The 21-slide guide and
PDF follow a five-minute sitting from the
course through controls, completion, reflection and return. Native state,
decision, request and record diagrams accompany three real source-component
crops. Authoring records sequence,
source boundaries and optical corrections. The implementation baseline is
4d605bc2a5fed0685683edce5c4a2b354cca90ce; the primary source pin is
1ceae52c7896a59ac74a4429d410f3218f0b91c6.
All 659 previous slides, scripts, narration manifest entries and MP3 hashes are preserved. Twenty-one tracks add 1,671.198 seconds (27.9 minutes). The library has 65 guides, 680 unique slides, 703 guide-slide instances and 399 cited paths. No source unit is assigned merely from citations. The combined Tara practice/assembly unit and full V1 program remain open.
Delivered explanation and evidence#
The guide distinguishes sparse course progress from the displayed count, the room feed from recommendation routes, timer position from listening, generic captions from authored transcripts and ambient media from voice guidance. It then traces guarded start requests, HTTP versus thrown completion failures, sequential source/canonical inserts and their fixed projection fields. Online reflection's Saved label has no save request in this handler; offline storage, tab-only fallback and unperformed replay are taught explicitly. The final case checks the owning write and later read before promising retention.
- Thirteen controlled browser cases execute the complete pinned TaraSitPlayer and AST-exact TaraRoom with actual Lilith primitives/CSS. Eight continuous source crops retain PNG dimensions and hashes in the capture manifest. HTTP, auth, clock, storage and optional-component seams are declared. This establishes source-component behavior, not a deployed authenticated member session.
- Ten immutable-source VM cases execute the full room/scripts and complete AST-selected repository methods. Pool responses, clock and IDs are controlled; emitted queries, payloads, sequential failure and readback are characterized. No live PostgreSQL execution or durable database acceptance is claimed.
- The settled presentation and preservation suite passes 108 cases. Nine separate asset/Eve regression cases pass. Targeted ESLint, Ruff, formatting, generated-artifact freshness and coverage checks pass. The fifteen cited sources are byte-identical between the source pin and implementation baseline.
- The final browser matrix passes 153 checks across desktop, compact desktop, 390/320-pixel mobile, three center embeddings and six enlarged-text cases. It retains 84 captures, 63 embedded axe checks and three deliberately rejected removals (capture, partial branch and canonical owner). No fit finding, document overflow or runtime error remains.
- The complete focused center review passes fourteen checks, preserves all 64 earlier guide HTML files and four shared UI assets, checks navigation/search/history/mobile focus and both audio boundaries, and exports all 21 PDF pages. Forty-two embedded captures are retained. Browser audio output was disabled.
- The PDF text report verifies the exact visible-body labels and order on every page. Optical review records the actual model-inspected draft, revised, mobile and final PDF images. Text extraction alone does not establish visible ink or legibility.
- New narration receives encoded-audio/pause validation and a final 21-track decode check. Speech continuity records exact prior-entry, script and MP3-hash equality against the 659-track publication baseline. All 680 final hashes are bound; prior tracks reuse their existing decode evidence.
Corrections and reproduction#
Optical review covered all 21 compact draft slides. Five were polished: a crowded speed-control label, two repeated request-diagram captions, a long record heading and missing explicit starting-state labels. The final semantic review caught the implicit starting states; they are now visible in each transition row. Notes and recordings were already correct and remain unchanged. Revised desktop and representative 320-pixel captures were inspected again. All 21 narrow slides passed geometry checks before narration. The source-capture provenance test caught CSS-bound rounding differing from PNG pixel dimensions. The harness now reads the PNG header; all eight captures were regenerated and the media entries corrected. Failed and draft logs remain in logs and attempts; they are not counted as passing checks.
Run python3 docs/presentations/presentation-center/verification/v1-tara-web-practice-2026-09-08/verify-receipt.py
from the repository root. The verifier binds source/authoring inputs, generated
artifacts, retained evidence and all recordings. Later edits invalidate this
frozen receipt; use its publication revision for historical verification.
The full V1 program stays active. Tara recommendation/catalog, practice assembly, native callers, companions, mentors and workbench depth remain unfinished, alongside the other V1 workstreams. Normal-speed human listening, independent reader comprehension, live member, database/device and production deployment acceptance were not performed.