# Account controls · delivery evidence · 8 September 2026

The new [seventeen-slide account guide](../../decks/member-account-controls.html)
adds a complete account learning route after Lilith web/PWA. A member's push-off
request connects surface selection, profile/preferences, persistence and refresh
races to specialist memory, voice, safety, notification, Telegram, data-rights,
connection and billing journeys. Native diagrams and record specimens explain
who owns each result. All seventeen tracks add 1,621.576 seconds (27.0 minutes)
of narration. The complete PDF is delivered alongside the HTML.

The [authoring record](../../authoring/v1-account-controls.md) documents source
reconciliation and every slide's job. Seven complete units from the pinned
account feature have reviewed explanatory assignments. This is scoped teaching
review, not complete assignment of the supporting source families or full V1.
The [V1 completion program](../../authoring/v1-completion.md) remains active.

## Verified candidate

| Evidence | Scope and result |
| --- | --- |
| `browser/report.json` | 125 checks: seventeen slides at four standalone widths, three embedded views, and six enlarged-text cases. Axe covers every embedded slide. All three lost-content controls are rejected; no runtime or desktop fit findings. |
| `center.json` | Fourteen complete focused checks, including library navigation/search/history, measured durations, mobile drawer/focus, every new slide, audio end/replay at desktop/mobile and the complete PDF. Every earlier guide HTML and all four shared UI files equal baseline `ea93aa4add757fe4ec8b3f2aa75811c95ee93ac2`. This establishes unchanged bytes, not a new acceptance claim for those guides. |
| `pdf-visual-text.json` | Exact title, subtitle, takeaway and all visible-body fields in order across seventeen PDF pages. The center additionally checks print geometry, table typography, folios and complete sequence. |
| `speech-reuse.json`, `logs/account-narration.log` | All 580 earlier manifest entries and MP3 hashes are unchanged. Seventeen new recordings were generated on one CPU thread; the renderer verified all 597 encoded tracks. |
| `logs/account-units.log` | 73 passing Python tests covering account assignments, builder, compositions, exact PDF visual fields and narration. |
| `logs/account-node-units.log`, ownership/regions logs | 32 passing Node tests across change scope, baseline chaining, audio scope, captures, durations, canonical handoff inventory and scrolled region order. |
| `logs/account-teaching.log` | Seven account units newly reviewed; all current assignments valid. The whole-library inventory remains explicitly incomplete. |
| `logs/account-eslint-final.log`, `logs/account-ruff.log` | Targeted JavaScript and Python lint passes. Final formatting/freshness output is retained with the receipt. |
| `state.json` | Frozen artifact, narration and evidence hashes for this increment. |

Model optical review inspected all seventeen final 1280px browser slides and the
full PDF sequence. PDF detail views cover the six-group table, ownership
handoff, refresh gate and connection record. Mobile detail review covers the
refresh branches and notification handoff; every mobile slide additionally has
semantic, reflow, accessibility and reachable-navigation automation. Browser,
PDF and center captures are retained. Sequence contact sheets establish pacing,
not a substitute for native text inspection.

No human reader session, normal-speed human listening or new live member action
was performed. The guide labels hypothetical cases and distinguishes the
walkthrough's recorded behavior from this presentation's verification.

## Findings resolved

The control-family map originally gave incoming arrows verbs that implied the
controls acted on the account hub. The final labels correctly describe controls
being edited or summarized through it. The profile explanation corrects the
feature prose's concurrent-revision interpretation: the actual client uses a
local mutation-generation guard and discards outer server revisions.

A missing handoff could previously escape the geometry auditor because it
checked only the exchanges left in the DOM. Guide review now compares complete
ordered owner/exchange inventories and labels with the canonical manuscript.
The deliberate missing notification readback is rejected by that check.

The first center run failed when mobile playback scrolled the slide: the hidden
masthead has a zero-sized box, so comparing it with the negative viewport
coordinate of the scrolled header falsely reported overlap. The corrected check
excludes only the authored hidden masthead and requires all other regions to be
present, nonempty and correctly ordered. Negative overlap/missing-content tests
pass. The failed report and log are retained under `attempts/` and `logs/`.

The initial unpublished draft preview passed all layout/source checks, then its
nested PDF check used macOS Python 3.9, which lacks `zip(strict=True)`. The same
export passed with Python 3.11; every final PDF/narration command uses that
required environment. No shared presentation CSS, playback or center UI changed.

## Reproduce

Use Python 3.11 plus the existing narration/PDF environment and repository
Playwright. Run expensive processes serially under `run-supervised.py`.

- `guide-review.mjs` with `OSHUN_PRESENTATION_REVIEW_CONFIG=authoring/v1-account-review-config.json` and an explicit output directory.
- `presentation-center.mjs --guides member-account-controls --base-revision ea93aa4add757fe4ec8b3f2aa75811c95ee93ac2` with an explicit center output directory.
- `check-eve-depth-pdfs.py --source-map authoring/v1-account-source-map.json --report <output.json>` for the exact visual text.
- `build-eve-oshun.py --check --preview` and `check-center-coverage.py --check` for final artifact and assignment freshness.
- `verify-receipt.py` beside this file for frozen delivery/evidence continuity.

Publication of this account increment leaves offline mechanics, room feature
depth, workbenches and the rest of V1's source/design review active.
