TODOS

Notebook editor follow-ups

Verify near-limit editing on Android API 26

What: Run the Notebook document's 10 KiB and 256 KiB input, recovery, and serialization benchmark on an Android API 26 emulator using the minimum-supported System WebView profile.

Why: The issue 5917 implementation intentionally uses a real-Chromium performance gate, which does not prove that near-limit typing meets the same latency budget on Readest's oldest supported Android runtime.

Context: The browser gate requires 50 operations with p95 below 16 ms at 10 KiB and below 50 ms at 256 KiB, with no operation above 100 ms. Reuse that workload on API 26 and move recovery storage away from synchronous localStorage if the device profile misses the budget.

Effort: S Priority: P2 Depends on: Issue 5917 Notebook editor implementation

Cloud Sync provider selection follow-ups (deferred by /autoplan, 2026-07-06)

Deferred from the Cloud Sync provider-selection plan (#4959/#4380). See the Decision Audit Trail in the plan for reasoning.

• Pre-switch "download all Readest Cloud books" affordance so a fresh device gets full library completeness when a third-party provider is selected. (S)
• Library-page sync-status indicator for the active third-party provider (fileSyncStore already exposes aggregate progress). (S)
• Account page active-provider chip. (XS)
• File-engine parity: reading stats + per-book viewSettings sync via the file layout (readingStatus/tags parity shipped as its own PR). (M)
• Server-side quota error code (code: 'quota_exceeded', mirroring the share import route) so the client stops string-matching 'Insufficient storage quota'; message drift silently restores retry behavior. (S)
• Pre-existing: Manage Sync "Books" category gates metadata rows but NOT binary uploads to Readest Cloud (queueUpload never consults isSyncCategoryEnabled('book') despite the category docs claiming it) — align behavior or docs. (S)
• Sentry cloudSyncProvider tag: Sentry tagging is Rust-mediated (set_webview_info pattern in sentry_config.rs); add a set_cloud_sync_provider command + before_send tag so sync-related reports carry the active provider. Console log lines ship in the meantime. (S, needs src-tauri)

Deferred items from the Edge TTS Web Audio plan review (/autoplan, 2026-07-04). Each was explicitly deferred, not forgotten — see the Decision Audit Trail in .agents/plans/2026-07-03-edge-tts-webaudio.md.

TTS listening engine follow-ups

• Cross-section gapless playback: preload and schedule the next section's first sentence so chapter boundaries are as seamless as sentence boundaries. (M)
• Lock-screen ±10s seek offsets in addition to prev/next sentence. (S)
• Persist measured sentence durations per book so a reopened chapter starts with an exact timeline instead of estimates. (S)
• Worker offload for decode + WSOLA if device profiling shows main-thread jank on low-end Android. (S)
• Provider-agnostic voice source hedge: local neural TTS (e.g. Piper/Kokoro WASM) plugging into WebAudioPlayer/SectionTimeline — the engine is designed for this; see "Strategic framing" in the plan. (L)
• Background chapter prefetch (convert timeline estimates to exact durations ahead of playback). (M)

Self-hosted premium TTS follow-ups (deferred by /autoplan, 2026-07-08)

Deferred from the Readest Voice plans (spec: docs/superpowers/specs/2026-07-08-selfhosted-premium-tts-design.md). See the Decision Audit Trail in .agents/plans/2026-07-08-readest-tts-integration.md.

• Dict-popup premium pronunciation: route the dictionary speak button through /api/tts/readest for eligible plans (reuses the wire contract as-is). (M)
• Voice preview samples in the TTS voice picker (pre-generated clips per voice, static hosting; no GPU call on tap). (M)
• Qwen3 instruct-based emotion/style control as a Pro setting (API already supports instruct; needs UX + catalog plumbing). (M)
• Warm-worker peak-hours schedule (RunPod min-workers=1 on a cron) if cold-start complaints materialize; premise-4 mitigation. (S)
• Client-side cold-start abandonment telemetry (did the user give up before first audio?) to complete the day-60 review data. (S)
• Comparative MOS-style voice regression harness vs Edge/native/managed APIs (post-demand; the launch gate is a one-shot A/B listen). (M)
• On-device Kokoro exploration for offline premium voices (desktop first; 82M params runs near-realtime on CPU) — if accepted as taste decision, this moves into a phase-2 plan instead. (L)
• Cancel in-flight RunPod jobs on client abort: forward AbortSignal through the route and call /cancel/{id} so a stop during a cold start does not burn a full GPU job. V1 keeps signal-less fetches deliberately (in-flight dedup shares one promise across preload and playback, Edge parity). (S)
• Extend the TTS browser e2e harness with a mocked /api/tts/readest so the premium engine's playback + highlight path gets automated coverage. (M)
• Per-user in-flight request cap (1-2) on /api/tts/readest to complement the char quota (needs shared state on Workers: KV or DO). (M)
• Delegate uncovered-language marks to Edge TTS instead of skipping them (cross-engine per-sentence delegation; v1 skips the sentence). (M)
• End-to-end request id: client generates one per sentence, route forwards it in the RunPod input, worker logs it — full-path correlation in one grep. (S)
• Generate the shared voice catalog (voices.json + catalogVersion) from readestVoices.ts instead of hand-syncing the worker fixture. (S)