Source revision: 1ceae52c7896a59ac74a4429d410f3218f0b91c6. Review date:
2026-09-06. This record accompanies the V1 web/PWA, Tara and Arete narration
rebuild. It does not establish whole-domain teaching or live product acceptance.
Read the complete pinned Tara and Arete feature pages, account/profile page, brand record, PWA service-worker helper and Tara playback-rate implementation. Also inspected the pinned Arete check-in refinement and treatment helpers and Tara's modality-family enum. Compared these with all ordered slides and explanations in the three guides. The source reading is broader than the two reviewed assignments below; it does not automatically mark any other unit reviewed.
Corrected and reviewed units#
- Tara: Playback rate is a real contract, lines 353–376. The feature page
groups non-finite and out-of-range values under an overly broad clamping
exception. The actual implementation checks finiteness first and always throws
PlaybackRateRangeErrorforNaNor either infinity. Clamping only admits finite out-of-range requests. The correctedtara-audio-accessibilityexplanation retains every frozen policy field, two-place precision and pitch intent; examples show0.93 → 0.95, default rejection of1.40, explicit clamping to1.25, and unconditional infinity rejection. The snapshot tests execute this standalone pinned contract, not a current device/player. - Arete: The check-in: status, engagement, and streak treatment, lines
104–132.
missis an explicit status, not an absent database row. The correctedarete-check-in-contracttakeaway and explanation preserve the habit/local-day unit, all five meanings, engagement bounds, reason requirements and shared treatment helpers. Skip and decline share grace but remain distinct choices. No database uniqueness or live persistence claim is inferred from a schema or helper.
The two assignments in
member-playback-and-checkin.json
bind the exact source units and assembled slide fingerprints. Later source or
slide changes invalidate that review. Notes, questions and sources remain
available without audio. Narration readiness and automated playback do not
establish a normal-speed listening review.
Remaining depth and faithful guide decomposition#
The existing guides introduce journeys and boundaries; they are not exhaustive reference guides. Continue with coherent mechanism and worked-case guides, not a fixed slide or guide quota.
Tara#
- Reconcile the exact mood/theme/modality/context/lineage reference datasets with their contracts. The feature prose says six modality families while its own displayed enum names seven. The pinned contract confirms all seven: stillness, guided, breathwork, sound, embodiment, visualization and devotional; a snapshot inventory regression preserves this correction. Cover every entry's eligibility, distress, fallback and lineage-comparison rules, rather than treating a taxonomy overview as a full inventory explanation.
- Teach exact breath phases and the alternate-nostril null-cadence distinction, plus all five duration buckets' defaults and recommendation spacing. These remain product contracts, not personalized medical suitability guidance.
- Give the session transition table, event consequences, timeline invariants,
completion thresholds and checkpoint/recent-completion constants their own
worked explanation. The source specifically disclaims a shipped
drift_idle_secondssetting; modeled drift is an event, not proof of an idle detector. - Extend continuation beyond the overview: full program, teacher and theme fields, all prerequisite timestamp/blocked-ID rules and scope/consent governance. Explain ritual assembly, cadence, quiet hours, calendar conflict, humane-miss policies, triggers, tone-review axes and teacher rights/locales.
- Map all consumer, depth and mobile companion surfaces to their owners. Cover
the domain adapter,
@tara/*stack and meditation-runtime family separately. A mobile companion component is not a dedicated mobile route tree. - Keep cross-device resume, richer haptics/downloads and immersive arc evidence distinct. Mentor Presence is additional source material: the feature page describes a default-off, opt-in, entitlement- and human-release-gated extension. The ten-slide guide does not yet teach its complete contracts or establish that its release gates passed.
Arete#
- The check-in refinement child unit remains unassigned: it also teaches exact mood/energy ranges, friction notes and typed evidence references. Do not extend the parent-unit review by association.
- Teach habit, goal and routine fields, all conditional cadence/archival and lifecycle invariants, per-habit grace configuration and the distinction from the recovery engine's default policy.
- Work through cadence-aware miss counting, recovery stages, protected versus archived momentum, completion updates and human-controlled rescoping. Map missed-day, streak, recovery and plan records to their exact transitions and evidence requirements.
- Separate the schema's eleven friction signals/nine interventions from the domain taxonomy's twelve categories/twelve interventions. Explain evidence bands, visibility/consent, approval and the complete weekly-review sections, limits and lifecycle instead of substituting a general coaching overview.
- Preserve the singular
/arete/offeringcomposition route versus plural/arete/offeringsgallery, all other consumer/depth surfaces and their BFF ownership. Service scaffolding is not the member persistence tier. Trace BFF stores and client-local state, current provider/UI seams and Living Offerings governance independently.
Web/PWA and account#
- Extend the four availability questions with complete route/shell, release, service-worker, account and design-system units. The account overview does not enumerate the full section set, every endpoint field, upload limits or substrate-specific control lifecycle.
- The pinned PWA helper's
queueBackgroundAction()returns true after posting a message to the worker; that return is not a durable queue acknowledgement or a server receipt.applyWaitingServiceWorker()likewise requests worker activation rather than proving refreshed content. Work through those request/effect boundaries with the consuming worker and UI before claiming an offline journey is taught or verified. Two pinned-helper regressions use a fake worker that never acknowledges or persists anything; they verify the return-value distinction, not a real browser-worker journey. - Teach registration guards/retry, waiting/install notifications, reading-route detection and standard versus reading update copy. Member-cache marking, privacy-first fallback and member-to-anonymous purge requests require the actual worker/storage path to prove deletion; a helper returning true alone is insufficient.
- Complete the account's nine route rows, panel section inventory, authenticated profile/preferences projections, revision rereads and substrate-owned memory, persona, channel, billing and data-rights behavior. A saved binding or preference does not establish downstream delivery or cross-device revocation.
Continue the full whole-monorepo and cross-cutting learning map in TODOS.md, including implementation, UI/UX, screenshots, generation, Domains and real Eve integration. These findings are not an exclusion list or a replacement for the full source-unit inventory.