# Nisaba narration and explicit partial-audio verification — 2026-09-06

The [full continuation checklist](../../TODOS.md) remains active. This increment
does not complete global narration, source teaching, live Eve integration,
Domains, or final editorial/listening review.

## Current recordings

Nisaba's seven remaining recordings were regenerated in one bounded, one-thread
CPU process that exited normally with code 0. The model, voice, pronunciation
and timing policy are unchanged. No narrator remained during browser checks. The
[render log](narration.txt) records the completed tracks and resource samples.

The [current-audio check](current-audio-check.json) verifies **96 recordings,
7,335.062 seconds and 1,147 sentence segments** against the actual manifest,
fingerprints, file hashes/bytes, decoded duration and encoded pauses. There are
still **285 pending recordings: 265 stale and 20 missing**. This deliberately
does not claim a passing full-inventory gate.

All nine slides in the current Nisaba guide now have current audio. The
[standalone report](nisaba-browser.json) and [log](nisaba-browser.txt) cover its
desktop/compact/mobile layout, accessibility, source/script alignment,
navigation, dialogs, print/PDF sequence and actual playback, pause/resume,
seeking, speed, end transitions, HTTP recovery and playback races. The reused
`room-maturity` recording also becomes current in the V1 web/PWA and Arete
guides; their other pending recordings remain visibly incomplete.

Mechanical audio checks and accelerated browser interaction are not a human
normal-speed listening pass. No live Nisaba provider, production reading service
or scholarly notebook deployment is certified by these tests.

## More useful partial browser evidence

The center harness now accepts `--draft --ready-audio`. It retains all-guide
layout and PDF verification while checking embedded audio for every guide whose
catalog narration is complete. At this checkpoint those are `portfolio-map`,
`nyx` and `nisaba`. Start, end, replay and stop are exercised at desktop and
mobile sizes, alongside drawer pausing and topic-switch cancellation.

The [center report](center-ready-audio.json) and [log](center-ready-audio.txt)
retain the exact scope: 38 guides and all 407 guide-slide instances for
layout/accessibility, all 38 PDF exports, and only those three guides for audio.
The report explicitly lists 35 skipped audio guides. It remains `draft: true`
and cannot become the full narrated release baseline.

Six [scope-selection regressions](audio-scope-tests.txt) guard the distinction:
ordinary draft runs skip audio; release mode never silently drops incomplete
guides; partial mode lists both included and excluded guides; it requires draft
mode and a full, non-smoke, non-focused layout run; and it rejects an empty
eligible set or truthy non-boolean readiness. The browser also asserts that
every declared audio guide produced its desktop and mobile end/replay result.

## Source-review findings

The [Nisaba learning map](../../drafts/nisaba-learning-map.md) follows a
complete reading of the 621-line pinned feature page, its 29 meaningful units
and the current nine-slide orientation. It records required contracts,
algorithms, language/criticism mechanisms, routes and worked examples without
treating an overview citation as complete unit coverage.

Pinned Git inspection finds 23 direct Nisaba package manifests, not the page's
22; its list omits `@nisaba/study-adapter`. The map distinguishes that adapter's
injected-store notebook operations from its separate unconditional
`persistNotebook()` refusal, and separates a tokenization module's existence
from a root-package re-export. Three new static snapshot regressions preserve
these boundaries. They do not execute a service or freeze current product code
to historical limitations.

All [54 Python tests](unit-tests.txt), including actual Misaki and MP3 checks,
pass. Ruff, scoped ESLint, Prettier and builder/report freshness are checked as
well. Teaching coverage remains **two reviewed units and 103,324 unassigned
meaningful units**. This map creates no teaching assignments or placeholder
catalog entries.

The [artifact receipt](checks.json) confirms the reports' scope and current
HTML, PDF and manifest hashes, and binds the tested harness, source map and
test/report files. The standalone and center supervising sessions both exited
normally with code 0. The [complete teaching gate](teaching-incomplete.txt)
still rejects a full-coverage claim, as required while those units are pending.

## Reproduction

Use the pinned Python environment, installed Chromium, resource admission checks
and one expensive gate at a time:

```sh
python -m unittest discover -s tools/presentations/tests -p 'test_*.py'
node --test tools/presentations/tests/center-audio-scope.test.mjs
python tools/presentations/build-eve-oshun.py --check --preview
CHROME_CHANNEL=chromium python tools/presentations/run-supervised.py node tools/presentations/tests/eve-oshun-deck.mjs --guide nisaba
CHROME_CHANNEL=chromium python tools/presentations/run-supervised.py node tools/presentations/tests/presentation-center.mjs --draft --ready-audio
```

Use separate `OSHUN_DECK_QA_DIR` and `OSHUN_CENTER_QA_DIR` folders to retain
each run. Full narration and normal-speed listening remain required, followed by
the non-draft full-library gates and every remaining checklist requirement.
