# Roadmap

Ideas and planned enhancements. Grouped by area. No commitment to order or timeline.
Items marked ~~strikethrough~~ have been implemented.

Last refreshed at v1.0.1 (2026-08-09).

---

## 1.0 GA Milestone (from 1.0-rc audit, 2026-05-25)

The app went public as v1.0.0-rc.1+ and is functionally complete. These
are the polish items that separate "RC quality" from a confident GA tag.
None are strictly blocking, but the items listed here (test suite +
accessibility cleanup in particular) are worth clearing when practical.
The `-rc` suffix was retired in v1.0.0 as a versioning-scheme change,
independent of these polish items.

### Test coverage (highest leverage)

LiftTrace has **zero automated test coverage** today. Every release is
"build, hope, release." This is the single biggest GA blocker.

Concrete target: a 5-test Playwright smoke suite covering the core flow:

- Login → /diary visible
- Log a workout: add exercise, complete a set, finish → summary appears
- View statistics → chart renders, no console errors
- Create a program → appears in list
- NutriTrace federation save+test (mocked) → "Connected" pill appears

Wire into GitHub Actions on every PR. Roughly a day to scaffold + write.

### ~~i18n extraction to 55-60% coverage~~ *(done and exceeded)*

Original target was ~340 keys total (from a 192 baseline). `src/i18n/en.json` now carries 1232 top-level keys — the sweep landed and then some. Further i18n work moves to the general "Full coverage" pass planned for v1.1.0 (see [feedback_i18n_full_sweep_cadence.md] in memory).

### Onboarding tour (first-run overlay)

Wizard finish now shows a brief "You're All Set" celebration (rc.4) but the diary still drops new users into a UI they've never seen. An overlay walkthrough, "this is where you log sets, this is your active program, this is Trace", would lift retention. Pattern: stepped tooltip system keyed on first-launch flag in `localStorage`. ~1 day.

### Accessibility cleanup

About 20 `svelte-ignore a11y-click-events-have-key-events` blocks live across Radio, Diary, Trace FAB. Each is a clickable `<div>` that lacks keyboard reach. Mechanical replacement with `<button>` + `on:keydown` handlers, plus visible focus indicators on the custom widgets. Matters more if targeting F-Droid or accessibility-conscious self-hoster lists. Roughly half a day.

### Smaller deferred audit items

- **Mobile / responsive at <360px**: Settings.svelte has no explicit mobile max-width container; on very narrow viewports some setting rows push trailing controls off-screen. Most fixes are local; one breakpoint pass covers it.
- **`min-width` audit**: exercise sort menu (`160px`), some radio cards, fixed-height progress SVG on ExerciseDetail (`100px`). All cosmetic but noticeable on edge-case viewports.
- **Loading spinners inside settings sub-sections**: most of the route loading states landed a shared `<Spinner>` component in rc.4; a second pass to use it inside long-running setting actions (sync now, full backup create) would round out the consistency.
- **Per-OIDC-provider verified indicator**: SettingsAuth has a per-provider `testProvider` action, but no persistent "verified" pill on the provider card. A small green check/red X next to the issuer URL would clarify state at a glance.

---

## Workouts & Sets

### ~~Warm-up set generator~~ *(done, v0.9.4-beta.13)*
Bar → 50% → 70% → 85% ramp; manual button + opt-in `autoGenerate` on template load; `warmup` flag propagated through all stats math so warm-ups don't inflate volume totals.

### ~~RPE / RIR per set~~ *(done, v0.9.4-beta.13)*
Opt-in chip + picker on SetRow, stored as `set.rpe`.

### ~~Rest timer rebuild~~ *(done, v0.9.4-beta.5→9)*
Persistent store, absolute `endTime`, countdown beeps at 3/2/1/0 with vibration, background Service Worker notification.

### ~~Selectable rest-timer tones~~ *(done, v0.9.4-beta.13)*
5 synth presets (Classic, Bells, Beeper, Gym, Minimal) with preview button.

### ~~Superset rest timer "Next: X" label~~ *(done, v0.9.4-beta.13)*
Shows correct next exercise (first in group) after a superset round, not the one just completed.

### ~~Auto-collapse default ON + persisted collapse state~~ *(done, v0.9.4-beta.5→9)*
3-state model (`collapsed` / `expanded` / null) respects explicit user intent across navigation.

### ~~Drag auto-scroll during diary reorder~~ *(done, v0.9.4-beta.5→9)*
Edge-zone RAF loop, `behavior: 'instant'` override of global smooth-scroll.

### Collapsed-card summary
When an ExerciseCard auto-collapses after completing all sets, show the best working set on the header (e.g. `Bench Press · 4/4 · 225×5`) instead of just the name.

### Cardio-specific fields on the set row
Distance / pace / heart rate on the SetRow itself. Useful for users mixing conditioning with lifting. Related to Bundle B (BLE HR) but broader.

---

## Programs & Templates

### Template auto-suggest
Suggest today's template based on program rotation and last workout date.

---

## Statistics

### ~~Body weight trend~~ *(done)*
### ~~Progressive overload auto-fill~~ *(done)*
### ~~Calendar heatmap~~ *(done)*
### ~~1RM~~ *(done)*
### ~~Plate calculator~~ *(done)*

### RPE trend chart
"Average RPE at working weight" line chart per exercise. Useful for spotting accumulated fatigue.

### Weekly summary, next steps ([#98](https://github.com/TraceApps/lifttrace/issues/98))
The summary now reports sessions against the weekly goal, sets, volume, time trained, PRs and sets per muscle group, each against the average week over the month before, with a link into Statistics. Still to come:
- **Translated emails.** Server emails are English for every locale (the numbers already follow the language setting). A family-wide job, since CookTrace and NutriTrace have the same gap.
- **Choosing what goes in.** Per-user picks for which metrics and which comparison, before any editable template.
- **Program-aware comparison.** Week 3 of this block against week 3 of the last one, from the `program_id` and `program_week` already stamped on every workout.
- **Admin-set schedule.** An instance-wide default day and time, the way scheduled backups are set.
- **An exact week link.** Statistics only has rolling ranges, so the email opens the last seven days rather than that calendar week.

---

## Exercises & Equipment

### ~~Custom exercises~~ *(done, v0.9.4-beta.5→9)*
Create / edit / delete with media picker (upload / paste URL / YouTube), "Custom" chip, Settings management section.

### ~~Custom equipment + multi-select "Available today" filter~~ *(done, 2026-06-29 dev/main)*
Exercises filter chip strip is now multi-select. User-defined equipment strings (Slackboard, Sandbag, etc.) get a dashed-border chip alongside the six normalized buckets when at least one exercise uses them. Persists via SERVER_SETTINGS (`customEquipment`).

### ~~Similar exercises~~ *(done, v0.9.4-beta.13)*
"Similar exercises" grid on ExerciseDetail, primary-muscle-overlap scoring.

### ~~Per-exercise JSON share + URL importer~~ *(done, 2026-06-29 dev/main)*
Single-exercise export from ExerciseDetail; importer accepts file pick or URL paste (github.com/blob auto-rewrites to raw.githubusercontent.com). Strips install-local fields on export, re-creates as `source='custom'` on import. Sets up the awesome-lifttrace-exercises community-repo pattern with no in-app machinery.

### Orphan upload cleanup
When a custom exercise is deleted, delete its `/uploads/exercises/<uuid>.ext` file too. Minor disk-leak for heavy users.

---

## Trace AI (LiftBot)

### ~~Smart Add (workout edition)~~ *(done, v0.9.x)*
Text + voice parse via LiftBot, hold-to-record on the FAB.

### ~~LiftBot data access~~ *(done, v0.9.x)*
Live context injected every chat.

### ~~LiftBot FAB frequency visualizer ring~~ *(done, v0.9.4-beta.5→9)*
32 SVG bars around the robot head, Web Audio `AnalyserNode`.

### ~~Trace FAB visualizer parity on Android~~ *(done, v1.0.0-rc.6)*
RENDER_LERP 0.08 → 0.35 so the frequency ring tracks music dynamics with the same amplitude as the PWA.

### Trace deload suggestions
Proactively suggest a deload week when a user has logged RPE 9+ across three consecutive workouts on the same lift.

---

## Import & Migration

Already supported: Strong, Hevy, FitNotes, Jefit.

### ~~Workout-history import~~ *(done, v0.9.4-beta.10→12)*
Strong + Hevy + FitNotes + Jefit CSV adapters, two-step preview → commit, skip/replace on duplicate dates.

### ~~Garmin FIT file import~~ *(done, issue [#6](https://github.com/TraceApps/lifttrace/issues/6))*
Adapter live at `server/lib/workout-import/garmin-fit.js` via `@garmin/fitsdk`. Wired into the import preview / commit flow through `server/routes/workout-import.js`. Maps Garmin's `exercise_category` enum to LT's exercise catalog via `garmin-exercise-map.js`.

### Apple Health / Fitness workouts
XML export from Health app; summary-level only (workout duration + heart rate), no set-level data.

### Google Fit / Health Connect
Less common for lifting, but possible.

### Fitbod, Gymshark Conquer, Gravitus
Lower priority, smaller userbases.

### RPE back-fill from imported notes
The Strong / Hevy importers currently stash RPE into set notes as `"RPE 8"`. Promote those to the real `set.rpe` field with a one-shot migration now that RPE exists.

---

## Body Stats & Photos

### ~~Body stats~~ *(done, v0.9.x)*

### ~~Progress photos timeline~~ *(done)*
Dated photos on their own `/progress` page, reached from Statistics and the Body Stats sheet, with a drag-to-compare before/after view. Photos never expire; they are the long-term record the feature exists for. Coaching clips (#57) are a separate concern with a separate lifetime, see that issue.

### Photo notes
A free-text note per photo, not per date. The deciding case is the pose label ("front relaxed", "side flexed"): the table deliberately allows several photos on one date, and a pose is inherently per-photo, while a date-level note ("start of cut") degrades perfectly well into a per-photo one. The implication only runs one way, so the photo is the right owner, and one notes field beats two.

Deliberately free text rather than a structured pose picker for now. A taxonomy chosen before anyone has formed a habit is a UI plus a migration betting on a guess, and free text stays fuzzy-matchable if smart compare (offering like against like) ever earns its place.

Bigger than the one nullable column suggests: notes are the first thing that makes a photo row editable, so this adds an update verb across the app route, the MCP tool and the REST route, plus the sync-push and backup-restore column lists. Full plan in `docs/plans/photo-notes.md`.

---

## External Integrations

### ~~NutriTrace federation~~ *(done, `write:workouts` scope)*
LT posts workouts into NT's wellness history via bearer-scoped `/api/v1/*` federation.

### Model Context Protocol (MCP) server
Expose LiftTrace to external AI clients (Claude Desktop, Cursor, Codex, VS Code) the same way NutriTrace does starting with `v1.2.0-dev01`. Voice / chat "log a set post-workout" and "how many reps at 225 last month?" are the natural fits.

Off by default; opt in with `MCP_ENABLED=1` and a new `mcp:read` scope on the token. Write and destructive tiers gated behind `MCP_WRITE_ENABLED=1` + `mcp:write` and `MCP_DESTROY_ENABLED=1` + `mcp:destroy`, each destructive call also needing `confirm: true`.

Natural tool set (~11):

- **Read**: `get_current_program`, `get_workout_history`, `get_exercise_prs`, `get_body_stats`, `list_exercises`
- **Write**: `log_workout`, `log_set`, `log_body_stat`, `add_cardio_entry`
- **Destructive**: `delete_workout`, `edit_set`, `create_custom_exercise`

Wait until NT's MCP has real usage feedback before porting so any design tweaks land in NT first. Roughly 2-4 days; see NT's `server/lib/mcp/` layout for the exact pattern.

---

## Radio (audio during workouts)

### ~~Streaming radio stations~~ *(done, v0.9.x)*

### ~~HLS radio station support~~ *(done, v0.9.4-beta.5→9)*
Lazy-loaded `hls.js` chunk for `.m3u8` streams (TuneIn-style).

### ~~Radio-Browser.info Browse sub-tab~~ *(done, v0.9.4-beta.5→9)*
Search ~40k stations, tap to add.

### ~~ICY `StreamTitle` now-playing~~ *(done, v0.9.4-beta.5→9)*
Proxy parses interleaved metadata, client polls `/now-playing`, displays in mini-player + full player + station list.

### ~~Station groups, rename, in-group reorder~~ *(done, v0.9.4-beta.10→12)*
Datalist suggestions, drag-and-drop cross-group, bulk-rename pencil on group header, up/down arrows for in-group.

### ~~/api/radio-proxy~~ *(done, v0.9.4-beta.5→9)*
Routes station streams through the server so the Web Audio visualizer isn't CORS-silenced.

### ~~Radio station metadata hardening~~ *(done, v0.9.4-beta.10→12)*
UTF-8 / Latin-1 fallback, HTML entity decode, ad-marker suppression, Icecast `status-json.xsl` + Shoutcast `/stats?json=1` fallback fetchers.

### ~~RDS Italia `Song*` + HLS ID3 (`TIT2`/`TPE1`) parsers~~ *(done, v0.10.0-beta.6)*
Three sanitizers (Java/JS/server) all recognize asterisk-delimited and ID3-tagged broadcaster formats.

### ~~PWA stream RDS on lockscreen~~ *(done, v0.10.0-beta.6)*
Now-playing poll pushes parsed title/artwork into `navigator.mediaSession.metadata` so the OS lockscreen reflects each song change.

### Additional providers

Current: Subsonic (Navidrome/Airsonic/Gonic), Jellyfin, Plex, Emby, HLS / Icecast / Shoutcast streaming stations via Radio-Browser.info directory.

- **Funkwhale**: Subsonic-compatible (may already work) + own API
- **Ampache**: Subsonic-compatible + Ampache API
- **Koel**: own REST API
- **LMS (Lyrion)**: Logitech Media Server API
- **Multiple servers**: browse multiple providers simultaneously (Symphonium-style)

---

## Android App (Capacitor)

### ~~Unified Android playback on Media3 ExoPlayer~~ *(done, v0.10.0-beta.6)*
Radio + library both flow through `RadioPlayerPlugin` + `RadioPlaybackService`. One MediaSession, one notification UX, lockscreen swaps Prev/Next/Stop layout dynamically. Drops `capacitor-music-controls-plugin`. PWA path unchanged.

### ~~Offline support~~ *(done, v0.9.x)*

### ~~Notifications / scheduler~~ *(done, v0.9.x)*

### ~~PWA rest timer notifications~~ *(done, v0.10.0-beta.6)*
Request `Notification.requestPermission()` from `startRest` (a real user gesture) instead of `_finish`'s setTimeout (silently no-ops).

### Native Android voice input
Current `SmartLogModal` voice path uses Web Speech API; reliability inside Android WebView is patchy. Switch to `@capacitor-community/speech-recognition` when a session lines up to touch that surface.

---

## Sync & Multi-Device

### ~~Cross-device workout save race fix~~ *(done, v1.0.0-rc.6)*
Editing the same workout from phone + PWA near-simultaneously could clobber the other device's just-completed sets with a stale in-memory snapshot; the workout store now refetches and merges before writing.

### ~~NutriTrace-style user management~~ *(done, v0.9.4-beta.10→12)*
Sign-out moved to sidebar footer, "My Profile" shortcut in User Management section, Settings top is clean.

---

## Export & Sharing

### ~~CSV workout export~~ *(done, 2026-06-29 dev/main)*
Long-format CSV (one row per set) downloads from WorkoutSummary. PWA path is a direct Blob download; Android writes to Cache + opens Share intent. Unilateral L/R splits become two rows.

### ~~Share workout card~~ *(done, v0.9.x)*
PNG render from completion summary.

### Per-entity CSV/JSON export
Workouts.csv, programs.json, exercises.csv separate from the existing full-backup ZIP. Workout CSV landed in rc.6 for one workout at a time; this is the batch / all-entity story.

---

## UI / UX Polish

### ~~Animated banner redesigned~~ *(done, v1.0.0-rc.6)*
Illustrated SVG banners retired across every page. Setting Banner Style to Animated now paints the same compact accent bar as Gradient plus a subtle motion effect (Shimmer / Drift / Pulse / Aurora, all honour Reduce Motion). Reclaims ~40px of vertical real estate.

### ~~Password manager password-generation fix~~ *(done, v1.0.0-rc.6)*
`passwordrules` attribute added to all 11 password-CREATION inputs so browsers and password managers generate passwords satisfying the LT policy.

### ~~Diary layout polish~~ *(done, v0.9.4-beta.15→16)*
"Now doing" pill, current-set highlight, progress fill on summary bar, dedupe workout name.

### ~~Settings copy consistency~~ *(done, v0.9.4-beta.15→16)*
Title Case for section names, sentence case for setting rows.

### Keyboard shortcuts overlay
`?` opens a cheatsheet listing Space / Arrow / etc. Radio player already has keyboard bindings; surface them.

---

## Distribution & Licensing

### Paid Android app (future)

Speculative. LiftTrace is currently free everywhere (self-host + free signed APK from the release page). This section captures what would need doing IF a paid Play Store / F-Droid distribution ever happens.

- **Attribution page / About panel** listing every active exercise / media / library source + its license (CC-BY-SA 4.0 for wger, Unlicense for free-db, AGPL for OSS ExerciseDB + hls.js, commercial TOS for RapidAPI, Radio-Browser.info attribution). AscendAPI requires visible credit for any project using their OSS dataset.
- **Build-tier flag** (e.g. `VITE_BUILD_TIER='commercial'`) that filters `SOURCES[]` so the paid Android build does not expose `oss.exercisedb.dev`. Their terms explicitly prohibit commercial/monetised use of the OSS endpoint.
- **Explicit warning in the wizard** when a user picks the OSS source: "Personal / non-commercial use only. Safe for self-hosters; paid-app users should use RapidAPI instead."
- **Prebuild a starter SQLite** in CI with just Free Exercise DB (~870 exercises, Public Domain) + its images, bundle into the Android APK assets. First-launch instantly useful with zero network.
- **Local GIF mirror option**: download GIFs to `/uploads/exercisedb/{id}.gif` at import time so URL rotation by upstream can't break the app. 150-300 MB disk, included in full backups. Toggle in Settings → Exercise Catalog.
- **Weekly scheduled re-import** for commercial ExerciseDB sources (URLs rotate every Monday 00:00 UTC per AscendAPI's caching guide). Low priority, the OSS + free sources don't rotate.
- **Commercial dataset snapshot**: buying the v1 dataset outright from AscendAPI's shop so a paid APK could bundle rich GIFs without an ongoing API dependency. One-time cost vs. subscription.

---

## Engagement / Achievements

### Achievements / badges
"100 workouts logged", "First 3-plate bench", "30-day streak", etc. Motivational layer, opt-in.

---

## Offline

### Offline PWA editing (family-wide, after the next main release)

Bring NoteTrace's browser offline model here, so the installed web app works in a
dead zone the way the Android app does. This is set for every Trace app once the
next main release is out.

The pattern to copy (NoteTrace `src/lib/offline-api.js` + `offline-edits.js`):
Workbox caches the app shell, an IndexedDB mirror answers reads when the server
can't be reached, an outbox holds edits and shows them at once, new rows get
temporary ids that are mapped to real ids after they go up, and the queue is sent
through the existing sync push endpoint so the server merges browser edits exactly
as it merges the phone's. A Web Lock stops two tabs sending at once and a
BroadcastChannel keeps them in step.

Deliberately NOT the Service Worker Background Sync API: Safari doesn't support
it, and iOS is the main reason for the work. Flush from the page instead, with
retries backing off from 3s to 30s plus `online` events.

The sidebar sync pill and the amber / red colour rule are already in place here,
so the state has somewhere to show.

iOS caveats to plan for: Safari evicts site data after about 7 days of no use
unless the PWA is on the Home Screen, so queued-but-unsent work needs to be
visible and installing needs a nudge.

LiftTrace specifics:

- Cover workouts, sets, exercises, programs, and settings.
- Still online afterwards: media uploads, Trace AI, radio, and anything admin.
- First slice: log a workout offline, which is the whole point in a gym basement
  with no signal.

---

## Tech debt

### Emby provider `/emby/*` route split
Today the Emby library-browsing methods delegate to `jellyfin.*` via `.apply(jellyfin, args)`, which means they hit `/api/subsonic/provider/jf/*` routes on the server (with Emby's URL in radioUrl). Works because Emby is a Jellyfin fork with API-compatible endpoints and both auth with `X-Emby-Token`. Would break if Emby ever diverges from Jellyfin's `/Items`-style routes. Fix = duplicate the jellyfin methods into the emby object with `/emby/` prefixes, or thread a `_prefix` parameter through `_proxyJson`.

### Android radio HE-AAC decoder workaround
Chromium WebView's direct `<audio src=...>` path on Android hits a deterministic `PIPELINE_ERROR_DECODE` at exactly stream-timestamp 5461333μs (packet #128) on iHeart-format HE-AAC streams (Z100, KIIS, etc.), same micro on every reconnect. Likely an SBR/PS feature boundary issue in Chromium's media pipeline. Current mitigation (v0.10.0-beta.x onward): WebView intercept proxy adds synthetic CORS headers, stream plays via direct `audio.src` for ~5s then errors; player.js auto-recovers via removeAttribute+load+new src cycle, giving continuous-with-gaps audio. Real fix paths to explore: (a) direct MSE pipeline packaging raw AAC ADTS into fragmented MP4 via mux.js; (b) native MediaCodec decode → MP3 re-encode through the proxy; (c) separate Android audio playback path via Java MediaPlayer / ExoPlayer as a Capacitor plugin (loses Web Audio visualizer for streams).

### `openid-client 5 → 6`
Only major-version upgrade left after the rc.5 and rc.7 sweeps closed Svelte 4→5, Express 4→5, bcryptjs 2→3, multer 1.4.5-lts.1→2.x, nodemailer 8→9. Complete rewrite, cleaner API. Defer unless iterating on OIDC. Monthly `npm audit` cadence continues (see memory `project_lifttrace_dep_audit.md`).

### ~~Security dep bumps~~ *(done, v1.0.0-rc.7)*
Multer 1.4.5-lts.1 → 2.2.0 (three high-severity CVEs on the LTS line: unhandled-exception DoS, crafted-request DoS, unclosed-stream memory leak), nodemailer 8.0.7 → 9.0.3 (five CVEs including raw-option bypass, TLS OAuth cert validation, CRLF header injection), vite bumped alongside for the Windows `server.fs.deny` bypass patch. `npm audit` at both root and server now reports zero vulnerabilities.

### ~~Settings sub-component split~~ *(done, v0.9.x)*
890-line orchestrator, one file per section.

---

## Post-1.0 deferred bundles (from rc.x forum feedback)

These came from public feedback after v1.0.0-rc.x. They're structural enough that they should NOT land mid-RC. RC is for stabilising what's out, not piling on. Hold both until at least one cycle of RC feedback quiet, then start Bundle A.

Recommended sequencing: Bundle A first (unlocks a category of workouts LT currently can't express at all + the schema work is a prerequisite for any future "distance" or other non-rep set type). Bundle B second (purely additive, safer rollout, can land any time after A is stable).

Hold the work until the user explicitly green-lights a bundle start. Until then this section is the design-decisions snapshot so when the day comes the work resumes from a known starting point rather than re-deriving everything.

---

### Bundle A: Time-based exercises + HIIT-style rest

**Original commenter ask** (paraphrased from forum thread):

1. Per-exercise countdown timer for time-based work ("2 minutes of weighted carries, start timer, get countdown, get a beep when it hits"). Should be settable per exercise.
2. Per-set + per-round rest for HIIT circuits ("20s rest between exercises, then 2-minute rest before starting another round").
3. The commenter's mental model: "rest is just an exercise with a timer": work and rest as the same primitive.

**Open architectural decision: which framing**

Two viable shapes, neither chosen yet. Pick before writing code.

**Option 1, Rest-as-a-set (closer to commenter's literal ask)**

- Add a single `set.kind: 'work' | 'rest'` flag.
- Both work and rest sets share one primitive: a timed segment.
- HIIT circuits become inline sequences: work / rest / work / rest / long-rest.
- Stats math (volume, PRs, time-under-tension) filters `kind='rest'` rows out so the workout log isn't polluted by "47 sets of Rest this month".
- One concept (timed segment + kind flag) instead of three new fields.
- **Visible cost**: rest periods become inline rows in the diary. Every HIIT circuit gets ~10 extra rows. Strength-only users still see one work row per set as today, so they only notice it when they build a HIIT workout.

**Option 2, Set metadata (original Bundle A as written)**

- Add `set.duration_sec` (null for weight×reps work, number for time-based).
- Add `set.rest_after_sec` (null = program default, number = per-set override).
- Add `superset.round_break_sec` (rest after the LAST exercise in a round).
- Add `exercises.default_set_type: 'reps' | 'time'` for the input picker default.
- Three new fields + a fourth on exercises. Cleaner stats (rest never exists as a row), messier vocabulary (rest has its own non-set identity).
- **Visible cost**: rest stays invisible between rows (current behavior). Per-set rest override is hidden behind long-press by default. Diary view stays clean.

**Recommendation**: Option 1. It's literally what the commenter asked for, the diary-row cost is acceptable because strength users never trigger it, and the schema is genuinely simpler (one flag vs. four new fields).

**Schema changes once an option lands**

Option 1 path:

- `workout_log` JSON per-set gains: `kind` ('work' default), `duration_sec` (null for rep-based work), `weight`/`reps` stay exactly as today for backward compat.
- `exercises` gains optional `default_set_type: 'reps' | 'time'` (idempotent ALTER). Defaults to 'reps' for everything existing.

Option 2 path:

- `workout_log` JSON per-set gains: `duration_sec`, `rest_after_sec`.
- `workout_log` JSON per-exercise gains: `round_break_sec` on the superset wrapper.
- `exercises` gains `default_set_type` same as Option 1.

Both paths are backward-compatible; existing workouts (no new fields) render exactly as today.

**Implementation phases** (each independently testable)

Phase 1, Schema + SetRow input rendering (~2-3 evenings) *(done, issue #89)*

Shipped as the timed-sets slice of Bundle A, without deciding the rest-framing question below: the fields it needs (`set.duration_sec`, and a `set_type` of `'reps' | 'time'` on the exercise instance and the library row) are the same under Option 1 and Option 2, so that decision is still open. Named `set_type` rather than `default_set_type` to match the existing `load_type`, which has the same per-instance / library / remembered-preference layering. Recorded set data outranks every default, so history never changes shape. Covered: Diary chip and set row, exercise editor, templates (a real `duration` field through the weeks matrix and per-set specs), Statistics and exercise detail (longest hold), records and the `pr.set` webhook, CSV export, Strong and Hevy import, Smart-Add, Trace, MCP and the REST API.

Original phase notes:

- Add the columns + ALTER migration.
- SetRow renders `duration_sec + weight` row when set is time-based, `reps + weight` row when rep-based (current behavior).
- Volume math + share-card render filter / handle time-based sets appropriately (don't contribute to volume; do contribute to time-under-tension if that becomes a stat).

Phase 2, Active set timer (~3-4 evenings) *(done, issue #89: built as a count-up hold timer with a 3s lead-in and a cue at the set's target time, rather than a pure countdown, so holding past the target still logs the real time)*

- Countdown UI for time-based sets. Reuses the rest-timer audio / haptic / Service-Worker-notification plumbing (`src/stores/restTimer.js` pattern). Auto-completes on timer end + writes the actual duration.
- Pause / resume / manual stop. Lockscreen behaviour already solved by the rest timer's machinery, so this is mostly reuse not new build.

Phase 3, Per-set + round rest overrides (~2 evenings)

- Option 1 path: rest rows just exist inline in the diary. Each rest set has its own duration field. Round break is just a longer rest set at the end of a round.
- Option 2 path: long-press a rest pill to edit just this rest; round-break pill on the last set of a superset round. Reuses the existing rest timer for the actual countdown.

Total: ~1.5-2 weeks of evenings.

**UI clutter analysis + mitigations**

Without progressive disclosure, Bundle A adds 3-5 new UI surfaces: ExerciseEditor (+1 field), Superset header (+1 setting), SetRow (input shape swap), per-set rest override UI, WorkoutSummary (potential TUT tile).

Mitigations that keep strength-only users from seeing any of it:

- Default `default_set_type = 'reps'` on every existing + new exercise. Time UI only renders when an exercise is explicitly set to time-based. 95% of users see zero change.
- Hide ExerciseEditor's set-type field behind an Advanced disclosure if zero visible churn there matters.
- Per-set rest override hidden behind long-press, never displayed inline (Option 2 path only).
- Round break only shown on supersets that opt in (non-null `round_break_sec`); plain supersets render exactly as today.
- WorkoutSummary TUT tile only renders if the workout had any time-based sets. Reps-only workouts look identical.

**Discipline note for when work starts**

Bigger UI risk than the features themselves: feature creep during build. If a bundle start happens, lock down which surfaces get the new UI BEFORE writing code, and reject anything that wants to add a global setting / Settings screen entry. Discipline matters more than architecture here.

**Decisions still open before writing code**:

- Option 1 vs Option 2 (recommend Option 1).
- Time-under-tension as a stats metric? Track total seconds per exercise alongside volume, or skip stats integration entirely?
- Visible to which screens? Statistics page, exercise history, share card all need a small story for time-based work.
- `default_set_type` on the exercise or on the program template's exercise-row override? My take: exercise. Template can override if needed but the exercise itself usually has a natural set-type identity (plank → time, bench → reps).

---

### Bundle B: Live BLE chest-strap heart rate

Cleaner scope than Bundle A, more isolated. Adds an external-device surface area to test but doesn't touch core workout logging.

**Plumbing** (~3-4 evenings)

- New `src/lib/bleHr.js` abstracts over Web Bluetooth (PWA, Chrome only) and `@capacitor-community/bluetooth-le` (Android native, also works for iOS later). Both expose GATT Heart Rate Service `0x180D` characteristic `0x2A37` the same way, standard payload, Polar / Wahoo / generic straps all expose it.
- Pair-and-connect flow: device picker → connect → subscribe to BPM notifications → emit samples through a Svelte store.
- AndroidManifest gains `BLUETOOTH_CONNECT`, `BLUETOOTH_SCAN`, `ACCESS_FINE_LOCATION` permissions.
- Settings → Connected devices section: pair button, remembered device, auto-reconnect on workout start.

**Storage + UI** (~3-4 evenings)

- `workout_log.hr_samples` new column (JSON `[{ ts, bpm }]`). Idempotent ALTER, default null.
- Live BPM chip during workout (top of Diary, near the timer).
- WorkoutSummary stats: avg HR, max HR, optional time-in-zones (HR zones derived from age via 220-age default, overrideable in Settings).
- CSV export gains `hr_avg_bpm`, `hr_max_bpm` columns.

Total: ~1.5 weeks of evenings.

**Decisions still open before writing code**:

- HR zones: derive default from age (220-age × {50%, 60%, 70%, 80%, 90%}) or let user override? My take: derive default + show Settings override.
- iOS later. Web Bluetooth doesn't work on iOS Safari. When iOS lands the only BLE option is the same Capacitor plugin. Same JS flow, same data shape. Plan: have iOS done from day 1 by using the Capacitor plugin on native instead of Web Bluetooth on PWA from the start.
- Battery. Continuous BLE notifications drain. Stop subscription on workout end + on app-pause-to-background-without-active-workout.

---

### Why not start either bundle now

RC is meant to stabilise what's out, not pile on. Both bundles are structural enough that mid-RC introduction risks regression noise that drowns out actual RC bug feedback. Hold until at least one cycle has passed where the RC isn't generating new bug feedback, then start Bundle A.

The user can override this and start either bundle whenever; these are deferred, not blocked.