# Tara selection and Lilith handoff · delivery receipt · 8 September 2026

The [25-slide selection guide](../../decks/tara-selection.html) and
[19-slide handoff guide](../../decks/tara-lilith-handoff.html) teach the full
feature catalogs, independent availability and distress rules, type selection,
explanatory fields and thirteen policy summaries. Each has a complete PDF and
narration. [Authoring](../../authoring/v1-tara-selection.md) records source scope,
sequence, visual decisions and application boundaries. Baseline:
`0a6efd8473a7497da0660d04f1953d245eaa2ecd`. Source pin:
`1ceae52c7896a59ac74a4429d410f3218f0b91c6`.

The library has 70 guides, 790 unique slides, 813 guide-slide instances and
415 cited paths. All 746 prior slides, scripts, manifest entries and recording
bytes remain exact. Forty-four new tracks add 3,508.925 seconds
(58.5 minutes). Full V1 and application acceptance remain open.

## Evidence

- [Source report](source/report.json): five complete modules, 64 unmodified
  owning tests and 21 added cases, including all 144 mood/distress/classifier
  combinations. Explicit temporary feature/persona facades isolate module
  dependencies. [Continuity](source-continuity.json) binds ten source/test paths.
  [Owning test output](source/owning-tests.json) retains actual Vitest results.
- [316 final browser checks](browser/report.json): 176 captures, 132 embedded
  axe runs, eight enlarged-text cases and five rejected removals: the theme
  catalog, retained modality outcome, invalid-type outcome, outer permission and priority selection branch.
  No unexpected errors or fit findings remain.
- [Focused center review](center.json): both guides, all 44 slides in two embedded
  contexts, four audio boundaries and 44 exported PDF pages. All 68 prior guide
  HTML files and four shared UI assets remain exact. Browser audio output is
  disabled. [Visible PDF text/order](pdf-visual-text.json) and
  [optical review](optical-review.json) provide separate evidence.
- The settled presentation suite passes 119 cases. Scoped narration decode,
  freshness, coverage, ESLint, Ruff, formatting and local-link checks pass.
  [Speech preservation](speech-reuse.json) compares all prior scripts, entries
  and MP3 bytes without claiming fresh human listening.

## Attempts and limits

The first source fixture had an extra parenthesis, caught by formatter and lint;
removing it left source expectations unchanged, then the complete source run
passed. Initial authoring used the default Python 3.9 and encountered the known
zip(strict=True) requirement; the Python 3.11 environment rendered both drafts.
The anxious comparison was changed from one mood row to its four candidate rows
before the first draft build. Draft review passed 88 desktop/narrow checks.
Optical review then removed repeated comparison takeaways, shortened priority
labels, exposed required resource kinds and moved mood fallback before the
combined selection diagram before final delivery. The first final browser run exposed a checker gap: priority diagrams were
skipped by the semantic audit. The checker now verifies their nodes, visible
branches, accessible relationships and selected-rule emphasis. The settled
review rejects both missing outcomes and a missing selection branch. The
[failed attempt](attempts/browser-1/report.json) is retained separately.
The first PDF text check lacked a priority-layout contract. It now checks both
terminal outcomes and every condition/branch label, including text painted after
the folio; a regression rejects missing outcomes, labels and relocated conditions.
Earlier draft captures deliberately retain their
pre-polish state. Failed logs are not passing evidence; original trailing
whitespace, where present, is preserved in gzip copies identified by
[log preservation](log-preservation.json).

Run `python3 docs/presentations/presentation-center/verification/v1-tara-selection-2026-09-08/verify-receipt.py`
from the repository root. Later library edits invalidate this frozen receipt;
use the delivery revision for historical verification. The
[full V1 program](../../authoring/v1-completion.md) remains active. Actual start
suppression, session/audio control, resource availability, clinical review,
follow-up and recovery need their owning consumers. Source policy data is not
current contact guidance or clinical acceptance. No live member data was used.
