# Tara sessions and recovery · delivery evidence · 8 September 2026

The [new 21-slide guide](../../decks/tara-ritual-sessions.html) follows an
interrupted practice through lifecycle history, completion decisions,
continuation, skip, recovery and the evidence needed for a saved return. Record
specimens, state relationships, decision gates and numerical cases distinguish
the session contract, domain completion result and player telemetry. The guide
follows `tara` in the member-experience sequence and includes 3,396 words of
speaker notes, three reader questions per slide, 32.2 minutes of narration and a
complete [PDF](../../decks/tara-ritual-sessions.pdf).

Two existing Tara slides were corrected: taxonomy consistency does not establish
recommendation enforcement, and the continuation schema does not compute actual
readiness or authorize its declared memory scope. Peer continuation fields now
use record anatomy; taxonomy columns have equal weight. The new tables use
meaningful headings and no positional emphasis. Shared renderer and center UI
code are unchanged.

The [authoring record](../../authoring/v1-tara-sessions.md) reconciles claims
against actual pinned schemas and library functions. Four complete inventoried
feature units have reviewed teaching assignments. Practice selection, player
checkpoint behavior, scheduling, companions, mentor presence and the Tara
content workbench remain in the [full V1 program](../../authoring/v1-completion.md).

## Verified scope

| Evidence | Result |
| --- | --- |
| `browser/report.json` | All 31 slides in both affected guides at four standalone widths and three embedded views, plus six enlarged-text cases: 223 checks, 100 captures, 93 embedded axe instances, no runtime or fit findings. Three missing-content controls are rejected. |
| `center.json` | Fourteen complete focused center checks covering navigation, search/history, measured durations, mobile drawer/focus, slide/source/transcript behavior, four audio end/replay boundaries and both complete PDFs. All 61 other guide HTML files and four shared UI files match the immutable baseline. |
| `pdf-visual-text.json` | Exact authored titles, subtitles, takeaways and visible body fields on all 10 + 21 PDF pages, in their expected order. Center checks also cover print geometry, table typography and folios. |
| `speech-reuse.json`, `logs/tara-narration.log` | Twenty-one new tracks plus two corrected tracks generated; all 641 recordings pass the renderer's encoded audio, hash, fingerprint, duration and pause checks. All 618 other manifest entries and MP3 bytes match baseline. |
| `logs/tara-presentation-unit.log` | 108 passing Python tests for builder, composition, table/ownership semantics, PDF fields, narration and V2 preservation. |
| `logs/tara-node-tests.log` | 60 passing Node tests, including 16 new immutable-source Tara characterizations and the existing member-contract, Tara-read, center-scope, duration, ownership and slide-region checks. |
| Source fixtures | Actual pinned schemas/functions execute with explicit boundary fixtures; cases cover inconsistent history, unchecked declarations, audio floors, prerequisite readiness, 100% without completion, optional minutes, skip refusal, retained recovery time, projection identity and device routing. No database, player or device synchronization is represented. |
| Freshness and lint logs | Scoped ESLint, Ruff, canonical formatting, builder and coverage freshness pass. Teaching coverage remains incomplete: 79 reviewed units, 10,018 structural units and 103,247 unassigned units. |
| `source-continuity.json` | All nine cited source files are byte-identical between the primary pin and implementation baseline. Source-unit assignments separately delimit reviewed teaching. |
| `state.json` | Frozen delivery inputs, retained evidence and every recording hash; `verify-receipt.py` also requires bound files to be tracked by Git. |

Model optical review inspected every new desktop slide at 1280 pixels and both
corrected overview slides. All 31 final PDF pages were reviewed as complete
sequences; native PDF pages 6, 9, 14 and 17 of the new guide and page 7 of the
overview establish specimen, table and worked-case detail. Final mobile record
anatomy, continuation and recovery cases were inspected at native size. The
remaining mobile slides have complete reflow, semantic, navigation and
accessibility automation. All final browser captures, 62 focused center slide
captures and 31 PDF page renders are retained.

No representative-reader session, normal-speed human listening or live member,
database, notification or device acceptance was performed. The technical checks
and model visual review do not substitute for that evidence.

## Findings and corrections

The first draft used the wrong table input key; the builder rejected it. The
corrected draft passed its standalone layout and PDF checks. Source-fixture
setup first needed the existing web package's Zod dependency and explicit
companion/navigation boundaries. A failing semantic expectation then showed
that `stateChanged` is declared but not recomputed by the session refinement.
Direct source review confirmed this; the explanation now teaches that limit,
and the accepted-edge characterization is retained.

The first teaching assignment check rejected a skip slide that omitted its
governing feature citation. The citation and assembled evidence fingerprint
were corrected. The first full guide review was incomplete: layout checks
preceded narration, and final-track playback timed out because that track did
not yet exist. The failed report and logs remain in `attempts/` and `logs/`.

After narration, the full matrix passed. Native optical review then found four
awkward labels across three slides. Shorter labels preserve all adjacent field
details; a plain-language recovery claim removes a stranded hyphen. The final
matrix passed again because those visible inputs changed. Eighty-eight captures
are byte-identical to the earlier passing review; twelve captures reflect only
the three revised slides. `render-continuity.json` records both sets, and the
complete pre-polish report and captures are retained. No speech regeneration
was needed for those composition labels.

Upstream commit `336013f511c0b4dcebfe47b50f2002b4f0f62798` was integrated before
final center/PDF verification. Its Eve computer-use evidence and generated
indexes do not alter the presentation inputs or reviewed Tara sources.
`upstream-integration.json` records this continuity. The center comparison base
remains `4d893c29bb7a24f7c1b1dde051ae4d075db0d6ae`; it is an immutable artifact
baseline, not an assertion that the remote branch is still at that commit.

## Reproduce

Use the existing Python 3.11 narration/PDF environment and repository
Playwright with Chrome. Run expensive commands serially under
`tools/presentations/run-supervised.py`.

- `guide-review.mjs` with `OSHUN_PRESENTATION_REVIEW_CONFIG=authoring/v1-tara-sessions-review-config.json` and an explicit review output directory.
- `presentation-center.mjs --guides tara,tara-ritual-sessions --base-revision 4d893c29bb7a24f7c1b1dde051ae4d075db0d6ae` with an explicit center output directory.
- `tests/check-eve-depth-pdfs.py --source-map authoring/v1-tara-sessions-source-map.json --report <output.json>`.
- `python3 -m unittest test_build test_composition test_tables test_ownership test_eve_depth_pdf test_narration test_v2_rollback_session` with the presentation tools/tests on `PYTHONPATH`.
- `node --test --test-concurrency=1` for `pinned-tara-sessions`, `pinned-member-contracts`, `pinned-tara-handoffs`, `center-change-scope`, `center-baselines`, `center-audio-scope`, `center-duration-labels`, `ownership-content` and `slide-region-order` under `tools/presentations/tests/`, each with its `.test.mjs` suffix.
- `build-eve-oshun.py --check --preview`, `check-center-coverage.py --check` and this receipt's `verify-receipt.py`.

This receipt closes the scoped technical delivery increment. The full V1 audit
and reader/listening acceptance remain open.
