Roadmap

This site, the session loop (/start, /end), hygiene and research.

• 10 verified on device
• 1 built, not device-verified

• M1 Blueprint: VISION, ARCHITECTURE, SCORE_TIMELINE_SCHEMA, PREPROCESSOR, ROADMAP L

  Cut point: Docs committed and cross-linked; stack pinned (Flutter, Android first).

  More

  Evidence: f0a390d, 72209da

  Proved by: docs · 2026-06-21

• M2 Sideloadable demo APK (release, debug-signed; no ADB needed) S

  Cut point: Dylan installs it on the S25 without a computer.

  More

  Evidence: 0704cc3 · build/app/outputs/flutter-apk/app-release.apk (v4 on 2026-07-05)

  Proved by: installed; no clear verdict yet (2026-09-20) · 2026-07-04

• M3 Outside product review folded into the docs (6 points) S

  Cut point: Sustained-read exit, loop-fatigue sourcing, Phase-4 gate, H2 hypothesis, H2.5 rung, LFS/OGG flags all in the docs.

  More

  Evidence: 0704cc3

  Proved by: docs · 2026-07-04

• M4 Jev (TypeSafe AI) research + the two-mode decision M Ultracode Fable high S0

  Cut point: Report saved under docs/research; two-mode rule in ARCHITECTURE; JevTagger row in the plan.

  More

  Evidence: docs/research/JEV-2026-09-20.md

  Proved by: docs · 2026-09-20

• M5 Strategy research: soundtrack generators, arrangement, artists, competition, hardware, licensing L Ultracode Fable high S0

  Cut point: Brief saved with verified/refuted claims; decisions written into the plan rows below.

  More

  Evidence: docs/research/STRATEGY-2026-09-20.md

  Proved by: docs · 2026-09-20

• M6 Roadmap site (this) — inventory, derivation, brief, session plan, paper theme L Ultracode Fable high S0

  Cut point: `npm run gate` green; both views render; every session card carries MODEL/EFFORT/ULTRACODE/ULTRA-REVIEW tags and a cut point.

  More

  Evidence: roadmap/ · npm run gate green (23 tests, astro check 0/0/0) · release 20260920192748

  Proved by: gate + https://libretto.tinkr.es/sessions 200 from outside · 2026-09-20

• M7 Management dashboard at /admin (books & licences, stems, competition, research) M Fable high S0

  Cut point: Four data-driven admin pages behind nginx basic auth.

  More

  Evidence: roadmap/src/pages/admin/* · https://libretto.tinkr.es/admin/

  Proved by: /admin/ = 401 without password, 200 with; four pages render · 2026-09-20

• M8 libretto.tinkr.es live: Cloudflare A record, nginx block, certbot, deploy script with neighbour check M Fable high S0

  Cut point: `curl -sI https://libretto.tinkr.es` = 200 from outside; /admin/ = 401 without password; neighbours unchanged.

  More

  Evidence: deploy/deploy-roadmap.sh --cert · cert expires 2026-12-19

  Proved by: curl from outside: / 200, /admin/ 401; neighbours 200/302/200/200 · 2026-09-20

• M9 /start and /end wired to the plan (digest script, brief-first briefing, inventory step, Sonnet cleanup agent) M Fable high S0

  Cut point: `bash .claude/scripts/start-check.sh` prints "Session 1 next" in one call; /end moves rows + regenerates + redeploys.

  More

  Evidence: .claude/scripts/start-check.sh · ~/.claude-personal/commands/{start,end}.md · .claude/agents/cleanup.md

  Proved by: digest prints the brief first in one call · 2026-09-20

• M10 Context hygiene: CLAUDE.md ≤ 6 KB, GOTCHAS.md, one-line memory index, ROADMAP.md as history S Fable high S0

  Cut point: Before/after byte counts in BUILD_LOG; nothing lost (grep-able).

  More

  Evidence: BUILD_LOG 2026-09-20: CLAUDE.md 15,943→4,367 B; MEMORY.md 11,434→1,964 B; docs/GOTCHAS.md 14,118 B

  Proved by: byte counts + grep of moved titles · 2026-09-20

• M11 Standing rule: every question through the AskUserQuestion tool S Fable high S0

  Cut point: One line in the global CLAUDE.md + a feedback memory.

  More

  Evidence: memory/feedback-questions-via-question-tool.md

  Proved by: docs · 2026-09-20

Real beds in the mixer and the on-device verdict that gates everything downstream.

• 4 verified on device
• 4 built, not device-verified
• 8 planned

• SN1 flutter_soloud mixer: all voices alive, 16 ms equal-power crossfade ticker L

  Cut point: A crossfade between two looping stems feels like a score transition on a device.

  More

  Evidence: ffbae74 · lib/mixer/audio_mixer.dart

  Proved by: Pixel 7: Dylan — "the transition feels good" · 2026-06-22

• SN2 Cue sequencer with 1200 ms hysteresis (no mix flapping) S

  Cut point: Jitter test passes; no audible flapping on device.

  More

  Evidence: ffbae74 · lib/mixer/cue_sequencer.dart

  Proved by: jitter test + Pixel 7 · 2026-06-22

• SN3 Mixer debug screen with live per-stem gain bars + source-family selector S

  Cut point: Gains visible live; family switch rebuilds the mixer.

  More

  Evidence: ffbae74, 3358344 · lib/app/mixer_debug_screen.dart

  Proved by: Pixel 7 · 2026-06-22

• SN4 Placeholder stem generator (seam-free synthetic WAV pads) S

  Cut point: 7 loop-safe placeholder beds render and loop without clicks.

  More

  Ugly on purpose; re-voiced caricature-distinct on 2026-07-05 so timing can be judged.

  Evidence: ffbae74, ae225c5 · tool/gen_placeholder_stems.py

  Proved by: Pixel 7 + emulator · 2026-06-22

• SN5 stems.json v2 + source-family A/B harness (placeholder / cc0 / ai, role ids family-agnostic) M

  Cut point: Swapping a family needs zero change in cue/sequencer/mixer code; families A/B live from the debug screen.

  More

  Evidence: 3358344 · lib/stems/stem_bank.dart, assets/stems/stems.json

  Proved by: analyze/test/APK green; never heard with real audio · 2026-06-22

• SN6 Seamless-loop prep tool (equal-power tail→head crossfade) S

  Cut point: Any raw clip becomes a click-free loop.

  More

  Evidence: 3358344 · tool/make_seamless_loop.py

  Proved by: synthetic input only · 2026-06-22

• SN7 Sourcing pack: Freesound CC0 searches + ElevenLabs SFX prompts, D-minor beds S

  Cut point: Dylan can source a full role set from the document alone.

  More

  Evidence: 3358344, 0704cc3 · tool/phase1_real_stems_sourcing.md

  Proved by: doc · 2026-06-22

• SN8 80 scene-level cues for Frankenstein + caricature-distinct placeholder voices M

  Cut point: Mood changes land mid-chapter at story beats; moods sound different from each other.

  More

  Evidence: ae225c5 · tool/gen_frankenstein_timeline.py

  Proved by: emulator: cue flipped mid-chapter; S25 verdict pending · 2026-07-05

• SN9 make_seamless_loop.py: downbeat-snap pre-pass (allin1) + concat mode for multi-take beds M Opus + Sonnet high S1

  Cut point: 2-3 short takes stitch into one 60-120 s bed with loop points on downbeats; tests on synthetic takes.

  More

  allin1 + Demucs are MIT. Never madmom pretrained weights (CC-BY-NC-SA).

• SN10 APK v5 with the four AI families S Opus high S2

  Cut point: Release APK built and handed to Dylan.

• SN11 CC0 family: Dylan's Freesound picks per role, loop-prepped, provenance in stems.json M Needs Dylan Opus high S3

  Cut point: 7 CC0 beds with Freesound ids + licence recorded.

  Waiting on: SN9 · make_seamless_loop.py: downbeat-snap pre-pass (allin1) + concat mode for multi-take beds

• SN12 Source-family A/B verdict on the S25 (CC0 vs best AI family) M Needs Dylan Opus high S3

  Cut point: A family is chosen and the reason written to memory + BUILD_LOG.

  Waiting on: ST1 · Generator shoot-out: Stable Audio 3 Small (local), Stable Audio 2.5 API, Eleven Music v2, ACE-Step — 7 roles each

• SN13 Mood-at-right-beats verdict, round 2 (80-cue timeline on real beds) S Needs Dylan Opus high S3

  Cut point: Dylan confirms changes land at story beats and moods are distinct, or a named fix list exists.

• SN14 Phase 1 exit: 45-60 min sustained read without loop fatigue M Needs Dylan Opus high S3

  Cut point: Dylan reads 45-60 min on the S25 and the score never pulls him out; Phase 1 declared closed.

  More

  THE gate: Phase 4 (SC4+) does not start before this.

• SN15 APK v6 (reader polish + chosen family as default) S Opus + Sonnet high S5

  Cut point: Release APK built and handed to Dylan.

• SN16 APK v7 (library + polish) S Opus + Sonnet high S10

  Cut point: Release APK built and handed to Dylan and the test group.

AI generators, adaptive arrangement, commissioned packs, author-recommended artists.

• 6 planned

• ST1 Generator shoot-out: Stable Audio 3 Small (local), Stable Audio 2.5 API, Eleven Music v2, ACE-Step — 7 roles each L Opus high S2

  Cut point: Families ai_sa3 / ai_sa25 / ai_11m / ai_ace in stems.json, loop-prepped, selectable from the debug screen.

  Waiting on: SN9 · make_seamless_loop.py: downbeat-snap pre-pass (allin1) + concat mode for multi-take beds

  More

  First real AI audio in Libretto — what Dylan disliked so far was the synthetic placeholders. Recipe + do-not-use list in tool/phase1_real_stems_sourcing.md.

• ST2 Key-adherence + cost report per generator (D minor held? USD per bed, minutes per bed) S Opus high S2

  Cut point: docs/research/GENERATORS-<date>.md with a table; spend ≤ 30 USD.

• ST3 ScoreArrangement design (horizontal re-sequencing on SoLoud 4.1 scheduled play + loop points) for artist tracks S Opus xhigh S7

  Cut point: ARCHITECTURE section written: segments (allin1/Demucs) → per-mood loops + transitions; what stays hand-authored.

• ST4 Composer commission #1 (work-for-hire): 8-10 roles, 60-120 s evolving beds, 24-bit/48k WAV, key/tempo sheet, stingers L Needs Dylan Opus high S108

  Cut point: Pack delivered and accepted; contract signed (WFH, derivative-work right explicit); 5-20k USD band.

  Waiting on: SN14 · Phase 1 exit: 45-60 min sustained read without loop fatigue

  More

  Indie-tier rate anchor 400-2,000 USD per finished minute (GameSoundCon 2025).

• ST5 Arrangement of the commissioned pack through the ScoreArrangement pipeline M Opus high S108

  Cut point: Pack becomes a stems.json family and drives a whole book.

  Waiting on: ST3 · ScoreArrangement design (horizontal re-sequencing on SoLoud 4.1 scheduled play + loop points) for artist tracks

• ST6 Author-recommended artist outreach (first author + first artist agreed in principle) M Needs Dylan

  Cut point: One author names an artist; one artist says yes to a scoped pack.

  Waiting on: CA8 · Artist letters: author-recommended artists shortlist + outreach template

EPUB rendering, reading position, comfort, the reader UX (Phases 2-3 follow-through).

• 3 verified on device
• 3 built, not device-verified
• 5 planned

• R1 EPUB reader (flutter_epub_viewer / epub.js), WebView isolated to lib/reader L

  Cut point: Seed EPUB renders and paginates on a device.

  More

  Evidence: b29ed63 · lib/reader/reader_screen.dart

  Proved by: Pixel 7 · 2026-06-23

• R2 ReadingPositionSource keystone: chapterId = spine-CFI prefix, percentThroughBook = epub.js progress M

  Cut point: Position survives a font reflow; hermetic unit tests green.

  More

  Evidence: b29ed63 · lib/position/

  Proved by: Pixel 7: 50.00% / /6/30 held across reflow; 11 tests · 2026-06-23

• R3 Home screen (Mixer | Reader) + mixer bootstrap moved verbatim S

  Cut point: Both screens reachable on device.

  More

  Evidence: b29ed63 · lib/app/home_screen.dart

  Proved by: Pixel 7 · 2026-06-23

• R4 Chapter-anchored cue lookup (`bookFromChapter`) with fallback to book percent M

  Cut point: Conductor prefers chapter geometry when chapter ids match; non-regressive otherwise.

  More

  DORMANT: reader emits CFI-spine ids (/6/30), timeline uses slugs (chapter-5). R7 lights it up.

  Evidence: 12fe1a1

  Proved by: 30 tests · 2026-07-01

• R5 Open from device: file_picker → plain reader for personal EPUBs (no score) S

  Cut point: A picked .epub renders; personal books never enter the repo.

  More

  allowScriptedContent:true is a live surface for user-supplied books → PL8 before H2.

  Evidence: 12fe1a1

  Proved by: analyze + tests · 2026-07-01

• R6 ReaderTheme (bg + ink always together) + reading UX redesign (comfort sheet, score strip, debug toggle) M

  Cut point: Text renders in Dark/Sepia/Light; theme switches live; debug chrome hidden by default.

  More

  Evidence: 3d98829 · lib/reader/reader_theme.dart, test/reader_theme_test.dart

  Proved by: emulator release; regression test · 2026-07-05

• R7 Reader CFI-spine → timeline-slug mapping (lights up the chapter-anchored lookup) M Opus + Sonnet high S5

  Cut point: Conductor takes the chapter path on the real Frankenstein EPUB; verified against its CFIs on device in the next sitting.

• R8 Font-size change debounce S Opus + Sonnet high S5

  Cut point: Rapid font taps no longer race the async reflow.

• R9 Reading-speed smoother on the position source S Opus + Sonnet high S5

  Cut point: A fast page-flip does not jump the mix.

• R10 Default ambient bed for device-opened (unscored) books S Opus + Sonnet high S5

  Cut point: Opening a personal EPUB plays a calm bed instead of silence.

• R11 Small cleanups: `.single` → `.firstOrNull` in open-from-device; Phase-2 debug app bar off the scored screen S Opus + Sonnet high S5

  Cut point: Both gone; analyze + tests green.

The offline preprocessor: parse, tag, merge into scene cues, validate.

• 3 built, not device-verified
• 6 planned

• SC1 Score timeline model + strict loader/validator + Frankenstein seed (Gutenberg #84) M

  Cut point: Timeline JSON validates; `cueAt(book)` unit-tested; golden asset test green.

  More

  Evidence: 1c0c8de · lib/timeline/score_timeline.dart

  Proved by: 8 tests + golden asset · 2026-06-30

• SC2 Score conductor: reading position → cue → mixer through the MixSink seam M

  Cut point: First cue commits immediately; later cues obey the 1200 ms hysteresis; unit-testable without SoLoud.

  More

  Evidence: 1c0c8de · lib/timeline/score_conductor.dart

  Proved by: 3 tests; emulator init chain · 2026-06-30

• SC3 Scored reader screen (session owner: StemBank → mixer → timeline → conductor) M

  Cut point: Reading Frankenstein changes the mix live.

  More

  Evidence: 1c0c8de · lib/app/scored_reader_screen.dart

  Proved by: emulator; S25 pending · 2026-06-30

• SC4 preprocessor/ Python subproject: EPUB parse → paragraphs with chapter + char offsets M Ultra review Ultracode Opus xhigh S6

  Cut point: Frankenstein parses into a paragraph table whose offsets match the reader's chapter geometry.

  Waiting on: SN14 · Phase 1 exit: 45-60 min sustained read without loop fatigue

  More

  Hard-gated on the Phase 1 sonic verdict (review point, 2026-07-04).

• SC5 ModelTagger interface + ClaudeTagger (Opus 4.8 default, Batch API + prompt caching) M Ultra review Ultracode Opus xhigh S6

  Cut point: One chapter tagged {mood, intensity, tempo, confidence} per paragraph; cost logged.

  Waiting on: SN14 · Phase 1 exit: 45-60 min sustained read without loop fatigue

• SC6 Strict schema validation of the generated cue JSON S Opus xhigh S6

  Cut point: Generated output passes the same validator the app loads with.

• SC7 Smoothing + merge of paragraph tags into per-scene cues M Ultracode Opus xhigh S7

  Cut point: Auto cue count and boundaries are in the same order of magnitude as the hand 80-cue timeline; diff reviewed.

  Waiting on: SC5 · ModelTagger interface + ClaudeTagger (Opus 4.8 default, Batch API + prompt caching)

• SC8 Stem + transition assignment and full-timeline validator M Ultracode Opus xhigh S7

  Cut point: A complete auto Frankenstein score-timeline.json loads in-app and drives the mixer on the emulator.

• SC10 Model taste-test: Opus 4.8 vs Haiku 4.5 vs Jev on golden chapters — agreement, calibration, cost, latency L Needs Dylan Opus high S8

  Cut point: Default tagger chosen and written into PREPROCESSOR.md; best auto timeline listened to on device.

  Waiting on: SC8 · Stem + transition assignment and full-timeline validator

Seed library, comfort settings, the whole-book read that closes H1.

• 7 planned

• P1 Two more public-domain books (Gutendex) scored end-to-end M Opus + Sonnet high S9

  Cut point: Two new timelines generated, validated, packaged as assets.

  Waiting on: SC8 · Stem + transition assignment and full-timeline validator

• P2 Library screen: pick a book, resume position M Opus + Sonnet high S9

  Cut point: Three scored books selectable; position restored per book.

• P3 Master volume + mood-intensity control S Opus + Sonnet high S10

  Cut point: Both persist and apply live.

• P4 Sleep timer (fade out, stop) S Opus + Sonnet high S10

  Cut point: Timer fades the mix and stops the engine.

• P5 Crossfade + hysteresis tuning knobs (debug) with sane defaults S Opus + Sonnet high S10

  Cut point: Defaults chosen from the S3 verdict; knobs hidden behind the debug toggle.

• P6 Persisted comfort settings + first-run defaults S Opus + Sonnet high S10

  Cut point: Font, theme, volume survive a restart; first run opens on a sensible page.

• P7 H1 exit: Dylan reads a whole book with Libretto and reaches for it again L Needs Dylan Opus high S12

  Cut point: Verdict recorded in VISION/BUILD_LOG; H1 declared done.

  Waiting on: SN14 · Phase 1 exit: 45-60 min sustained read without loop fatigue

Forks, git-LFS, OGG, scripted-content sandbox, iOS.

• 2 verified on device
• 2 built, not device-verified
• 5 planned

• PL1 Flutter toolchain + Android app skeleton (es.tinkr.libretto) M

  Cut point: Empty app builds, runs on the Pixel 7 emulator, `flutter test` green.

  More

  Evidence: ccee4f8

  Proved by: emulator run + test · 2026-06-22

• PL2 Vendored fork: flutter_inappwebview_android (AGP 9 proguard fix) M

  Cut point: Debug + release APK build on AGP 9.0.1.

  More

  Frozen fork; drop when upstream fixes or AGP pinned to 8.x.

  Evidence: b29ed63 · third_party/flutter_inappwebview_android

  Proved by: APK builds; runs on device · 2026-06-23

• PL3 Vendored fork: file_picker 11.0.2 (AGP 9 + builtInKotlin=false Kotlin fix) S

  Cut point: Release APK builds with file_picker on AGP 9.

  More

  Evidence: 0704cc3 · third_party/file_picker/FORK.md

  Proved by: APK builds · 2026-07-04

• PL4 INTERNET permission in the main manifest (release-only WebView spinner fixed) S

  Cut point: Release APK renders the reader (no eternal spinner).

  More

  Evidence: 3aa8ace

  Proved by: emulator, release build · 2026-07-04

• PL5 git-LFS decision + configuration (or audio masters kept out of the repo, documented) S Opus + Sonnet high S1

  Cut point: A 300 MB WAV drop cannot produce a bad commit; the rule is in CLAUDE.md.

• PL6 OGG-in-app decision + one OGG seamless loop proven on SoLoud S Opus + Sonnet high S1

  Cut point: Decision written (WAV master / OGG shipped or WAV only) with the APK-size numbers behind it.

• PL7 Fork drop check: upstream flutter_inappwebview_android + file_picker on AGP 9 S Opus + Sonnet high S1

  Cut point: Either a fork is dropped, or FORK.md records why not and the next check date.

• PL8 Scripted-content sandbox for user-supplied EPUBs L Ultra review Ultracode Opus xhigh S101

  Cut point: A hostile EPUB fixture cannot run script; user books still render.

  More

  HIGH before any stranger loads a book. allowScriptedContent:true today.

• PL9 iOS platform: flutter create --platforms ios, signing, TestFlight build L Needs Dylan Opus high S102

  Cut point: A TestFlight build installs on an iPhone; WAV/OGG loops verified on iOS.

  More

  Needs an Apple developer account and a cloud Mac.

Public-domain sources, licensing partners, the legal edges of "score your own books".

• 5 planned

• C1 Admin Books table fed from the same book manifest the app ships S Opus + Sonnet high S9

  Cut point: One manifest, two readers (app + /admin/books).

• C2 Catalog content model: licence, source, provenance, timeline version per title M Opus + Sonnet high S105

  Cut point: Every catalog title carries a machine-checked licence + source; Gutendex import script.

• C3 Licensing partner scoping: PublishDrive / Ingram CoreSource / Draft2Digital — deal shape, revenue share, DRM obligations M Needs Dylan Opus high S109

  Cut point: A written comparison and one opened conversation.

• C4 Soundstripe app-embedding licence quote (the one stock catalog built for apps) S Needs Dylan Opus high S109

  Cut point: Quote in hand; go/no-go against a commission.

• C5 Legal read: format-shifting / personal-use scoring under Costa Rica Law 6683, US DMCA §1201, EU S Needs Dylan

  Cut point: A one-page note by a lawyer; user-facing claims adjusted.

  More

  DRM stripping is never a path; "Open from device" stays scoped to files the user legally holds DRM-free.

Device render, media, test-user approvals, the campaign itself.

• 10 planned

• CA1 Friends & family test group installs APK v7 S Needs Dylan Opus high S11

  Cut point: At least 3 people reading with Libretto and reporting.

  Waiting on: SN16 · APK v7 (library + polish)

• CA2 Two books driven to "100% approved" by Dylan and the test group L Needs Dylan Opus high S11

  Cut point: Fix list → fixes → re-approval; both books signed off.

  More

  Campaign gate, half 1.

• CA3 Industrial-design brief for the H3 reader (form, display, camera/IR placement, audio) M Needs Dylan Opus + Sonnet high S13

  Cut point: One-page brief approved by Dylan.

  Waiting on: G2 · Gaze accuracy log + go/no-go note for the device concept

• CA4 3D render / mockup of the device L Needs Dylan Opus + Sonnet high S13

  Cut point: Render set (front, side, in-hand) approved by Dylan.

  More

  Campaign gate, half 2.

• CA5 Marketing media: stills + a short video of the app and the render M Needs Dylan Opus + Sonnet high S13

  Cut point: Media folder ready for a campaign page and social posts.

• CA6 Landing section on libretto.tinkr.es (public-facing, not the roadmap) S Opus + Sonnet high S13

  Cut point: A visitor understands the product in 20 seconds.

• CA7 Campaign page draft + reward tiers M Needs Dylan Opus high S14

  Cut point: Kickstarter draft with tiers, goal 2-8k USD for one commissioned stem pack.

  Waiting on: CA4 · 3D render / mockup of the device

• CA8 Artist letters: author-recommended artists shortlist + outreach template M Needs Dylan Opus high S14

  Cut point: Five names with a one-page pitch each; rights reality (sync + master) explained in the letter.

• CA9 Campaign goal + timeline decision S Needs Dylan Opus high S14

  Cut point: Goal, launch window (H2 beta) and fallback written down.

• CA10 Crowdfunding campaign launch L Needs Dylan

  Cut point: Campaign live at H2 beta with the approved books and render.

  Waiting on: CA2 · Two books driven to "100% approved" by Dylan and the test group · CA4 · 3D render / mockup of the device · CA7 · Campaign page draft + reward tiers

Store, accounts, payments, onboarding, legal, launch.

• 8 planned

• PU1 Accounts + sync (positions, library) L Ultra review Opus xhigh S104

  Cut point: Sign in on two devices; position and library follow.

• PU2 Catalog + scored packs (public-domain seed library, downloadable timelines) M Opus + Sonnet high S105

  Cut point: Catalog lists ≥ 10 scored public-domain books.

  Waiting on: P2 · Library screen: pick a book, resume position

• PU3 Payments + "score your own books" subscription (Play Billing / App Store / Stripe web) L Needs Dylan Ultra review Ultracode Opus max S106

  Cut point: A test purchase unlocks scoring; refunds + cancellation paths proven.

  Waiting on: O3 · Online mode: "score as you read" chapter-ahead tagging with offline fallback + app mode switch · PU1 · Accounts + sync (positions, library)

• PU4 Onboarding flow M Opus + Sonnet high S107

  Cut point: A stranger reaches a scored page within 60 s of install.

• PU5 Store assets (icon, screenshots, listing copy) S Needs Dylan Opus + Sonnet high S107

  Cut point: Assets approved by Dylan.

• PU6 Privacy policy, terms, support channel M Needs Dylan Opus + Sonnet high S107

  Cut point: Published pages; support inbox answered.

• PU7 Public beta + launch checklist M Needs Dylan Opus high S110

  Cut point: Play internal test → open beta; checklist all green.

  Waiting on: PL8 · Scripted-content sandbox for user-supplied EPUBs · PU1 · Accounts + sync (positions, library)

• PU8 Trademark "Libretto" (name + logo) S Needs Dylan

  Cut point: Filing submitted.

"Score as you read" for unscored books; the preprocessor as a service.

• 3 planned

• O1 JevTagger: second ModelTagger over TypeSafe Jev (Choice on the mood enum, Score for intensity) M Needs Dylan Opus high S8

  Cut point: Tags a golden chapter through the Jev API with calibrated confidence per paragraph.

  Waiting on: SC5 · ModelTagger interface + ClaudeTagger (Opus 4.8 default, Batch API + prompt caching)

  More

  Needs Dylan's Jev waitlist key. Cloud-only, text-only; never in the live reading loop.

• O2 Scoring service: the preprocessor as an authenticated, rate-limited VPS endpoint with a job queue L Ultra review Ultracode Opus xhigh S103

  Cut point: An EPUB posted to the endpoint returns a validated timeline; abuse limits proven.

  Waiting on: SC8 · Stem + transition assignment and full-timeline validator

• O3 Online mode: "score as you read" chapter-ahead tagging with offline fallback + app mode switch L Ultra review Ultracode Opus xhigh S103

  Cut point: An unscored EPUB gets a score while reading; airplane mode still reads with the default bed.

  Waiting on: O2 · Scoring service: the preprocessor as an authenticated, rate-limited VPS endpoint with a job queue

Phone-camera gaze as a ReadingPositionSource; the gate before hardware money.

• 5 planned

• G1 Camera-gaze smoke test: MediaPipe iris → regression → coarse gaze → throwaway CameraGazePositionSource feeding ScoreConductor L Ultracode Opus xhigh S4

  Cut point: Gaze drives a cue change end-to-end on video, on the S25 or a laptop webcam.

  More

  Moved up from H2.5 on 2026-09-20: cheap, no purchase, shapes the device render. Frames never leave the device.

• G2 Gaze accuracy log + go/no-go note for the device concept S Opus xhigh S4

  Cut point: Measured error (cm / lines) at reading distance written to docs/research.

• G3 Gaze + page-estimate fusion, calibration flow, confidence gating L Ultracode Opus max S201

  Cut point: Fused position beats the page estimate alone on a 20-minute read; falls back cleanly in bad light.

  Waiting on: G1 · Camera-gaze smoke test: MediaPipe iris → regression → coarse gaze → throwaway CameraGazePositionSource feeding ScoreConductor

• G4 On-device reading-behaviour heuristics (skim / re-read / pause) → intensity nudges M Opus high S202

  Cut point: Nudges are audible, subtle, and computed entirely on-device.

  Waiting on: G3 · Gaze + page-estimate fusion, calibration flow, confidence gating

• G5 H2.5 verdict sitting: does gaze → score feel magical? M Needs Dylan Opus high S203

  Cut point: Verdict written; H3 capital gate opened or closed.

  Waiting on: G4 · On-device reading-behaviour heuristics (skim / re-read / pause) → intensity nudges

The eye-tracking e-reader. Epics only until H2.5 de-risks the thesis.

• 8 planned

• D1 Android e-ink reader as test device (Boox Go 7 / Palma 2, or Hisense A9 for a front camera) + APK on e-ink M Needs Dylan Opus high S204

  Cut point: Libretto reads on e-ink with acceptable refresh and audio out.

  More

  Kindle jailbreak ruled out for Flutter (2026-09-20): flutter-pi needs DRM/KMS.

• D2 OS decision: gaze on a coprocessor (embedded Linux) vs on the main SoC (Android/AOSP) XL Needs Dylan

  Cut point: Decision memo with power, cost and vision-stack trade-offs; prototypes stay on Android until then.

  Waiting on: G5 · H2.5 verdict sitting: does gaze → score feel magical?

• D3 Device prototype: IR gaze module, display, audio, battery XL Needs Dylan

  Cut point: A bench prototype reads a book with gaze-driven score.

  Waiting on: D2 · OS decision: gaze on a coprocessor (embedded Linux) vs on the main SoC (Android/AOSP)

  More

  Ground truth: Pupil Labs Core (~3.6k EUR) or Gazepoint GP3 HD (~2.7k USD). Consumer Tobii is not licensed for gaze data.

• D4 ODM RFQs (RK3566-class e-ink boards) + open-hardware references (Open Book Touch, Modos Flow) XL Needs Dylan

  Cut point: Three quotes with MOQ and NRE; one reference board on the bench.

  Waiting on: D2 · OS decision: gaze on a coprocessor (embedded Linux) vs on the main SoC (Android/AOSP)

• D5 Word-level cue anchoring (gaze precision → cue timing on the dot) XL Needs Dylan

  Cut point: Cues fire within one line of the intended word on the device.

  Waiting on: D3 · Device prototype: IR gaze module, display, audio, battery

• D6 On-device reflex classifier over gaze features (only if a local System-One / open-weight model exists) XL Needs Dylan

  Cut point: A local model classifies reading state under 20 ms on the device; no frame leaves it.

  Waiting on: D3 · Device prototype: IR gaze module, display, audio, battery

  More

  Conditional epic from the Jev research: cloud reflex models are the wrong shape for this loop.

• D7 Funding (crowdfunding proceeds, loan, or investment) for a production run XL Needs Dylan

  Cut point: Capital secured for the first run.

  Waiting on: CA10 · Crowdfunding campaign launch · D3 · Device prototype: IR gaze module, display, audio, battery

• D8 Manufacturing + fulfilment of the first run XL Needs Dylan

  Cut point: Devices in backers' hands.

  Waiting on: D7 · Funding (crowdfunding proceeds, loan, or investment) for a production run