Libretto
Sessions
48 numbered sessions · 112 rows scheduled · 7 unnumbered H3 epics
H1 — soundtrack lab
Exact session numbers. Ends when Dylan reads a whole book with Libretto and reaches for it again.
21 sessions · 72 items · session 5 next
-
Session 0
finished M + L + L + M + M + M + S + S- M4 Jev (TypeSafe AI) research + the two-mode decision M
Cut point: Report saved under docs/research; two-mode rule in ARCHITECTURE; JevTagger row in the plan.
- M5 Strategy research: soundtrack generators, arrangement, artists, competition, hardware, licensing L
Cut point: Brief saved with verified/refuted claims; decisions written into the plan rows below.
- M6 Roadmap site (this) — inventory, derivation, brief, session plan, paper theme L
Cut point: `npm run gate` green; both views render; every session card carries MODEL/EFFORT/ULTRACODE/ULTRA-REVIEW tags and a cut point.
- M7 Management dashboard at /admin (books & licences, stems, competition, research) M
Cut point: Four data-driven admin pages behind nginx basic auth.
- M8 libretto.tinkr.es live: Cloudflare A record, nginx block, certbot, deploy script with neighbour check M
Cut point: `curl -sI https://libretto.tinkr.es` = 200 from outside; /admin/ = 401 without password; neighbours unchanged.
- M9 /start and /end wired to the plan (digest script, brief-first briefing, inventory step, Sonnet cleanup agent) M
Cut point: `bash .claude/scripts/start-check.sh` prints "Session 1 next" in one call; /end moves rows + regenerates + redeploys.
- M10 Context hygiene: CLAUDE.md ≤ 6 KB, GOTCHAS.md, one-line memory index, ROADMAP.md as history S
Cut point: Before/after byte counts in BUILD_LOG; nothing lost (grep-able).
- M11 Standing rule: every question through the AskUserQuestion tool S
Cut point: One line in the global CLAUDE.md + a feedback memory.
- M4 Jev (TypeSafe AI) research + the two-mode decision M
-
Session 1
finished S + S + M + S- PL5 Audio-master policy: masters out of git (no LFS) + a pre-commit budget guard S
Cut point: A 300 MB WAV drop cannot produce a bad commit; the rule is in CLAUDE.md.
- PL6 Shipping format decided: OGG Vorbis q4, with a placeholder/placeholder_ogg format A/B in the app S
Cut point: Decision written (WAV master / OGG shipped or WAV only) with the APK-size numbers behind it.
- SN9 make_seamless_loop.py: downbeat-snap pre-pass (allin1) + concat mode for multi-take beds M
Cut point: 2-3 short takes stitch into one 60-120 s bed with loop points on downbeats; tests on synthetic takes.
- PL7 Fork drop check: file_picker fork dropped (upstream 13.1.0), inappwebview fork kept with a FORK.md S
Cut point: Either a fork is dropped, or FORK.md records why not and the next check date.
- PL5 Audio-master policy: masters out of git (no LFS) + a pre-commit budget guard S
-
Session 2
finished L + S + S + S + S- ST1 Generator shoot-out: Stable Audio 3.0, Stable Audio 3 Small (local), Eleven Music v2, ACE-Step — 7 roles each L
Cut point: Families ai_sa30 / ai_sa3 / ai_11m / ai_ace in stems.json, loop-prepped, selectable from the debug screen.
- ST2 Key-adherence + cost report per generator (D minor held? USD per bed, minutes per bed) S
Cut point: docs/research/GENERATORS-<date>.md with a table; spend ≤ 30 USD.
- SN10 APK v5 with the four AI families S
Cut point: Release APK built and handed to Dylan.
- SN19 Pop-free crossfades: sine-law equal-power entry + audio-thread volume ramps S
Cut point: Dylan steps cues on the S25 and hears no pop or lurch at a cue change.
- ST8 Blind A/B listening bench for stem families S
Cut point: Families compared blind, sides shuffled per role, verdict recorded as evidence beside the measurements.
- ST1 Generator shoot-out: Stable Audio 3.0, Stable Audio 3 Small (local), Eleven Music v2, ACE-Step — 7 roles each L
-
Session 3
finished M + M + L + M + S- ST10 Per-book score design: Dungeon Crawler Carl (world, characters, emotions → stem roles, key, cue vocabulary) M
Cut point: docs/research/SCORE-DCC.md names the stem roles (places, characters, System AI, combat), the key/tonal plan, the prompts, and where the book-derived files live on the device; Dylan approves it.
- ST11 DCC stem family on Eleven Music v2: generated, loop-prepped, seam- and third-checked M
Cut point: A book family (e.g. dcc_11m) exists via generate_beds.py with every seam +0 frames and no bed carrying a clashing third; spend stays inside the 30 USD cap.
- SC13 DCC score timeline: scene cues over the owned EPUB, chapter + char-offset anchored L
Cut point: A score-timeline.json for DCC book 1 validates against the schema, with per-scene cues using the ST10 roles; it lives outside git and loads with the book.
- R14 Runtime score pack: an opened EPUB finds its own timeline + stem family M
Cut point: Opening DCC from the device plays its own score on the S25; a book with no pack opens with the default family and no cues, never a crash.
- ST12 Re-roll on a clashing third: generate_beds.py --max-tries keeps the first bed that layers S
Cut point: --max-tries N re-rolls a bed whose third reads major or both, keeps the first that layers (else the least F-sharp), and a real family ships through it.
- ST10 Per-book score design: Dungeon Crawler Carl (world, characters, emotions → stem roles, key, cue vocabulary) M
-
Session 4
finished M + L- R15 Score packs off the APK: a book pack carries its timeline AND its beds, pushed to the phone M
Cut point: dcc_11m leaves assets/ and the APK; push_score.py pushes timeline + beds + a pack manifest to the app files dir; StemBank loads the family from disk and DCC plays on the S25 as it does today.
- SN20 Streaming mixer: only the current and next cue beds are loaded, from disk L
Cut point: The mixer loads/streams just the beds the current and the upcoming cue need (not every bed of the family); a 32-bed pack keeps native heap near today (~150 MB on the S25); the loop seam and pop-free crossfades are re-proven by ear on the S25.
- R15 Score packs off the APK: a book pack carries its timeline AND its beds, pushed to the phone M
-
Session 5
next up M + M + L- ST13 DCC score v2 design: an environment bed per place + a feel bed per emotion/character M
Cut point: SCORE-DCC.md v2 lists every distinct place in book 1 (counted from the text, ~15-25) and ~12 feel beds (the 8 moods + awe, triumph, character motifs) with their prompts and lengths (env 3 min, feel 4 min); Dylan approves it.
- ST14 DCC v2 bed library on Eleven Music: ~32 beds, third- and tonal-centre-gated M
Cut point: Every ST13 bed exists via generate_beds.py (--max-tries), seams +0 frames, no clashing third, tonal beds D-minor rank <= 6; spend inside the 60 USD cap (estimate ~18 USD with re-rolls).
- SC14 DCC v2 timeline: place + feel per beat, about twice the beats, the place layer steady L
Cut point: The book is re-read with a two-axis brief; every cue carries one environment and one feel bed; the environment changes only when the place does; the timeline validates in Python and in the app parser.
Waiting on:
ST14· DCC v2 bed library on Eleven Music: ~32 beds, third- and tonal-centre-gated
- ST13 DCC score v2 design: an environment bed per place + a feel bed per emotion/character M
-
Session 6
S + S + M + S + S + M + S- ST15 DCC v2.1 place beds: 20 pure field-recording ambiences on Stable Audio 3.0 S
Cut point: All 20 DCC2_ENV beds re-made from the v2.1 prompts (no drone, no music, 3 min) on Stable Audio 3.0, imported with generate_beds.py --generator import, levelled, seams +0, published in dcc2_11m.
Waiting on:
ST14· DCC v2 bed library on Eleven Music: ~32 beds, third- and tonal-centre-gated - ST16 DCC feel beds that fit: system, show, defiance, respite and guide re-briefed and re-made S
Cut point: The five feels that scored 1-2 on the blind bench get new prompts and Stable Audio 3.0 takes; a re-run of the bench scores each at 3 or better.
Waiting on:
ST14· DCC v2 bed library on Eleven Music: ~32 beds, third- and tonal-centre-gated - SN22 Moment cues: one-shot sounds fired on their sentence (a blast, a door shaking) M
Cut point: The timeline carries moments (anchor char offset + one-shot file + gain); the conductor fires each once when its anchor comes into view; the mixer plays one-shots over the beds; unit-tested and heard on the S25 on at least 3 DCC moments.
Waiting on:
SC14· DCC v2 timeline: place + feel per beat, about twice the beats, the place layer steady - SC15 DCC v2.1 heard on the S25: place-first, quiet, not repetitive, and fitting the passage S
Cut point: Dylan reads DCC on the S25 with the v2.1 pack (ST15, ST16, SN22 in) and no longer calls it repetitive or distracting; a blind bench re-run beats 2.92.
Waiting on:
SC14· DCC v2 timeline: place + feel per beat, about twice the beats, the place layer steady ·ST15· DCC v2.1 place beds: 20 pure field-recording ambiences on Stable Audio 3.0 ·ST16· DCC feel beds that fit: system, show, defiance, respite and guide re-briefed and re-made ·SN22· Moment cues: one-shot sounds fired on their sentence (a blast, a door shaking) - SN17 OGG q4 vs WAV listening A/B on the S25 (placeholder vs placeholder_ogg) S
Cut point: Dylan cannot pick the OGG family out on the S25 - or q6 is adopted and the budget in docs/AUDIO_ASSETS.md is updated.
- SN14 Phase 1 exit: 45-60 min sustained read without loop fatigue M
Cut point: Dylan reads 45-60 min on the S25 and the score never pulls him out; Phase 1 declared closed.
Waiting on:
SC15· DCC v2.1 heard on the S25: place-first, quiet, not repetitive, and fitting the passage - SC11 score-timeline schema extension: point/moment cues S
Cut point: score-timeline.json schema extended with a point/moment cue type (chapter + char-offset, one-shot stinger reference, no duration); validator updated; documented in SCORE_TIMELINE_SCHEMA.md as planned-then-implemented.
- ST15 DCC v2.1 place beds: 20 pure field-recording ambiences on Stable Audio 3.0 S
-
Session 7
L + S + S + L- G1 Camera-gaze smoke test: MediaPipe iris → regression → coarse gaze → throwaway CameraGazePositionSource feeding ScoreConductor L
Cut point: Gaze drives a cue change end-to-end on video, on the S25 or a laptop webcam.
- G2 Gaze accuracy log + go/no-go note for the device concept S
Cut point: A 2-minute real-reading session, hand-labeled, scored on three tiers — line-transition recall ≥80%/≤20% false-positive, region-level (correct third of the page) ≥75%, stretch line-level ±1 ≥60% — and gaze must beat the time-based ReadingPositionSource on the same session.
- G8 Patent/FTO skim: gaze-based reading-position tracking S
Cut point: Patent/FTO skim on gaze-based reading-position tracking, referencing docs/GOTCHAS.md's "not ownable, don't spend on copyrighting the idea" finding; confirms it still holds now that gaze is the pivot's core mechanism, logged under docs/research.
- G9 Dark-room IR gaze spike: Arducam+IR LEDs, darkness + glasses, before the desk prototype locks a camera L
Cut point: Same tiered bar as G2 (region ≥75%, line-transition ≥80%/≤20% FP, stretch line ±1 ≥60%), re-measured with an Arducam OV9281 (global-shutter, no IR-cut) + 850/940nm IR LEDs rig in near-darkness + a dim front light, with and without glasses (single-vision, AR-coated, progressive); 850nm-vs-940nm (provisional until the G12 sun test) and camera+IR duty-cycle-at-rest choices logged; non-camera alternatives (EOG, near-eye photodiode, mmWave, thermal, event cameras) recorded as considered and rejected for a handheld at 30-45cm.
Waiting on:
G2· Gaze accuracy log + go/no-go note for the device concept
- G1 Camera-gaze smoke test: MediaPipe iris → regression → coarse gaze → throwaway CameraGazePositionSource feeding ScoreConductor L
-
Session 8
M + S + S + S + S + S- R7 Reader CFI-spine → timeline-slug mapping (lights up the chapter-anchored lookup) M
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
Cut point: Rapid font taps no longer race the async reflow.
- R9 Reading-speed smoother on the position source S
Cut point: A fast page-flip does not jump the mix.
- R10 Default ambient bed for device-opened (unscored) books S
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
Cut point: Both gone; analyze + tests green.
- SN15 APK v6 (reader polish + chosen family as default) S
Cut point: Release APK built and handed to Dylan.
- R7 Reader CFI-spine → timeline-slug mapping (lights up the chapter-anchored lookup) M
-
Session 9
M + M + S- SC4 preprocessor/ Python subproject: EPUB parse → paragraphs with chapter + char offsets M
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 - SC5 ModelTagger interface + ClaudeTagger (Opus 4.8 default, Batch API + prompt caching) M
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
Cut point: Generated output passes the same validator the app loads with.
- SC4 preprocessor/ Python subproject: EPUB parse → paragraphs with chapter + char offsets M
-
Session 10
M + M + S- SC7 Smoothing + merge of paragraph tags into per-scene cues M
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
Cut point: A complete auto Frankenstein score-timeline.json loads in-app and drives the mixer on the emulator.
- ST3 ScoreArrangement design (horizontal re-sequencing on SoLoud 4.1 scheduled play + loop points) for artist tracks S
Cut point: ARCHITECTURE section written: segments (allin1/Demucs) → per-mood loops + transitions; what stays hand-authored.
- SC7 Smoothing + merge of paragraph tags into per-scene cues M
-
Session 11
M + L- O1 JevTagger: second ModelTagger over TypeSafe Jev (Choice on the mood enum, Score for intensity) M
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) - SC10 Model taste-test: Opus 4.8 vs Haiku 4.5 vs Jev on golden chapters — agreement, calibration, cost, latency L
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
- O1 JevTagger: second ModelTagger over TypeSafe Jev (Choice on the mood enum, Score for intensity) M
-
Session 12
M + M + S- P1 Two more public-domain books (Gutendex) scored end-to-end M
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
Cut point: Three scored books selectable; position restored per book.
- C1 Admin Books table fed from the same book manifest the app ships S
Cut point: One manifest, two readers (app + /admin/books).
- P1 Two more public-domain books (Gutendex) scored end-to-end M
-
Session 13
S + S + S + S + S + S + L- PL10 Per-ABI APK split for the sideload build S
Cut point: The APK Dylan sideloads is roughly halved with no asset change and still installs on the S25.
- P3 Master volume + mood-intensity control S
Cut point: Both persist and apply live.
- P4 Sleep timer (fade out, stop) S
Cut point: Timer fades the mix and stops the engine.
- P5 Crossfade + hysteresis tuning knobs (debug) with sane defaults S
Cut point: Defaults chosen from the S4 verdict; knobs hidden behind the debug toggle.
- P6 Persisted comfort settings + first-run defaults S
Cut point: Font, theme, volume survive a restart; first run opens on a sensible page.
- SN16 APK v7 (library + polish) S
Cut point: Release APK built and handed to Dylan and the test group.
- PL8 Scripted-content sandbox for user-supplied EPUBs L
Cut point: A hostile EPUB fixture cannot run script; user books still render.
- PL10 Per-ABI APK split for the sideload build S
-
Session 14
S + L- CA1 Friends & family test group installs APK v7 S
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
Cut point: Fix list → fixes → re-approval; both books signed off.
- CA1 Friends & family test group installs APK v7 S
-
Session 15
L- P7 H1 exit: Dylan reads a whole book with Libretto and reaches for it again L
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
- P7 H1 exit: Dylan reads a whole book with Libretto and reaches for it again L
-
Session 16
S- CA6 Landing section on libretto.tinkr.es (public-facing, not the roadmap) S
Cut point: A visitor understands the product in 20 seconds.
- CA6 Landing section on libretto.tinkr.es (public-facing, not the roadmap) S
-
Session 17
M- CA8 Artist letters: author-recommended artists shortlist + outreach template M
Cut point: Five names with a one-page pitch each; rights reality (sync + master) explained in the letter.
- CA8 Artist letters: author-recommended artists shortlist + outreach template M
-
Session 18
M- R12 Renderer char-offset<->screen-rect map M
Cut point: For the current layout (font size, margins), the reader exposes a char-offset<->screen-rect map so a screen coordinate (tap or, later, gaze) resolves to a word and its char-offset; verified across at least two font sizes on device.
Waiting on:
P7· H1 exit: Dylan reads a whole book with Libretto and reaches for it again
- R12 Renderer char-offset<->screen-rect map M
-
Session 19
M + M + S- SC12 Preprocessor moment tagging (sparingly) M
Cut point: Preprocessor tags moment candidates sparingly (a handful per book, not per paragraph) on a golden chapter; over-tagging flagged as a defect, not a feature.
Waiting on:
SC11· score-timeline schema extension: point/moment cues ·SC5· ModelTagger interface + ClaudeTagger (Opus 4.8 default, Batch API + prompt caching) - SN18 Stinger asset family + one-shot mixer layer M
Cut point: A stinger asset family exists (source + license recorded like any stem) and the mixer plays a one-shot stinger over the current bed with no crossfade, without disturbing the bed's own loop.
Waiting on:
P7· H1 exit: Dylan reads a whole book with Libretto and reaches for it again - R13 Phone-side tap-to-trigger prototype for moment cues S
Cut point: Phone-side debug mode: tapping a word near a tagged moment fires its stinger over the bed, using the time-based position estimate + R12's map — no gaze hardware involved. Dylan judges whether the moment-cue feel is worth building gaze precision for.
Waiting on:
R12· Renderer char-offset<->screen-rect map ·SC11· score-timeline schema extension: point/moment cues ·SN18· Stinger asset family + one-shot mixer layer
- SC12 Preprocessor moment tagging (sparingly) M
-
Session 20
L- G12 Sunlight IR gaze test: park/beach midday, 940nm + bandpass + pulsed differencing, before the camera locks L
Cut point: G9's tiered bar (region ≥75%, line-transition ≥80%/≤20% FP) re-measured outdoors in direct midday sun, open shade and a bright park/beach setting, with the rig hardened for sun: 940nm LEDs + a matched narrow bandpass filter, short global-shutter exposure with LED pulses synced to it, LED-on/LED-off frame differencing to subtract ambient IR; with and without glasses; sunglasses tested and, where they block IR, detected with a clean fallback. When confidence drops (glare, squint, sun on the lens) position falls back to the time-based ReadingPositionSource and moment cues go fail-silent, verified on the same session. Final 850-vs-940nm choice logged with both dark (G9) and sun evidence.
Waiting on:
G9· Dark-room IR gaze spike: Arducam+IR LEDs, darkness + glasses, before the desk prototype locks a camera
- G12 Sunlight IR gaze test: park/beach midday, 940nm + bandpass + pulsed differencing, before the camera locks L
Desk prototype
Exact session numbers. The existing app as embedded-Linux device software on a Pi 5, a real e-ink panel driven by hand, the gaze rig on real hardware, phone<->device pairing.
12 sessions · 17 items · session 300 next
-
Session 300
next up M + S- D11 Real APK sideloaded on a borrowed AOSP e-ink device (OS-answer test) M
Cut point: Real APK sideloaded on any cheap/borrowed AOSP e-ink device; settles Android/AOSP vs. a custom Flutter e-ink embedder; no gaze hardware or G2 required.
- D24 Open-hardware e-ink reader build write-ups read end to end (PineNote, Open Book Touch, Modos Flow) S
Cut point: All three build write-ups read end to end; lessons logged to docs/research.
- D11 Real APK sideloaded on a borrowed AOSP e-ink device (OS-answer test) M
-
Session 301
L- D9 Pi 5 + flutter-pi: flutter_soloud mixer proven off-phone L
Cut point: The flutter_soloud mixer plays a real crossfade through BT/speaker on a Pi 5 running flutter-pi, off the phone, zero changes to lib/mixer.
Waiting on:
P7· H1 exit: Dylan reads a whole book with Libretto and reaches for it again
- D9 Pi 5 + flutter-pi: flutter_soloud mixer proven off-phone L
-
Session 302
M- D10 Waveshare IT8951 e-Paper HAT driven directly, refresh cadence policy M
Cut point: Own script pushes frames to the Waveshare HAT over USB/SPI; full- vs. partial-refresh latency logged; a refresh-cadence policy picked.
Waiting on:
P7· H1 exit: Dylan reads a whole book with Libretto and reaches for it again
- D10 Waveshare IT8951 e-Paper HAT driven directly, refresh cadence policy M
-
Session 303
S + S- D12 Combined-load check: audio + e-ink refresh proven simultaneously on one Pi 5 S
Cut point: Audio and e-ink refresh proven running simultaneously on one Pi 5; SPI/CPU/power contention logged.
Waiting on:
D9· Pi 5 + flutter-pi: flutter_soloud mixer proven off-phone ·D10· Waveshare IT8951 e-Paper HAT driven directly, refresh cadence policy - D13 Phone app pairs with the Pi rig over BT/USB S
Cut point: Phone app pairs with the Pi rig over BT/USB; phone can see the device's current book/position.
Waiting on:
D9· Pi 5 + flutter-pi: flutter_soloud mixer proven off-phone
- D12 Combined-load check: audio + e-ink refresh proven simultaneously on one Pi 5 S
-
Session 304
M + L- D14 Pi+HAT+battery+speaker in a printed PETG shell, off-battery form/fit M
Cut point: Pi+HAT+battery+speaker close inside a printed PETG shell, run off the Stage-2 LiPo pack; form/fit only.
Waiting on:
D9· Pi 5 + flutter-pi: flutter_soloud mixer proven off-phone ·D10· Waveshare IT8951 e-Paper HAT driven directly, refresh cadence policy ·P7· H1 exit: Dylan reads a whole book with Libretto and reaches for it again - D2 Android/Boox-sideload vs. custom Flutter e-ink embedder, ratified L
Cut point: Decision memo with power, cost and vision-stack trade-offs, ratified against D11's sideload results; prototypes stay on Android until then.
Waiting on:
D11· Real APK sideloaded on a borrowed AOSP e-ink device (OS-answer test)
- D14 Pi+HAT+battery+speaker in a printed PETG shell, off-battery form/fit M
-
Session 305
L- G3 Gaze + page-estimate fusion, calibration flow, confidence gating L
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
- G3 Gaze + page-estimate fusion, calibration flow, confidence gating L
-
Session 306
M- G4 On-device reading-behaviour heuristics (skim / re-read / pause) → intensity nudges M
Cut point: Nudges are audible, subtle, and computed entirely on-device.
Waiting on:
G3· Gaze + page-estimate fusion, calibration flow, confidence gating
- G4 On-device reading-behaviour heuristics (skim / re-read / pause) → intensity nudges M
-
Session 307
M- G5 Verdict sitting: does gaze → score feel magical on real hardware? M
Cut point: Verdict written; feeds the D21 pcb-horizon go/no-go (does not replace it).
Waiting on:
G4· On-device reading-behaviour heuristics (skim / re-read / pause) → intensity nudges
- G5 Verdict sitting: does gaze → score feel magical on real hardware? M
-
Session 308
M- G6 Gaze accuracy re-measured with the Arducam+IR rig on the Pi+HAT rig M
Cut point: Same tiered bar as G2/G9 (region ≥75%, line-transition ≥80%/≤20% FP), re-measured with the Arducam+IR rig on the Pi+HAT rig at real reading distance; glasses pass included; gaze still beats the time-based ReadingPositionSource on the same session.
Waiting on:
G2· Gaze accuracy log + go/no-go note for the device concept ·D10· Waveshare IT8951 e-Paper HAT driven directly, refresh cadence policy ·G9· Dark-room IR gaze spike: Arducam+IR LEDs, darkness + glasses, before the desk prototype locks a camera ·G12· Sunlight IR gaze test: park/beach midday, 940nm + bandpass + pulsed differencing, before the camera locks
- G6 Gaze accuracy re-measured with the Arducam+IR rig on the Pi+HAT rig M
-
Session 309
M- D25 Real APK + gaze rig on a purchased Boox tablet (gaze-on-hardware sideload) M
Cut point: Real APK + gaze rig on a purchased Boox tablet; full chapter reads with the score changing live and gaze assisting position, screenshots/video captured.
Waiting on:
G2· Gaze accuracy log + go/no-go note for the device concept ·P7· H1 exit: Dylan reads a whole book with Libretto and reaches for it again
- D25 Real APK + gaze rig on a purchased Boox tablet (gaze-on-hardware sideload) M
-
Session 310
M + L- CA3 Industrial-design brief for the reader device (form, display, camera/IR placement, audio) M
Cut point: One-page brief approved by Dylan.
Waiting on:
D10· Waveshare IT8951 e-Paper HAT driven directly, refresh cadence policy ·D14· Pi+HAT+battery+speaker in a printed PETG shell, off-battery form/fit - CA4 3D render / mockup of the device L
Cut point: Render set (front, side, in-hand) approved by Dylan.
Waiting on:
D10· Waveshare IT8951 e-Paper HAT driven directly, refresh cadence policy ·D14· Pi+HAT+battery+speaker in a printed PETG shell, off-battery form/fit
- CA3 Industrial-design brief for the reader device (form, display, camera/IR placement, audio) M
-
Session 311
M + S- G10 Armed-cue gaze mode: idle-to-armed ramp, fail-silent, 150ms word-level bar M
Cut point: The tracker idles at low rate, ramps to 30-60Hz within a few lines of a marked moment, and fires a stinger only on a confident fixation on the target word. Fail-silent verified (low confidence skips the moment, never guesses); word-level bar measured as the share of armed moments firing within ~150ms of the first fixation on the target word, with zero early fires across the test session.
Waiting on:
G6· Gaze accuracy re-measured with the Arducam+IR rig on the Pi+HAT rig ·R13· Phone-side tap-to-trigger prototype for moment cues - G11 Dozing detection: eyes-closed fade-out and position save S
Cut point: Eyes closed for N seconds fades the soundtrack out and the reading position is remembered; N tuned so a normal blink or paragraph re-read never triggers it.
Waiting on:
G6· Gaze accuracy re-measured with the Arducam+IR rig on the Pi+HAT rig
- G10 Armed-cue gaze mode: idle-to-armed ramp, fail-silent, 150ms word-level bar M
PCB + enclosure
Exact session numbers. First custom board, first real case, first vendor conversations — the real go/no-go before manufacturing money.
6 sessions · 8 items · session 400 next
-
Session 400
next up M + S + S- D15 First KiCad board routed and submitted to fab M
Cut point: A simple 2-layer board routed in KiCad, ERC/DRC clean, submitted to JLCPCB.
- D17 PMIC/rail-sequencing literacy spike S
Cut point: Rail-sequencing diagram for the chosen panel exists, checked against the vendor datasheet.
- D20 Good Display/E Ink contacted for the waveform file + hobbyist pricing S
Cut point: Good Display/E Ink emailed re: waveform file + hobbyist-quantity pricing; reply or non-reply logged.
- D15 First KiCad board routed and submitted to fab M
-
Session 401
M- D16 Charger + fuel-gauge bring-up on the desk prototype M
Cut point: BQ24074 + MAX17048 wired, SOC reporting over I2C on the desk prototype.
Waiting on:
D9· Pi 5 + flutter-pi: flutter_soloud mixer proven off-phone
- D16 Charger + fuel-gauge bring-up on the desk prototype M
-
Session 402
L- D18 First integrated custom PCB rev fabbed and bring-up tested L
Cut point: First integrated custom PCB rev (fuel gauge + charger + IT8951 breakout) fabbed and bring-up tested, using D20's waveform file and D17's rail plan.
Waiting on:
D15· First KiCad board routed and submitted to fab ·D16· Charger + fuel-gauge bring-up on the desk prototype ·D17· PMIC/rail-sequencing literacy spike ·D20· Good Display/E Ink contacted for the waveform file + hobbyist pricing
- D18 First integrated custom PCB rev fabbed and bring-up tested L
-
Session 403
M- D19 Enclosure CAD v1, dimensioned to the real board+panel+battery M
Cut point: Enclosure CAD v1, dimensioned to the real board+panel+battery, SLA-printed.
Waiting on:
D14· Pi+HAT+battery+speaker in a printed PETG shell, off-battery form/fit ·D18· First integrated custom PCB rev fabbed and bring-up tested
- D19 Enclosure CAD v1, dimensioned to the real board+panel+battery M
-
Session 404
M- G7 IR gaze module on the custom board, IEC 62471 Exempt Group documented M
Cut point: IR gaze module integrated on the custom board; LED choice, drive current, optics and duty cycle documented against IEC 62471, targeting Exempt Group — the same classification Face ID, Windows Hello and all-day Tobii accessibility trackers sit in — including the G12 pulsed outdoor mode.
Waiting on:
D18· First integrated custom PCB rev fabbed and bring-up tested ·G6· Gaze accuracy re-measured with the Arducam+IR rig on the Pi+HAT rig
- G7 IR gaze module on the custom board, IEC 62471 Exempt Group documented M
-
Session 405
L- D21 PCB-horizon exit: hand-built unit reads a full session, gaze-assisted, on battery L
Cut point: A hand-built unit — custom PCB rev 1 + SLA/FDM case — runs a full reading session, with gaze assist, on battery.
Waiting on:
D18· First integrated custom PCB rev fabbed and bring-up tested ·D19· Enclosure CAD v1, dimensioned to the real board+panel+battery ·G7· IR gaze module on the custom board, IEC 62471 Exempt Group documented
- D21 PCB-horizon exit: hand-built unit reads a full session, gaze-assisted, on battery L
Shelved — public app, deferred
Provisional numbers, on hold. Store, accounts, online mode, payments — resumes only if the pcb gate fails or nearer-term revenue is needed.
9 sessions · 15 items · session 102 next
-
Session 102
next up L- PL9 iOS platform: flutter create --platforms ios, signing, TestFlight build L
Cut point: A TestFlight build installs on an iPhone; WAV/OGG loops verified on iOS.
- PL9 iOS platform: flutter create --platforms ios, signing, TestFlight build L
-
Session 103
L + L- O2 Scoring service: the preprocessor as an authenticated, rate-limited VPS endpoint with a job queue L
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
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
- O2 Scoring service: the preprocessor as an authenticated, rate-limited VPS endpoint with a job queue L
-
Session 104
L- PU1 Accounts + sync (positions, library) L
Cut point: Sign in on two devices; position and library follow.
- PU1 Accounts + sync (positions, library) L
-
Session 105
M + M- PU2 Catalog + scored packs (public-domain seed library, downloadable timelines) M
Cut point: Catalog lists ≥ 10 scored public-domain books.
Waiting on:
P2· Library screen: pick a book, resume position - C2 Catalog content model: licence, source, provenance, timeline version per title M
Cut point: Every catalog title carries a machine-checked licence + source; Gutendex import script.
- PU2 Catalog + scored packs (public-domain seed library, downloadable timelines) M
-
Session 106
L- PU3 Payments + "score your own books" subscription (Play Billing / App Store / Stripe web) L
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)
- PU3 Payments + "score your own books" subscription (Play Billing / App Store / Stripe web) L
-
Session 107
M + S + M- PU4 Onboarding flow M
Cut point: A stranger reaches a scored page within 60 s of install.
- PU5 Store assets (icon, screenshots, listing copy) S
Cut point: Assets approved by Dylan.
- PU6 Privacy policy, terms, support channel M
Cut point: Published pages; support inbox answered.
- PU4 Onboarding flow M
-
Session 108
L + M- ST4 Composer commission #1 (work-for-hire): 8-10 roles, 60-120 s evolving beds, 24-bit/48k WAV, key/tempo sheet, stingers L
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 - ST5 Arrangement of the commissioned pack through the ScoreArrangement pipeline M
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
- ST4 Composer commission #1 (work-for-hire): 8-10 roles, 60-120 s evolving beds, 24-bit/48k WAV, key/tempo sheet, stingers L
-
Session 109
M + S- C3 Licensing partner scoping: PublishDrive / Ingram CoreSource / Draft2Digital — deal shape, revenue share, DRM obligations M
Cut point: A written comparison and one opened conversation.
- C4 Soundstripe app-embedding licence quote (the one stock catalog built for apps) S
Cut point: Quote in hand; go/no-go against a commission.
- C3 Licensing partner scoping: PublishDrive / Ingram CoreSource / Draft2Digital — deal shape, revenue share, DRM obligations M
-
Session 110
M- PU7 Public beta + launch checklist M
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)
- PU7 Public beta + launch checklist M
H3 — epics
Small-batch manufacturing. Unnumbered on purpose: these stay epics until the pcb-horizon exit (D21) opens the funding gate and they can be split into real sessions.
7 epics · no session numbers
- 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· Android/Boox-sideload vs. custom Flutter e-ink embedder, ratified - 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 shipped device.
Waiting on:
D21· PCB-horizon exit: hand-built unit reads a full session, gaze-assisted, on batteryMore
The armed-cue mechanism (fail-silent, ~150ms bar) is already prototyped at G10 (hw) and re-verified at G7/D21 (pcb); this epic carries it onto the manufactured run.
- 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:
D21· PCB-horizon exit: hand-built unit reads a full session, gaze-assisted, on batteryMore
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 ·D21· PCB-horizon exit: hand-built unit reads a full session, gaze-assisted, on 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 runMore
MOQ asymmetry named explicitly: boards scale down cheaply (50-500 units), the molded case does not (500-1,000). Certification: FCC Part 15C pre-cert radio module (cheap) vs custom RF; EU CE-RED plus the 2026 Cyber Resilience Act cybersecurity gate, which can turn a ~1,500 EUR self-assessment into a 60,000-120,000+ EUR Notified Body program — see D22.
- D22 FCC pre-cert module chosen; EU CRA scope-and-cost read M Needs Dylan
Cut point: FCC Part-15C pre-cert module chosen; EU CRA scope-and-cost read.
Waiting on:
D21· PCB-horizon exit: hand-built unit reads a full session, gaze-assisted, on battery - D23 Small-batch contract manufacturers vetted, RFQs sent M Needs Dylan
Cut point: 2-3 small-batch CMs vetted, RFQs sent.
Waiting on:
D21· PCB-horizon exit: hand-built unit reads a full session, gaze-assisted, on battery
The jump in the numbers is deliberate: H1 sessions run 1–99, the desk prototype (hw) starts at 300 and PCB+enclosure (pcb) at 400, so a session number always says which stage it belongs to and each stage can grow without renumbering anything downstream. H2 (101–199) is shelved, not deleted — its numbers stay reserved in case the public app resumes.