The full checklist remains active. This is a partial narration
and teaching increment, not complete Sophia/whole-monorepo coverage, live Eve
integration, Domains acceptance or final visual/listening review. Source claims
retain revision 1ceae52c7896a59ac74a4429d410f3218f0b91c6.
Current audio and standalone playback#
Two sequential one-thread CPU batches rendered eight previously pending tracks; both exited normally with code 0. The stock Kokoro voice, model revision, pronunciation and timing policy are unchanged. Logs: first batch, second batch.
The audio audit verifies 153 current tracks, 12,727.069 seconds and 1,969 sentence segments against speech fingerprints, hashes/bytes, decoded duration and encoded pauses. 228 tracks remain pending: 208 stale and 20 missing. Sophia now has complete audio. The shared substrate track also updates Iris's available duration without completing Iris. These mechanical checks do not establish normal-speed listening quality.
The full Sophia standalone suite exited 0 and passed all eight slide instances; its log retains the run. Coverage includes desktop/compact/mobile layout, axe accessibility, source/script alignment, navigation, dialogs, print sequence, actual audio decode and play/pause/resume/seek/speed/end transitions, HTTP failures, missing/corrupt audio, retry, delayed-play races and portable local-file behavior. No narrator was running. The report binds the exact delivered HTML.
Center and targeted visual review#
The complete center run exited 0, passing all 14
checks, 38 guides, 407 embedded slide instances and 38 PDF exports. Its
--draft --ready-audio scope covers 32 desktop/mobile audio boundaries for 16
ready guides and explicitly excludes the other 22 guides from audio acceptance.
All 814 initial-viewport captures, 45 bound center artifacts and 38 delivered
PDFs are revalidated by the receipt. Navigation, hierarchy, duration semantics,
search, browser history, drawer/focus handling and the local interaction budget
pass. This is not full-library audio acceptance.
The visual ledger binds 25 inspected images across four Sophia slides to canonical slide, HTML, final PDF and narration hashes. Each was inspected in desktop, compact desktop, full-height mobile, embedded desktop/mobile initial viewports and an actual final PDF-page render. The failure table also has a mobile table-end view. Embedded mobile alone is not a full-scroll review; the full standalone mobile and PDF views supply the lower content. Notes and questions were reviewed for all four slides.
No material defect was found in these inspected states. Final all-slide visual review and normal-speed listening remain open. This targeted ledger is not a claim to have manually inspected all 814 automated captures.
Source fidelity and teaching#
The learning map records the source review, four reviewed assignments and remaining depth. Four explanations were corrected or enriched:
- Extraction uses the summary verbatim or its title when that summary is blank. Jaccard token sets, stable ties and exact rendered-text deduplication preserve all structured claims and can leave inline marker gaps.
- Input citation counts determine default state/abstention and separate route confidence bands; they do not establish unique or independent corroboration.
- Synthesis cleanup checks identifier membership, not semantic support. A claim saying green can survive against a passage saying blue when it cites the real identifier. Label coercion is not factual verification.
- Removing every claim can leave model prose and invented inline markers in an abstained envelope. That successful sanitized result does not automatically trigger extraction. Thrown errors/timeouts do; a timeout does not cancel the underlying provider promise. Configuration also admits the shared runtime, while partial legacy overrides fail closed before shared resolution.
Fourteen new Node regressions execute pinned source with controlled Nisaba reads and provider responses. Three original route functions are extracted via the TypeScript AST without changing their bodies. The real meter wrapper runs with a recording replacement for its library builder; the configuration test controls shared provider/client-option dependencies. These tests do not establish mounted HTTP/authentication, live retrieval, provider, database, confidence-meter library, global display-policy or deployed-service acceptance. No BFF runtime was changed.
The frontend skill guided retention of the extractive/worked flows, plain enhancement columns and failure table. Detailed counterexamples and exact thresholds remain in the explanation, without extra chrome or decorative motion. Repository Playwright, axe and Poppler supply the available verification workflow; no dedicated presentation/PDF skill or audio-listening tool is available.
Automated checks and reproduction#
All 54 Python tests and 51 Node tests pass in the Python log and the Node log. The teaching-completeness gate exits 1 as expected: 15 reviewed units, 103,311 unassigned meaningful units and no invalid or stale assignment errors. This is an incomplete global gate, not complete coverage.
Builder and teaching-inventory freshness, scoped Ruff/ESLint/Prettier and whitespace checks pass in the static log. The receipt binds artifacts, reports, logs, reviewed visuals and relevant tools; all global completion claims remain false.
Use the configured Python environment and browser before reproducing:
python -m unittest discover -s tools/presentations/tests -p 'test_*.py'
node --test tools/presentations/tests/*.test.mjs
python tools/presentations/build-eve-oshun.py --check --preview
python tools/presentations/check-center-coverage.py --check
python tools/presentations/check-center-coverage.py --require-complete
The final command intentionally rejects completeness. Run browser suites
sequentially through run-supervised.py, with no narrator:
node tools/presentations/tests/eve-oshun-deck.mjs --guide sophia, then
node tools/presentations/tests/presentation-center.mjs --draft --ready-audio.
Export center PDFs last so the report binds delivered bytes. Use separate
temporary QA directories, CHROME_CHANNEL=chromium, and the resource checks in
AGENTS.md. The observed environment uses Playwright 1.60.0, Chromium
148.0.7778.96 and axe 4.11.4. No full-workspace build or sweep was needed for
these static presentation/source-test changes.
For the current-audio subset, pass the audit's comma-separated currentIds to
render-eve-oshun-narration.py --check --ids. The unfiltered full-inventory
gate is still required after all pending tracks are finished. Historical
receipts remain checkpoints, not evidence for subsequently changed bytes.