or the broader TODO objective.
Reproduction#
Use the prepared presentation dependencies and repository resource checks.
Run expensive commands serially with tools/presentations/run-supervised.py.
Choose fresh output directories for OSHUN_LABEL_QA_DIR, OSHUN_DECK_QA_DIR,
OSHUN_CENTER_QA_DIR and OSHUN_TABLE_QA_DIR; Linux uses
CHROME_CHANNEL=chromium.
node tools/presentations/tests/v2-companion-label.mjs
node tools/presentations/tests/eve-oshun-deck.mjs --guide v2-product-surfaces
node tools/presentations/tests/presentation-center.mjs --guides v2-product-surfaces
node tools/presentations/tests/table-pilot.mjs
python -m unittest discover -s tools/presentations/tests -p 'test_*.py'
node --test --test-concurrency=1 tools/presentations/tests/*.test.mjs
python tools/presentations/render-eve-oshun-narration.py --check
python tools/presentations/build-eve-oshun.py --check --preview
python tools/presentations/check-center-coverage.py --check
Next: complete source/teaching review and integrate the saved V2 contracts guide. The V1 Zod-versus-V2 Unreal contract distinction still needs explicit teaching; do not mark its entire architecture unit covered by the existing draft.