In progress
Admit a bounded friends-and-family TestFlight candidate
The first outside round is intended for a small friends-and-family group after the exact candidate passes Trakt and OpenAI onboarding, Today and retained library rendering, typed and voice Ask, watched/watchlist actions, silent automatic Encore delivery, Settings/key recovery, navigation, cast/person, and basic accessibility checks. TestFlight upload remains a separate action after this stabilization gate; local builds and historical archives do not count as that proof. **Deferred provider review recorded August 11:** TMDB's published developer terms include language about AI-based applications. Thomas has explicitly accepted that unresolved terms question for this free, noncommercial friends-and-family TestFlight path, so it is not an active candidate blocker. This is an owner product decision, not a claim of legal or provider clearance. Keep the published-terms note visible for later review, and reopen provider permission before any paid, subscription, revenue-generating, or otherwise commercial release. The OpenAI privacy data-flow and App Store answers still need review against the exact signed candidate.
- Last changed
- August 11, 2026
- Last verified
- Local candidate preflight in progress
- Protect
- Outside testing remains BYOK, has no developer-funded OpenAI key or paid subscription, and cannot begin from an old archive or an unverified build.
- Evidence required
- Clean native and Lab verification, current provider/privacy disclosures, exact signed candidate evidence, one joined retained-data physical-iPhone pass, and Thomas's explicit TestFlight upload decision.
Ready for Thomas
Decide the dark-art treatment on The Executioners
The shared luminance-aware lift is implemented without a title-specific image swap. Thomas still needs to decide whether the retained dark image is now clear enough.
- Last changed
- August 9, 2026
- Last verified
- August 9, 2026 · iPhone 17 Pro Simulator
- Protect
- Keep the exact retained episode still, full-bleed top edge, episode identity, actions, parent-show route, and accepted cinematic height.
- Evidence
- Current baseline screenshots 19–24 at checkpoint fae1d235.
In progress
Finish the retained-real-data product pass
Continue inspecting Today, Shows, Movies, Discovery, details, cast/person, Settings, onboarding, keyboard dismissal, artwork, and navigation as one real product. Fix shared causes and rerun every accepted surface touched by them. The August 12 usability audit found six shared presentation gaps without changing retained data: Today can briefly claim that nothing is scheduled while the useful-library projection is still restoring; Encore search lacks the shared keyboard escape controls and its retry tile does not explain the failed state; a single genuine cast result is stranded in a two-column layout; the Settings setup reminder omits personalized discovery and library intelligence; show ratings visually reserve a full rating tile for an unavailable source; and movie progress can read `Continue from Resume at …`. The active correction slice is limited to those shared causes. It must preserve the approved Today cards, newest-first seasons, genuine artwork fallback order, Cape Fear/Food Truck/Nicholas/Married Encore rendering, provider-read boundaries, and automatic Encore policy. The August 13 Lab refresh re-ran Today, Shows, Movies, Discovery, consumer Settings, Cape Fear, Great Food Truck Race Progress/episode/cast/person, Nicholas Galitzine, and Married with Children Encore against the current retained isolated-test-account Simulator. That review exposed one shared defect before publication: episode rows inside a routed detail destination received the default navigation environment action, so a tap could do nothing. Navigation is now injected into both the root and every routed destination; Cape Fear and Food Truck episode journeys passed on rerun. This fixes the shared route rather than either title. Provider settlement remains separately blocked by the expired isolated test-account Trakt credential.
- Last changed
- August 13, 2026
- Last verified
- August 13, 2026 · retained-data read-only Simulator matrix
- Protect
- Do not alter an accepted page while correcting another page, and never replace missing provider data with invented fixtures.
- Evidence
- Current product baseline and retained UI gap evidence.
In progress
Complete conversational Ask as the app-wide concierge
Grounded ShowNext tools, natural visible-page follow-ups, episode and season understanding, voice recovery, and broader conversational behavior still need a joined retained-data pass. Apple Foundation Models remain bounded R&D rather than the concierge brain.
- Last changed
- August 6, 2026
- Last verified
- August 6, 2026 · internal Simulator evidence
- Protect
- Keep OpenAI compiled in but off by default, BYOK as the current lane, visible sources, exact page context, consent, spoiler limits, and preview-before-mutation safety.
- Evidence
- Conversational tool architecture and OpenAI Realtime evidence.
In progress
Complete library actions without teaching magic phrases
Thomas explicitly approved real episode-action testing on the isolated isolated Trakt test account on August 13. Progress rows, multi-select, season actions, mark-through actions, and episode detail now share exact Trakt episode identities and one durable mutation/read-back path. A normal episode mark is one action, not a preview or a second confirmation. Provider snapshot, pending action, and verified read-back state are merged so a later mark cannot erase an earlier visible mark. Interrupted and ambiguous requests persist a scheduled read-back before any retry, and launch restoration is owned by the account runtime rather than a disposable view task. Controlled evidence passed, but Thomas's retained-data walkthrough rejected the result: a Reacher row remained spinning without changing progress, a bulk episodes 1–3 action did not settle, and an episode-detail mark did not remain visible on return to Progress. The retained outbox confirmed one transport-ambiguous episode and two unsent pending episodes. The active repair removes the redundant normal-row dot and secondary action sheet, keeps the existing left swipe shared with Today as the single-episode action, shows selection controls only after `Select`, and unifies compact and route-rich episode identities so detail, Progress, restoration, and bulk state cannot diverge. One shared projection must drive the unmistakable gray watched row, checked `WATCHED` label, inverse action, selected-season count, season-strip count, progress bar, and episode detail state. A retained relaunch then exposed the shared external blocker: the saved isolated test-account Trakt refresh token is rejected with HTTP 400, so no single or batch action can currently reach provider settlement. The DEBUG row now keeps that same learned gesture: it says `SAVING` while the write settles, then either remains visibly gray as `WATCHED` or rolls back and says `NOT SAVED · SWIPE TO RETRY` instead of looking successfully watched. The isolated DEBUG Simulator continues to use the test account for repeatable and destructive verification, while normal connected accounts—including Thomas's phone account—must receive the same user-initiated watched experience intended for TestFlight users. Shipping gateway admission is unchanged. Completion now requires reconnecting the isolated test account, then real add, cumulative bulk, detail/list, relaunch, provider read-back, and cleanup proof there; controlled tests alone cannot return this item to owner review. This authorizes ordinary in-app actions Thomas performs while testing; it does not authorize a separate bulk-maintenance script or unattended personal-account mutation campaign. Thomas rejected a reduced Progress gesture that copied only Today's left-swipe mutation. Progress vertical cards must use Today's complete interaction: tap opens, left swipe reveals the exact watched action, right swipe reveals More, and action colors, motion, feedback, edge lighting, verified treatment, and retry behavior stay consistent. `Select` is the sole Progress-only addition for batch changes. Earlier rendered state tests remain useful for count/progress convergence but no longer count as interaction-parity evidence. **August 13 physical-iPhone rejection:** the installed Shows collection could reorder while Thomas was scrolling because individual poster metadata changed the live Popular sort. The same build exposed Mark Watched on a main-account episode while an obsolete username gate refused the write, so the action could only end as `NOT SAVED`. Thomas explicitly rejected that laboratory restriction: his connected phone account is the real pre-TestFlight experience and must support normal in-app actions. The active shared repair must freeze the visible Popular order while metadata hydrates, preserve legitimate catalog membership changes without snapping the scroll position, and make watched-action capability derive from a real connected runtime—not a hard-coded account name—on Progress and episode detail. This rejection is not resolved by a successful build or a Cape Fear-specific exception: it requires a long retained Shows scroll, controlled shared-action regression, and a real single-action/read-back pass before acceptance. **August 13 internal repair evidence:** Popular order is now frozen for the visible Shows session while metadata hydrates, with explicit filter/sort changes remaining the only user-driven reorder. A retained-data iOS 27 UI test scrolled seven pages, held for eight seconds while metadata continued loading, and kept the same real show card at the same screen position. Controlled UI journeys also passed for single watched state, cumulative bulk selection, inverse action, Today-compatible swipe behavior, and the truthful read-only account rendering. This is internal Simulator evidence, not provider settlement or owner approval. The authorized real write/read-back gate remains blocked until the isolated test account is reconnected; the app must not substitute a personal-account write. **August 13 account-capability correction:** Thomas clarified that his account on his phone is the real pre-TestFlight product experience, not a protected read-only laboratory. The username restriction was removed from visibility, submission, launch restoration, and season-action restoration. Five focused iOS 27 UI journeys passed with zero failures, and the physical-device Debug app built and signed successfully. No phone install or Trakt action was performed during this internal verification; Thomas's next in-app action is the real provider evidence. **Permanent account-testing boundary:** the personal account is writable through normal ShowNext actions on Thomas's physical iPhone. Simulator, UI automation, and repeatable development verification use the isolated test account only and must never sign into or mutate the personal account. A Simulator-only connection guard enforces this without changing physical-device or TestFlight product capabilities.
- Last changed
- August 13, 2026
- Last verified
- August 13, 2026 · retained isolated-test-account rejection after controlled tests
- Protect
- The model interprets natural intent; ShowNext owns exact identity, Trakt operations, ambiguity, read-back, and visible cumulative state. Normal user-initiated product actions work for every connected Trakt account; account names never determine product capability.
- Evidence
- Current tool architecture, mutation-engine receipts, retained isolated-test-account review, and the August 13 episode-action audit.
In progress
Make Season Encores a complete reward system
The visual reward experience, hook quality, completion celebration, notifications, imported-library backfill, failure recovery, and true cost of a large archive need continued measurement. Automatic generation stays enabled for an eligible BYOK account; a missing key pauses the queue without discarding it. Automatic back-catalog results must accumulate silently instead of repeatedly interrupting the active page. Any later unread badge must support multi-select and Mark all seen without deleting permanent Encores.
- Last changed
- August 9, 2026
- Last verified
- August 9, 2026 · retained Simulator and queue audit
- Protect
- Genuine artwork, rare verified facts, permanent keepsakes, automatic metered BYOK generation, sequential durable queueing, and token/cost receipts.
- Evidence
- Encore archive, ceremony, queue, and current-baseline evidence.
In progress
Review the simplified consumer Settings experience
The root now prioritizes Accounts & Connections, Your Experience, Privacy & Data, and Help & About. Trakt disconnect lives inside Trakt, ShowNext deletion lives inside Manage Data, OpenAI security detail is disclosed on demand, and developer tools require seven taps on the About version row in DEBUG builds. Thomas approved the seven-tap Developer Tools reveal on August 10. The Trakt attention state now says `Connected · using saved library` instead of claiming an incomplete sync, explains that the latest check did not finish, preserves the last-confirmed date, and offers an explicit read-only Check Again action. The retained isolated-test-account rendering is ready for Thomas's visual judgment; the Check Again request was intentionally not triggered during the visual review. The August 11 physical-iPhone candidate exposed a presentation bug: the visible Trakt detail button requested a confirmation owned by the hidden parent Settings view, so Disconnect could appear inert. The correction keeps the destructive confirmation on the visible Trakt destination and retains the existing account runtime as the only owner of the actual disconnect operation.
- Last changed
- August 11, 2026
- Last verified
- August 11, 2026 · physical-iPhone disconnect defect reproduced by Thomas
- Protect
- Required Trakt-first and verified OpenAI-BYOK-second onboarding, transactional Save → Verify → Enable, consumer-first root groups, destructive actions inside their detail pages, and hidden DEBUG-only developer controls.
- Evidence
- Focused required-onboarding and Settings UI checks plus the normal retained-data Simulator launch; truthful Trakt attention-state test and retained isolated-test-account rendering.
Ready for Thomas
Converge every catalog entry on the accepted full detail experience
The walkthrough found that the installed retained Simulator artifact was older than current source. The current source has now been rebuilt and installed without clearing retained data. The shared full-bleed destination container restores a reliable left-edge back gesture across media, episode, and person pages. Shows now begins at Popular on its first render and Movies begins at Newest Available; compact Filter and Sort controls remain in the collection header. Automated UI evidence confirms the full episode actions, Discovery's Add to Watchlist and Open Progress menu actions, keyboard dismissal, the streamlined Settings hierarchy, and the visible Trakt disconnect confirmation. Thomas's rendered retained-data judgment remains the acceptance gate. Physical iPhone behavior and provider writes remain separate. The August 12 retained polish pass also identified two shared presentation corrections and one confirmed provider limitation. Compact cast cards should preserve the top of genuine portrait artwork without changing the accepted full-screen person hero. Movies must describe their Watch Next shelf as movies, not episodes. The Great Food Truck Race Season 19 episodes 3–7 have no exact still from either TMDB or TVmaze. Thomas's latest rendered direction replaces the empty state with genuine exact-season artwork, then genuine show artwork only when the season also has no poster; a later real episode still always wins. Shows Progress presents the latest numbered season first while preserving chronological episode order inside each season. The Xcode 27 build-for-testing and strengthened opt-in retained-data Food Truck audit pass. The audit verifies newest-first Season 19/18 ordering, genuine art for poster-less Seasons 4, 5, 7, and 18, fallback art for Season 19 episodes 3–7, and preservation of exact stills for episodes 1–2. The proof did not trigger or spend on the automatic Encore queue; the approved enabled policy was left unchanged. Thomas approved the Movies wording during this review; this correction did not alter Movies. Step 2 now covers Ask, OpenAI voice, source-grounded Season Encore rewards, personalized discovery, and library intelligence; it cannot be dismissed until Save → Verify → Enable succeeds. First-run error recovery still needs a later physical/onboarding pass.
- Last changed
- August 12, 2026
- Last verified
- August 12, 2026 · retained-data read-only Simulator R&D walkthrough
- Protect
- The accepted Today hierarchy, Cape Fear/Food Truck cinematic detail language, real retained Trakt state and artwork, Discovery ranking evidence, compact library controls, and the existing visible Trakt disconnect correction.
- Evidence
- Current source inspection; Xcode 27 Debug Simulator build; focused joined-navigation, episode-detail, and visible Trakt-disconnect UI regressions; retained-data Today launch without clearing the application container.
Needs proof
Complete physical-iPhone and accessibility acceptance
The latest preserved visual state has not completed a new joined physical-phone pass. Microphone, OpenAI voice, gestures, interruption, routes, direct VoiceOver, and real-device performance require Thomas's separate ready confirmation.
- Last changed
- August 9, 2026
- Last verified
- Simulator evidence only for the latest shared-cause pass
- Protect
- Simulator, physical phone, VoiceOver, audio-route, TestFlight, and live-service evidence remain separate claims.
- Evidence
- Current product baseline and prior phone-review chronology.
Parked
TestFlight and App Store release work
TestFlight begins only after the app is genuinely ready for outside product feedback. Distribution archive, App Store Connect, StoreKit pricing, shared OpenAI funding, and public release remain separate decisions.
- Last changed
- August 9, 2026
- Last verified
- Not a current candidate
- Protect
- Do not turn an old archive, build, or isolated test into release-readiness evidence.
- Evidence
- Release-readiness reset and current product baseline.