# Exact source, local request and unverified effect

September 7, 2026 UTC. Source pin: `1ceae52c7896a59ac74a4429d410f3218f0b91c6`;
implementation base: `a30f4eaeea7d52de95b9b0375f82bc4944fbbbe3`. This is a
composition-pilot increment, not an accepted benchmark or completed library. The
full objective remains in [TODOS.md](../../TODOS.md).

## Meaning and implementation

The existing `pwa-install-update` slide now shows exact lines 147–152 of
`applyWaitingServiceWorker`. Its absent-worker guard returns false; a waiting
worker receives `OSHUN_SKIP_WAITING`, followed by an immediate true return.
Posting exceptions escape, and repeating a call can post again. None of these
local results proves worker activation, refreshed content or a server effect.
The signature and final function brace are explicitly outside the excerpt.

The earlier Install → Cache → Update → Clear arrows implied a lifecycle the
source did not establish. The new visual makes a narrower mechanism concrete,
while preserving the installation/native-release, reading-context, queue and
privacy explanations in the notes. This is a changed teaching focus, not merely
new geometry for identical visible copy. Stable IDs, the eleven-slide guide
order, source pin and four citations remain unchanged.

The frontend skill informed the artifact-led hierarchy, native code/annotation
bands, sparse rules and single returned-result emphasis. Code and interpretation
remain separate, with source line ranges and accessible bindings. Builder
validation compares the specimen with the pinned Git bytes; annotations must
partition its complete source range. One annotation is valid when appropriate;
no fixed count or artificial directed graph is required.

Actual review led to two further corrections. Compact playback inherited a 17px
symbol style; a scoped 22px rule restores its intended role. At 320px,
identifiers broke into short trailing fragments. Native `wbr` opportunities at
identifier/path boundaries improve wrapping without inserting characters; source
ranges stay together. Code and annotation bodies remain 18px, headings 20px and
source labels 16px. These are provisional engineering floors, not
reader-validated typography thresholds. Earlier repeated location rows and
redundant prose were removed before tightening compact spacing.

[The artifact brief](../../authoring/artifact-pilot.md),
[complete-guide briefs](../../authoring/v1-web-pwa-guide.md) and
[neutral reader protocol](../../authoring/reader-review.md) retain the intended
meaning and unfinished acceptance work. No screenshot, live worker run,
willing-reader session or normal-speed listening outcome is claimed.

## Verification scope

The full presentation [Python](python-final.log) and [Node](node-final.log)
suites pass **117 and 147 tests**. Eight new Python cases cover exact pinned
bytes, provenance, invalid ranges, variable annotation count, escaping,
source-preserving wrapping, accessible bindings, speech separation, PDF order
and retained installation/privacy context. Three new Node cases bind this
specimen and exercise posting exceptions and repeated requests. Existing
controlled helper tests cover absent/waiting workers and queue acknowledgement.
They execute the pinned helper with controlled receiver dependencies, not a
registered browser worker or deployed product.

The final [composition matrix](composition.json) passes **77 captures**,
including thirteen source-artifact contexts. Eleven negative controls must fail
and pass after restoration: four existing ownership controls plus hidden code,
vertical clipping, undersized symbol, clipped annotation, altered source return,
wrong visible line number and wrong source pin. Exact code characters, source
line identities, symbol/path/pin/label text, annotation order, actual text
bounds and role-specific sizes are checked after reflow. CSS text enlargement
and forced fallback are labeled stress tests, not a browser-zoom or localization
acceptance claim.

The complete eleven-slide [web/PWA standalone suite](web-browser.json) passes
all twelve groups: desktop/compact/mobile, accessibility, source/transcript
dialogs, navigation/history, print/PDF and actual narration controls, seeking,
speed, endings, missing/corrupt audio and playback races. It includes four
source-artifact readability reports: desktop, compact, mobile and print.

The [complete center run](center-ready-audio.json) passes all fourteen groups:
38 guides, 388 unique slides, 414 instances, 828 embedded captures and all 38
PDFs. The 91 table instances yield 182 desktop/mobile readability reports;
ownership and source artifacts each yield three screen/print reports. Eighteen
ready guides pass 36 desktop/mobile playback boundaries. This is
**draft/ready-audio mode, not the final fully narrated release gate**.

[Fresh audio verification](audio-current.json) decodes and pause-checks all 177
current tracks: **16,103.359 seconds and 2,478 sentence segments**. The
replacement PWA track is 173.872 seconds with 24 segments. The other 176 current
manifest entries and MP3 bytes are unchanged. The backlog remains 211 tracks:
191 stale and 20 missing. Decode/encoded-pause checks are not normal-speed
listening or listener pronunciation acceptance.

[Static checks](static-checks.json) cover the final unit suites, builder/preview
and teaching-report freshness, composition audit, Ruff, scoped ESLint,
formatting and whitespace. The complete teaching gate deliberately rejects the
incomplete library: 26 reviewed, 10,018 structural and 103,300 unassigned units,
with no invalid assignments. No product code, API, persistence or deployed
runtime is modified. Controlled TypeScript execution is not a product-wide
typecheck or a live PWA integration/release gate.

Heavy checks ran serially on the 15 GiB RAM / 31 GiB swap host. The center's
lowest reported available memory was 12.5 GiB; no broad Nx or product build
sweep was run. [State](state.json) binds current inputs and pinned sources;
[integrity](integrity.json) binds the exact receipt files. Raw reports and logs
are copied without formatting. Empty-output static checks retain their recorded
exit status without an empty log file.

## Optical review and limits

[The visual ledger](visual-review.json) records **59 review items**: eight
baseline views, thirteen final source views, 24 changed pilot regression images,
eleven web/PWA desktop slides, one complete-guide contact sheet and two renders
of final PDF page 6. These are model observations at original image detail, not
human findings. Forty other pilot captures retain exact prior bytes and their
scoped observations. The source page was reviewed at 1280px page-fit width and
2000px full-size width. Tall mobile images establish scrollable reading, not
whole-slide fit inside a phone's initial viewport.

The source treatment separates guard, request and return in all inspected
contexts. Thin rules and the emphasized return survive PDF export. The complete
guide still has migration work: Lilith's four-room scope is shown as three
numbered groups, account/design slides repeat a three-column silhouette, release
enforcement has unlabeled edges, and the later worked update remains a linear
five-stage explanation without visible choice branches. The new source example
does not close those findings or the complete typography/reader standard.

All eight baseline and 77 final focused PNGs, and the standalone web/PWA PNGs,
are retained. All 828 center PNG hashes were checked during collection; only the
22 web/PWA initial-viewport PNGs are retained here. The other 806 remain
run-local automated outputs, not durable optical-review evidence. The retained
[pre-wrap narrow](diagnostics/pre-wrap-narrow.png) and
[pre-symbol-fix playback](diagnostics/pre-symbol-fix-compact-playing.png) images
document intermediate findings; their earlier passing diagnostic report did not
include the final role-specific symbol and provenance checks.

## Reproduce

Use the prepared presentation dependencies, fresh output directories and the
repository resource checks. Run expensive commands separately:

```sh
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
CHROME_CHANNEL=chromium OSHUN_COMPOSITION_QA_DIR=/tmp/oshun-source-new python tools/presentations/run-supervised.py node tools/presentations/tests/composition-pilot.mjs
CHROME_CHANNEL=chromium OSHUN_DECK_QA_DIR=/tmp/oshun-web-new python tools/presentations/run-supervised.py node tools/presentations/tests/eve-oshun-deck.mjs --guide v1-web-pwa
CHROME_CHANNEL=chromium OSHUN_CENTER_QA_DIR=/tmp/oshun-center-new python tools/presentations/run-supervised.py node tools/presentations/tests/presentation-center.mjs --draft --ready-audio
node docs/presentations/presentation-center/verification/annotated-source-2026-09-07/verify-evidence.mjs
```

For a fresh decode/pause check of this receipt's current subset, pass its
comma-separated `audio-current.json` → `current[].id` list to
`render-eve-oshun-narration.py --check --ids`. The unrestricted narrator check
remains the final all-track gate after the backlog is rebuilt. The receipt
verifier checks saved evidence and current byte identity; it does not run
browser automation, decode audio or replace human review.
