# Tara web practice and player · delivered increment · 8 September 2026

This increment follows the actual `/tara` → `/tara/sit/[id]` journey, then
traces start, completion, reflection and the later room read. It belongs in the
[full V1 program](v1-completion.md). The implementation baseline is
`4d605bc2a5fed0685683edce5c4a2b354cca90ce`; the primary source pin remains
`1ceae52c7896a59ac74a4429d410f3218f0b91c6`.

Visual thesis: the cream field guide connects recognizable product controls to
native diagrams of their actual state and data. Use focused source-component
captures where recognition helps; give transitions and read/write boundaries
their own diagrams. Keep sample data and implementation qualifications legible.

Content plan: choose a sitting from the path; inspect the room's progress
overlay and route resolution; use the player; distinguish its timer, captions,
transcript, ambient sound and voice control; follow start/completion writes;
inspect reflection outcomes; return to Tara and verify what was retained.
Separate this web flow from the shared recommendation/catalog and native audio
models. Those subjects need their own complete explanations, not a claim that
they automatically power the room.

Interaction thesis: the existing slide navigation, transcript/source reveals and
narration remain sufficient. Source-component captures may be panned with the
established keyboard-accessible product-view layout, with complete static native
callouts and a legible PDF crop. Do not add decorative motion or change the
product runtime to make a capture look more complete.

## Source and capture method

Use the complete pinned TaraSitPlayer component, the complete AST-selected
TaraRoom function and actual Lilith primitives/tokens/CSS. Controlled input
comes from the pinned room builder and an explicit completed-ID fixture. Declare
any host adapters, optional components or HTTP responses replaced by the
isolated capture harness. Crop only the implemented workspace being taught; do
not present the harness as a deployed page, live member or authenticated
backend. Retain input/output hashes and reproducible interaction assertions.

The [discovery record](v1-tara-practice-discovery.md) describes the complete
player, room and audio-manager reads. The current increment additionally read
the full practice-model builder (804 lines), actual start/completion repository
methods and their SQL definitions. That reveals important boundaries: default
metadata is not proof of actual accessibility/media delivery; completion writes
generate new source IDs, perform sequential inserts and do not preserve the
player's partial-completion vocabulary in the canonical event projection. Record
those details in their owning slides with exact source characterization.

## Working story

1. A selected sitting, a local finish and a later return are separate outcomes.
2. Show the implemented course rows with declared sample progress.
3. Explain the twelve-item room overlay, first unfinished row and all-done edge.
4. Identify the different room, today and recommendation feeds.
5. Resolve a sitting ID and the metadata passed to the player.
6. Show the actual transport, timer and accessibility controls.
7. Follow the local start/pause/drift/finish transitions.
8. Calculate scrubbing, elapsed position and the 80% early-finish boundary.
9. Separate generic progress captions from the authored sitting transcript.
10. Trace ambient audio and the separately available guidance endpoint.
11. Follow the start request and its retry/receipt boundary.
12. Follow completion HTTP success, HTTP failure and thrown failure separately.
13. Inspect source records, canonical projections and sequential-write failure.
14. Explain the room's completion readback and the distinct streak calculation.
15. Show the online reflection label and what its handler actually performs.
16. Follow offline local storage, failed local storage and unperformed replay.
17. Distinguish a new web mount from canonical checkpoint/native playback.
18. Follow contextual room links versus generic completion links.
19. Resolve the opening case with the actual write and later read evidence.

This list is an authoring sequence, not a slide-count cap or completion claim.
Keep the guide in drafts until optical review, source characterizations,
narration, PDF and affected center checks establish its delivery. Existing Eve
and V1 manuscripts and recordings must remain exact. No source unit is assigned
without complete teaching review. Companion/mentor/workbench and the remaining
V1 subjects stay open.

## Candidate review

The 21-slide candidate has 3,170 words of explanation and three reader questions
per slide. Model optical review covered the complete compact-desktop draft. Four
slides received polish: the speed-control row label, two repeated handoff
captions and the canonical record title. All four revised desktop captures and
representative 320-pixel layouts were inspected again. A narrow geometry check
captures all 21 slides. Three implemented product crops support recognition; the
remaining diagrams explain actual state, data and ownership.

The image manifest originally derived pixel dimensions from rounded CSS bounds.
A provenance test caught Chromium's different raster-bound rounding. The harness
now reads dimensions from the emitted PNG header and was rerun successfully; all
eight source captures and the three manuscript media entries agree. Source
component and VM tests have separate scopes documented in the delivery receipt.

The guide is sequenced directly after Tara's overview and before ritual sessions
and scheduling. Earlier guide manuscripts, HTML and narration remain unchanged.
The final delivery checks and 21 recordings pass; see the retained
[receipt](../verification/v1-tara-web-practice-2026-09-08/README.md). The full
V1 program and human acceptance remain open.

The final semantic browser review found that the local-transition diagram's
actions and destination states were visible, but its starting states were only
in structured metadata. Each row now names its exact starting state(s) in the
visible detail. This fifth composition correction preserves the already-correct
explanations and recordings. The failed attempt remains separate from the
settled final matrix.
