The full continuation checklist remains active. This increment
does not complete global narration, whole-monorepo teaching, live Eve
integration, Domains, UI/UX coverage or the final editorial/listening review.
All source claims retain revision 1ceae52c7896a59ac74a4429d410f3218f0b91c6.
Current recordings and playback#
Four sequential, one-thread CPU batches rendered 20 previously pending tracks and replaced three recordings after substantive explanation corrections. All four processes exited normally with code 0. The stock voice, model, pronunciation and timing policy are unchanged. The render logs retain each completed track and resource sample: batch 1, batch 2, batch 3, batch 4.
The current-audio check verifies 116 recordings, 9,332.714 seconds and 1,460 sentence segments using actual fingerprints, file hashes/bytes, decoded duration and encoded pauses. 265 recordings remain pending: 245 stale and 20 missing. This is not a full-inventory pass. The decode recheck log records a normal exit for all 116 selected tracks against the final manifest.
Orientation and all six Eve guides now have complete audio. Their full standalone browser suites passed normally, covering 74 slides in total:
| Guide | Slides | Report |
|---|---|---|
| Orientation | 10 | Result |
| Foundations | 11 | Result |
| Execution | 10 | Result |
| Context | 9 | Result |
| Trust | 10 | Result |
| Quality | 11 | Result |
| Operations | 13 | Result |
Each report has an adjacent -browser.txt log. These non-draft runs cover
desktop/compact/mobile layout, axe accessibility, source/script alignment,
navigation, dialogs, PDF sequence, actual playback/pause/resume/seek/speed/end
transitions, HTTP recovery, missing/corrupt audio, playback races and portable
local-file behavior. No narrator was running during browser checks.
The shared status slide also changed the release-acceptance guide. Its 12-slide draft suite passed layout, accessibility, source/script, navigation and PDF checks. That guide's remaining audio is incomplete; its draft run does not claim actual-audio coverage.
Mechanical audio checks and accelerated browser interaction are not a normal-speed listening review. No live Eve tool, provider or product-runtime acceptance is certified here.
Center verification and capture integrity#
The center report and log record a normal exit with all 38 guides, 407 guide-slide instances and 38 PDF exports checked. Navigation, hierarchy, source/prerequisite links, search, history, local interaction budgets, measured durations, drawer/focus behavior, layout and accessibility pass.
Actual embedded start/end/replay/stop boundaries pass on desktop and mobile for
the ten fully narrated guides: portfolio map, orientation, the six Eve guides,
Nyx and Nisaba. The report contains 20 viewport/guide results and explicitly
lists the 28 guides skipped for audio. --draft --ready-audio remains an
incomplete-library mode, not the final non-draft release baseline.
The center now saves every embedded desktop/mobile initial viewport instead of three sample slides per guide. Its report binds all 814 captures to their ordered guide/slide identities and actual persisted image hashes. Seven new capture regressions reject missing, duplicated, reordered or misidentified captures; scope/path substitution; invalid hashes; changed image bytes; and missing files. Together with the six audio-scope cases, all 13 Node tests pass.
Capture availability does not establish visual review. Embedded mobile initial-viewport images can end before the slide does; full-height standalone mobile, table-end and PDF images retain the additional content for inspection. The complete reproducible capture set remains in the run's temporary QA folder; only the images actually inspected for the review below are copied here.
Teaching corrections and targeted visual review#
The opening learning map records complete source-page reading and the five new hash-bound assignments:
- Status vocabulary now separates four definitions instead of combining built but unwired with proposed work, and teaches every gate and dated completion boundary.
- Task-family routing now explicitly teaches deterministic high-confidence signals, model judgment for ambiguity, escalation and neighboring-intent tests.
- Four context layers now teach every purpose, lifetime and user-control expectation, with a worked non-substitution example.
- The unchanged confirmation-record and client-action explanations already cover their full reviewed source sections, including the complete field and outcome inventories.
The frontend skill guided the plain four-column status comparison and retention of the existing restrained table/flow language. The targeted visual ledger binds 37 inspected images, five unique slides in six guide contexts, canonical slide/audio hashes and current guide HTML/PDF hashes. It covers standalone desktop, compact, full-height mobile, the context table's horizontal end, embedded initial viewports and the six corresponding pages from the current exported PDFs. Explanatory notes and useful questions were reviewed as well. No material defect remained in those inspected states. This is not an every-slide or full-guide normal-speed quality pass.
All 54 Python tests, including actual Misaki and MP3 checks, pass. Builder/coverage freshness, Ruff, scoped ESLint, Prettier and diff checks pass in the final static recheck. The complete teaching gate correctly rejects completion: seven reviewed units and 103,319 unassigned meaningful units remain the actual accounting, with no invalid or stale assignments.
The artifact receipt validates the saved reports against the current HTML, PDFs, narration manifest and reviewed images, and binds the relevant authoring, harness, test and evidence files. New main-branch changes must be integrated without changing the immutable source pin or silently reusing stale artifact evidence.
Reproduction#
Use the pinned Python environment, installed Chromium and repository resource checks. Run expensive targets one at a time with distinct QA output folders.
python -m unittest discover -s tools/presentations/tests -p 'test_*.py'
node --test tools/presentations/tests/center-*.test.mjs
python tools/presentations/build-eve-oshun.py --check --preview
python tools/presentations/check-center-coverage.py --check
CHROME_CHANNEL=chromium python tools/presentations/run-supervised.py node tools/presentations/tests/eve-oshun-deck.mjs --guide orientation
CHROME_CHANNEL=chromium python tools/presentations/run-supervised.py node tools/presentations/tests/eve-oshun-deck.mjs --guide release-acceptance --draft
CHROME_CHANNEL=chromium python tools/presentations/run-supervised.py node tools/presentations/tests/presentation-center.mjs --draft --ready-audio
Repeat the standalone command for each Eve guide in the table. For this partial
audio checkpoint, pass the report's comma-separated currentIds to
render-eve-oshun-narration.py --check --ids; after all pending tracks are
finished, the unfiltered full-inventory check and full narrated release suites
remain mandatory.