# Eve/V1 library verification

**Historical completed-phase record.** The 2026-09-06 machine-transfer
checkpoint adds content and changes speech timing across the library. The
reports below do not certify those later changes. See
[the handoff state](verification/handoff-2026-09-06/state.json) and
[remaining verification tasks](TODOS.md) for current scope and gaps.

The delivered portal contains 35 guides, 349 unique slides, 349 narration tracks
and 35 PDFs containing 374 pages. Repeated context accounts for the difference
between unique slides and guide pages. All 244 foundation slides are retained,
with 105 added explanations and worked cases. The reviewed source revision is
`1ceae52c7896a59ac74a4429d410f3218f0b91c6`.

## Current portal and exports

The [full presentation-center report](verification/presentation-center.json)
records a completed run with no runtime or accessibility violations:

- Every guide's sequence, source context, prerequisites and coverage mappings.
- Every focused slide inside desktop and mobile portal frames, including layout,
  accessibility and the initial horizontal position of tables.
- Search, empty results, keyboard focus, sidebar persistence, mobile drawer,
  deep links, browser history, guide details and sequential navigation.
- Actual first- and last-track playback for every guide on both viewports: 70
  guide-ending, replay and stop scenarios. Opening the drawer and changing
  topics correctly pause playback.
- Every guide PDF's page count, title sequence and numbered footers, with
  artifact hashes recorded after export.

Guide selection measured 105 ms at the 95th percentile and 176 ms maximum,
within the 2-second and 5-second local interaction budgets. The tested browser
was Chrome 152.0.7977.76 through Playwright 1.60.0. Viewports were 1440 × 900
for the desktop portal and 390 × 844 for mobile.

## Standalone and narration checks

The [full standalone-bank run](verification/standalone-bank.json) checked all
349 slides at desktop, compact desktop and mobile sizes; source and transcript
alignment; accessible explanations; PDF printing; and actual playback, speed,
seek and end transitions for every track. HTTP and portable-file scenarios
exercised missing or corrupt media, retry, denied autoplay and delayed playback.

Visual review then corrected one heading that said “three” while its table
contained four examples. `eve-context-failure` now reads “Fluent answers can
still use the wrong evidence,” and its narration was regenerated. Its notes and
source references were unchanged. Hash comparison confirms the other 34 guide
HTML files and 348 MP3s remained identical. The full current portal run above
includes the corrected slide, its audio and the updated PDF. The
[current bank source, control and HTTP checks](verification/current-bank-controls.json)
also passed after the correction. The earlier full-bank hash is retained as
evidence of that run; it is not presented as the current bank hash.

All 349 current tracks separately passed spoken-text/settings fingerprints, file
inventory, byte count, SHA-256, probed duration and full FFmpeg decode. Actual
Misaki frontend tests verify the requested **oh-shuhn** pronunciation for Oshun,
OSHUN, Ọṣun and decomposed Unicode spelling.

## Source, static and visual review

The builder's freshness check passed for the portal, focused decks, internal
regression fixture, coverage maps, catalog and scripts. All 18 Python tests,
Ruff, scoped ESLint and Prettier checks passed. These tests include source
escaping, unresolved references, exact excerpts, diagram structure, prerequisite
ordering, duration aggregation, preservation of foundation slides, stale or
corrupt audio and narration process recovery.

Visual review covered all 105 added slides in desktop and PDF renders, the first
and last pages of every focused PDF, and representative portal and mobile
states. Final portal table captures begin at the first column; keyboard and
accessibility inspection can otherwise leave a capture at an intermediate scroll
position.

The [verification summary](verification/summary.json) records the final
inventory, audio totals, correction hashes and check scope. The reports verify
the library artifacts and behavior. Product acceptance remains governed by the
evidence identified in each presentation.

Reproduction commands and dependencies are in the [library README](README.md).
