# Tara offline audio and retention · delivered increment

Source snapshot: `1ceae52c7896a59ac74a4429d410f3218f0b91c6`. Implementation
baseline: `eb2bfed6bbcd418e6958503e78ee81fa39bca190`.

Visual thesis: use the established warm paper surface, large dark type and one
rust accent to distinguish record ownership and make the return path easy to
trace. Content plan: introduce the member's offline return, identify the three
owners and records, follow enrollment and integrity, then explain resolution,
quality, retention, budgets, clearing and storage. Close with an investigation
that follows the same sequence. Interaction thesis: preserve keyboard slide
navigation, the center's persistent topic transition and its narration controls;
use the existing reduced-motion behavior and semantic disclosure surfaces.

Each slide has one takeaway, two specific questions and three narration
paragraphs. The ownership diagram emphasizes the initiating cache explicitly;
peer comparison rows carry equal weight. Worked cases supply concrete inputs,
with short visual copy and fuller explanations in the notes. The guide includes
an ownership exchange, record specimens, ordered cases, comparisons and a plan
budget table.

The source fixture executes four complete modules: TaraOfflineCacheStore,
DownloadManager including createExpoFileTransferPort, offline asset budgets and
MemoryOfflineStorage. Ten unmodified owning Jest tests retain their own transfer
doubles, with an explicit settings snapshot seam. Seventeen additional cases
execute the source over declared dependencies. The native adapter cases use
controlled File, Directory, download-task and Crypto implementations; the digest
seam computes SHA-256 of the actual string supplied to it. No native filesystem,
network, mounted hook, playable URI, durable storage or user acceptance is
claimed. Metadata examples reuse the exact builder output retained in the
[earlier native-caller receipt](../verification/v1-tara-audio-2026-09-08/native-caller/report.json).

The source cases distinguish descriptive audio checksums from encoded-file
hashes, raw-byte and base64-text presence digests, non-audio fallback selection,
quality replacement, insertion/truncation order, operation-triggered expiry,
default memory storage, audio-only clearing, externally owned preference state,
repair plus requeue, declared versus transferred size, manifest-only removal and
hydration validation. These findings characterize the pinned implementation;
they do not change the product or establish broader native acceptance.

The [delivery receipt](../verification/v1-tara-offline-2026-09-09/README.md)
records the complete source fixture, 144 browser checks, 80 captures, sixty
embedded axe runs, four enlarged-text cases and four rejected content removals.
Center review covers twenty slides in forty embedded captures, two audio
boundaries and the complete PDF. All twenty final PDF pages match the reviewed
draft pixels exactly. The initial closing slide extended 7.25 pixels below the
desktop frame; shorter action rows restored fit while preserving narration.
Reflection, session checkpoint consumers, quick reset, sleep, companions,
mentors, workbenches and the rest of V1 remain open.
