Companion: add three Chinese sleep proverbs to the bedtime nag

Extends the BEDTIME rotation with classic Chinese proverbs on sleep. The zh
lines are faithful, natural-reading renderings of the proverbs' sense (the
verified classical 原文 isn't recoverable; the source English lives in old
collections) — swap in exact originals later if available. The "spouse" line
is kept non-gendered on purpose.

Claude-Session: https://claude.ai/code/session_016Yr6jELuRc7hyzYLccQKZd

.env.example

@@ -17,6 +17,20 @@ LLM_MODEL=                        # checkpoint model (small/fast). Never hardcod
 LLM_CHAT_MODEL=                   # Ask Petal model (Mandarin-native). Falls back to LLM_MODEL if empty.
 LLM_TIMEOUT=30s
 
+# Read-aloud (TTS). Off unless TTS_ENDPOINT is set — when empty, the /api/tts route
+# is not mounted and the frontend falls back to the browser's Web Speech API.
+# Backed by a local Piper HTTP server (python3 -m piper.http_server).
+# Each Piper HTTP server loads ONE voice, so English and Chinese need separate
+# instances (different ports). zh is only routed when both TTS_ENDPOINT_ZH and
+# TTS_VOICE_ZH are set; otherwise Chinese falls back to Web Speech.
+TTS_ENDPOINT=                     # e.g. http://127.0.0.1:5005 — empty disables server TTS
+TTS_ENDPOINT_ZH=                  # e.g. http://127.0.0.1:5006 — Chinese Piper instance
+TTS_VOICE_EN=en_US-amy-medium     # Piper voice id for English
+TTS_VOICE_ZH=zh_CN-huayan-medium  # Piper voice id for Chinese
+TTS_CACHE_DIR=./data/tts          # on-disk store for synthesized clips (content-addressed)
+TTS_TIMEOUT=15s
+TTS_AUDIO_FORMAT=mp3              # mp3 | opus | wav — mp3/opus transcode Piper's WAV via ffmpeg
+
 # --- Deferred (not wired in the local-dev build) ---
 
 # Auth (Authentik OIDC) — deferred; single hardcoded local user for now

BUILD_PLAN.md

@@ -105,6 +105,35 @@ Multi-session build. **Source of truth for what's done and what's next.** Update
 - [x] **English phonetic (pivot from pinyin)** — for a native-Mandarin English learner the useful pronunciation aid is the English IPA, not pinyin (she reads Chinese fluently). `scripts/build_phonetic.py` extracts ECDICT's `phonetic` column (same source/freq-gate as the gloss); `phonetic.json.gz` embedded + lazily loaded; `Result.Phonetic` resolved via the same de-inflecting candidate walk; shown as `/ˈrɪvər/` in the WordCard. **Full dataset built from ECDICT: 46,579 words (361KB gz)**, in line with the other lexicon assets. The script also has a `--seed` mode (71 curated common words) that ships as a fallback / works without the csv. Curated seed entries (clean IPA) override the ECDICT form where both exist.
 - Verified: go build/vet/test (incl. new `TestExportAll`), tsc, vite build all clean; live smoke vs throwaway binaries — `/word/river|running|serendipity|rivers` all return phonetic (`rivers`→`river` de-inflected), export-all returns a valid 2-entry zip with de-duped CJK filenames + dated name, route doesn't collide with `/{id}`.
 
+### Phase 12 — Collocation coach 🔜 (planned 2026-06-26)
+**Why:** ESL writers nail grammar but miss *which words go together* — "do a decision" → "make a decision", "strong rain" → "heavy rain". These aren't *wrong*, so the grammar pass won't flag them; they're just non-native. Gentle "natives usually say…" hints are the single highest-leverage upgrade for making her writing sound native. **Build this first** — it's small and de-risks the migration-rebuild pattern Phase 13 also needs.
+
+**Key insight:** the suggestion pipeline is already generic over a `pass` + a `pendingScope` "family" (`runPass` in `internal/suggestions/handlers.go`; grammar + voice already prove it). Collocation drops in as a **third family** and reuses the entire accept/dismiss/re-anchoring/rail/Mandarin-explanation machinery — no new frontend rendering layer.
+- [ ] `internal/llm/collocation.go` — `CollocationInterval` (~25s) + `RunCollocation(...)`, reusing the existing `ParseCheckpoint` parser (same as `RunVoice`). No new parsing.
+- [ ] `internal/llm/prompts.go` — `CollocationMessages(contentText, tone)`. **This prompt is the whole feature**: flag only non-wrong-but-non-native word pairings; explicitly defer real grammar errors to the grammar family; phrase every explanation as warm "natives usually say…" with a Mandarin gloss — never "error/wrong".
+- [ ] `internal/db/models.go` + migration `0005` — new suggestion type `collocation`. ⚠️ The `type` column has a `CHECK(...)` and **SQLite can't `ALTER` a CHECK** → migration must **rebuild** the `suggestions` table (create new w/ extended CHECK, copy rows, drop, rename, recreate `idx_suggestions_doc_id`). Not a one-line ALTER.
+- [ ] `internal/suggestions/handlers.go` — add `collocationScope` (`deleteWhere: "type = 'collocation'"`, `forceType: collocation`), a `CollocationLimit` on `Handler` (+ wire in `New`), register `POST /{id}/collocation`. Mirrors `voice`, ~15 lines. Also extend `normalizeType` to accept the new type.
+- [ ] Frontend: `suggestionMeta.ts` collocation entry (💡 icon + its own warm color so cards read as friendly tips); `client.ts` `collocation(docId)`; a gentle trigger like the existing "Check my voice" pill — **"Make it sound natural 🌸"** — rendering straight into `SuggestionRail`/`SuggestionCard`.
+- [ ] Verify via the millenia UI-test harness (real-browser editor check) + go/tsc/vite clean. **Est. ~½ day** (mostly the new prompt + the table-rebuild migration).
+
+### Phase 13 — Vocabulary garden (spaced repetition) 🔜 (planned 2026-06-26)
+**Why:** the lexicon (`internal/lexicon`) is a stateless static-dataset lookup — **nothing records which words she's looked up.** Capturing them turns passive lookups into real vocabulary, and the review surface ties straight into the "petal garden" delight idea (words become blossoms; the sleeping kitten naps among them). Build **after** Phase 12.
+- [ ] New `internal/vocab` package + migration `0005`/`0006` — `vocab_words` table: `word, gloss, phonetic, example` (sentence captured at lookup for context), `doc_id`, + SM-2-lite scheduling: `due_at, interval_days, ease, reps, lapses, last_reviewed`; `UNIQUE(user_id, word)`.
+- [ ] Auto-capture: when `WordCard` calls `lookupWord`, also fire `POST /api/vocab` (upsert; new word → `due_at = now + 1 day`). Looking words up *is* the data source — zero extra effort from her.
+- [ ] Endpoints: `POST /api/vocab` (record/upsert), `GET /api/vocab/due` (cards due now), `POST /api/vocab/{id}/review` (grade again/good/easy → reschedule), `GET /api/vocab` (full garden), `DELETE /api/vocab/{id}`.
+- [ ] SR scheduler: gentle SM-2-lite / Leitner intervals (1d → 3d → 7d → 16d → 35d). No streak-shaming. This is the only genuinely new logic.
+- [ ] Frontend Garden panel (sibling to `HistoryPanel`/`DocList`): each learned word a blossom, more reps → more bloomed; the `PetalCompanion` kitten naps among them. Flashcard review: captured sentence with the word blanked → recall → EN↔中文 flip → again/good/easy (direction flips for recognition *and* production). A 🤍 "save to garden" on `WordCard` for explicit saves alongside auto-capture.
+- [ ] Verify via the millenia UI-test harness + go/tsc/vite clean. **Est. ~1½–2 days** (new package + table + panel/review UI; reuses existing panel/companion patterns).
+
+### Phase 14 — Companion warmth + bedtime nag + night mode ✅
+**Why:** the companion kitten is the heart of Petal's "built for her" feel. Three additions: (1) a wider, fresher pool of **encouraging phrases** so cheers don't repeat as quickly; (2) when she's still writing **late at night (≥11pm)**, the kitten gently nags her to go to bed; (3) at the same hour the whole app drifts into a calm **night mode** — dark moonlit theme + the falling petals become **falling stars**. Caring, never scolding — the sleepy-cat gag makes "you should be sleeping too 🐱💤" land perfectly. Self-contained, frontend-only.
+- [x] `web/src/components/Companion/tips.ts` — `ENCOURAGEMENTS` grown from 5→10 bilingual zh-first lines so cheers rotate fresher. New `BEDTIME: Line[]` array — four warm/playful lines (user-supplied English wit: "I bet your bed is missing you right now", "A tired writer is a bad writer", "Sleep is a wondrous enabler", "Hear that? No… everyone is sleeping and you should be too") with gentle Mandarin leads.
+- [x] `web/src/components/Companion/useCompanion.ts` — bedtime check folded into the existing 10s heartbeat (after the idle-return + break check, before the generic tip): `isBedtime()` = local hour ≥ 23 or < 4 (`new Date().getHours()`, the writer's machine clock). Only fires while actively writing (idle branch returns first). Own `lastBedtime` ref + `BEDTIME_GAP` 30min cooldown; respects `PROACTIVE_GAP`. New `BubbleTone` `'bedtime'` paces `readBubbleMs` (BUBBLE_MS + 4s lingers a touch longer for a wind-down read). Window knobs `BEDTIME_FROM`/`BEDTIME_TO` so the 11pm–4am range is one edit to retune.
+- [x] No tone-based bubble styling exists, so no CSS needed; the tone is metadata for pacing only. Sound stays the rotating pop.
+- [x] **Night mode** — `web/src/lib/night.ts` centralizes `isBedtime()` + the `BEDTIME_FROM`/`BEDTIME_TO` window (shared with the companion nag so they always agree). `web/src/hooks/useNightMode.ts` re-checks every 60s and toggles a `petal-night` class on `<html>`. `index.css` adds an `html.petal-night` block that **only re-points the palette tokens** (`--color-bg/surface/border/plum/muted/accent…`) to a dark moonlit set — every Tailwind color utility reads them via `var()`, so the whole UI flips with zero component changes (verified: `.text-plum{color:var(--color-plum)}`). Accent/type colors kept (they pop on dark); 600ms bg/color fade for a gentle dusk transition; print stays white (the `#fff` override is inside `@media print`).
+- [x] **Falling stars** — `PetalFall` gains a `night` prop. Particles are **chunky cartoon power stars** (`makeCartoonStar`, Mario/Kirby-style: fat 5-point shape, glossy radial fill, puffy round-join colored outline, corner shine + soft glow halo so they pop off the dark sky) in 5 candy colors (`CARTOON_COLORS`), mixed ~70/30 with small four-point twinkle sparkles (`makeStarSprite`/`STAR_PALETTE`) for depth. Every star clearly **spins** (random direction, ~0.5–2.2 rad/s so the quick ones really whirl while slow ones drift for contrast; guaranteed min speed since a 5-point star is symmetric every 72°), **falls straight down** (no sway — that's a petal thing), and shimmers via a shallow alpha pulse (not blink). Effect re-inits on the day↔night flip. App wires `const night = useNightMode()` → `<PetalFall night={night} />`. All sprites are canvas-drawn (offline, no asset files) — `sprites[]` is an image array, so a real PNG/SVG star could drop in later without restructuring.
+- [x] Verified: tsc clean, vite build OK, companion vitest 45/45. **Real-browser screenshots** (local Playwright + Chromium, clock mocked to 23:30): day = warm cream + pink sakura petals; night = dark plum-indigo + twinkling stars + glowing sleepy kitten. Both pretty (acceptance criterion).
+
 ### Deferred (post-v1-local)
 - [ ] Authentik OIDC auth + session middleware  ← **on hold: user doing foundational work first**
 - [ ] Copyleaks Tier-2 + webhook HMAC
@@ -113,8 +142,12 @@ Multi-session build. **Source of truth for what's done and what's next.** Update
 ### Next-up (post-v1 product, agreed with user 2026-06-26)
 - [x] **Phase 9 — ESL superpowers**: inline Chinese gloss on hover/select; "say it more naturally" / tone-rewrite. ✅ (see Phase 9 above)
 - [x] **Phase 10 — organization & polish**: cross-doc search, tags, tablet/touch polish, warm LLM-down failure states. ✅ (see Phase 10 above; "tags only" chosen over folders, FTS5 over LIKE)
+- [ ] **Phase 12 — collocation coach**: gentle "natives usually say…" hints for non-native word pairings, as a third suggestion family. 🔜 (see Phase 12 above; build first)
+- [ ] **Phase 13 — vocabulary garden**: spaced-repetition review built from looked-up words, surfaced as a blooming garden. 🔜 (see Phase 13 above)
+- [x] **Phase 14 — companion warmth + bedtime nag + night mode**: more encouraging phrases, a gentle "go to bed" nudge after 11pm, and a calm dark theme + falling stars at night. ✅ (see Phase 14 above)
 
 ## Session log
+- 2026-06-26: **Phase 14 complete** (companion warmth + bedtime nag + night mode). `tips.ts`: `ENCOURAGEMENTS` 5→10 lines; new `BEDTIME` array (4 lines, user-supplied English wit + gentle Mandarin leads). `useCompanion.ts`: bedtime branch in the 10s heartbeat (after idle-return + break, before the generic tip); only nudges while actively writing; own `lastBedtime` ref + 30min `BEDTIME_GAP`, respects `PROACTIVE_GAP`; new `'bedtime'` `BubbleTone` lingers ~4s longer. **Night mode** (added same session, user request): `lib/night.ts` centralizes `isBedtime()` + window (now shared by the nag too); `hooks/useNightMode.ts` toggles `petal-night` on `<html>` (60s re-check); `index.css` `html.petal-night` re-points only the palette tokens → whole UI flips via `var()` (no component edits), 600ms dusk fade, print stays white; `PetalFall` gains a `night` prop → chunky cartoon power stars (`makeCartoonStar`, Mario/Kirby-style, 5 candy colors) mixed ~70/30 with small twinkle sparkles, gentle spin + shallow shimmer, effect re-inits on flip; App: `useNightMode()` → `<PetalFall night={night}/>`. tsc + vite clean, companion vitest 45/45; verified with real-browser Playwright screenshots (clock mocked to 23:30) — day petals/cream vs night stars/dark-plum, both pretty. Bedtime window is `BEDTIME_FROM`/`BEDTIME_TO` (local clock) for easy retune.
 - 2026-06-26: **Phase 11 complete** (writer power-ups, batch requested as "do it all"). Seven features: (1) in-doc **Find & Replace** — `SearchHighlight` decoration extension + `FindReplace` bar (Ctrl/Cmd+F, match-case, replace-all back-to-front, DOM scroll that doesn't trigger the selection bubble); (2) **read-aloud** Web Speech util + 🔊 in WordCard & selection bubble; (3) **keyboard/touch access** — Ctrl/Cmd+D caret lookup, Ctrl/Cmd+J rewrite, touch long-press (refactored `handleContextMenu` → shared `openWordLookup(pos)`); (4) **export-all** backup zip (`GET /api/docs/export-all`, `TestExportAll`, sidebar download links); (5) **smart typography** input-rules extension (curly quotes/em-dash/ellipsis, ASCII-only so CJK untouched); (6) **duplicate doc + sidebar sort + outline popover**; (7) **English phonetic** (pivoted from pinyin — IPA is what an English learner needs; pinyin annotates Chinese she already reads) via `scripts/build_phonetic.py` + embedded `phonetic.json.gz` + `Result.Phonetic` + WordCard `/ˈrɪvər/` line — **full 46,579-word dataset built from ECDICT** (the csv re-download worked; `--seed` mode kept as a csv-free fallback). Also folded in this session: the **selection-bubble vs copy/paste fix** (bubble deferred to pointer-up + container `pointer-events:none` so it never sits where you click). go build/vet/test + tsc + vite all clean; live smoke verified word-phonetic (incl. de-inflection) + export-all zip (de-duped CJK names, route priority). Next: deferred bucket (auth/Copyleaks/deploy), still on hold per user.
 - 2026-06-26: **Phase 10 complete** (organization & polish). Scope confirmed with user: all four areas, **tags** (not folders), **FTS5** search. Backend: migration `0004_tags_and_search` (tags + document_tags + `documents_fts` trigram virtual table with sync triggers + back-fill); `db.Tag` model + color constants; `internal/docs/tags.go` (tag CRUD + idempotent assignment + `tagsByDoc` helper, doc list now carries tags); `internal/docs/search.go` (`GET /api/search`, FTS for ≥3 runes + LIKE fallback for 1-2, Go-built sentinel-highlighted rune-aware snippets, owner-scoped). Mounted `/api/tags` + `/api/search` in main.go. Frontend: `useTags`, `TagChip`/`TagPicker`/`SearchBox`, rewritten `DocList`/`DocListItem` (chips + filter bar + search), `api.search`/tag methods + `splitSnippet`/`tagColorVar`; responsive sidebar drawer (hamburger + scrim, <768px) + `pointer:coarse` tap-target/affordance CSS; tap-to-open + outside-pointerdown-close for suggestion cards (touch); `useCheckpoint` `llmDown` flag → warm bilingual "小助手在休息" StatusBar note. Tests: `tags_test.go`, `search_test.go` (incl. update-trigger re-index). All builds/tests/vet/tsc/vite clean; live smoke vs binary on :8061 (dead LLM host) verified search EN/CJK/2-char, full tag lifecycle, check→502 warm path, bundle contents; FTS backfill of pre-existing docs verified. **All v1 phases (0–7) + post-v1 product (8–10) done.** Remaining: deferred bucket (auth/Copyleaks/deploy), on hold per user.
 - 2026-06-26: **Phase 9 complete** (ESL superpowers: inline Chinese gloss + tone-rewrite). Decisions confirmed with user: gloss is an **offline EC dictionary** (instant, LLM-down-proof, fits the embedded-lexicon ethos), rewrite is a **selection bubble**. Data: `scripts/build_gloss.py` builds `internal/lexicon/data/gloss.json.gz` from ECDICT (66MB csv → 1.3MB gz, 57k freq-≤50k words, cleaned/trimmed). Backend: `lexicon` gloss map + `Gloss()`/`Result.Gloss` + `GET /api/gloss/{word}`; `llm.RunRewrite` + rewrite prompt/`styleGuidance`; `internal/suggestions/rewrite.go` (`POST /api/docs/:id/rewrite`, stateless, owner-scoped). Frontend: `GlossTip` hover tooltip (350ms delay, reuses `wordAt`, CJK-safe) + gloss line in `WordCard`; `SelectionBubble` + `RewritePreview` wired through `EditorCore` (onMouseMove/onSelectionUpdate, request-token guards, clears on edit/doc-switch); `api.glossWord`/`api.rewriteSelection`; CSS for the three new surfaces (+ print-hidden). Tests added in `lexicon` and `suggestions`. All builds/tests/vet/tsc/vite clean; live smoke vs fake vLLM on :8055 verified gloss + rewrite + 400/404/502 paths. Next: **Phase 10 (organization & polish).**

cmd/server/main.go

@@ -19,6 +19,7 @@ import (
 	"gitea.parodia.dev/drwily/petal/internal/lexicon"
 	"gitea.parodia.dev/drwily/petal/internal/llm"
 	"gitea.parodia.dev/drwily/petal/internal/suggestions"
+	"gitea.parodia.dev/drwily/petal/internal/tts"
 	"gitea.parodia.dev/drwily/petal/web"
 )
 
@@ -88,6 +89,14 @@ func main() {
 			log.Fatalf("image store: %v", err)
 		}
 		api.Mount("/images", imgHandler.Routes())
+
+		// Read-aloud: proxy short passages to a local Piper TTS server. Only
+		// mounted when TTS_ENDPOINT is configured; otherwise the frontend falls
+		// back to the browser's Web Speech API on its own.
+		if ttsHandler, ok := tts.New(cfg); ok {
+			api.Mount("/tts", ttsHandler.Routes())
+			log.Printf("read-aloud enabled (TTS endpoint=%s)", cfg.TTSEndpoint)
+		}
 	})
 
 	// Everything else: serve the embedded SPA (with index.html fallback for client routing).

deploy/README.md

@@ -0,0 +1,65 @@
+# Deploying read-aloud (Piper TTS) to millenia
+
+Petal's read-aloud generates audio with a local **Piper** neural-TTS server and
+transcodes it to mp3 with **ffmpeg**. Both run on millenia (192.168.1.212);
+nothing leaves the box. If Piper is down or `TTS_ENDPOINT` is unset, the frontend
+falls back to the browser's Web Speech API automatically.
+
+## 1. Piper as a systemd service (one-time)
+
+```bash
+# from this repo, on your workstation:
+scp deploy/piper.service deploy/setup-piper.sh 192.168.1.212:/tmp/
+ssh 192.168.1.212 'cd /tmp && sudo ./setup-piper.sh'
+```
+
+`setup-piper.sh` creates `~/piper/venv`, installs `piper-tts[http]`, downloads the
+`en_US-amy-medium` voice into `~/piper/voices`, installs+enables `piper.service`
+(loopback :5005), and smoke-tests it. Idempotent.
+
+Check it any time: `systemctl status piper`, `journalctl -u piper -f`.
+
+## 2. Point petal at Piper + redeploy the binary
+
+Add to petal's environment (its `.env` or launch env):
+
+```
+TTS_ENDPOINT=http://127.0.0.1:5005
+TTS_VOICE_EN=en_US-amy-medium
+TTS_AUDIO_FORMAT=mp3
+```
+
+Then ship the rebuilt binary (`go build -o petal ./cmd/server` already done) and
+restart the petal `:8088` session. Confirm the log line:
+`read-aloud enabled (TTS endpoint=http://127.0.0.1:5005)`.
+
+## 3. Verify
+
+```bash
+# on millenia — end-to-end through petal, including ffmpeg transcode:
+curl -sf -X POST localhost:8088/api/tts \
+  -H 'Content-Type: application/json' \
+  -d '{"text":"hello there","lang":"en-US"}' -o /tmp/petal-tts.mp3 \
+  && file /tmp/petal-tts.mp3   # expect: Audio file ... MPEG ... layer III
+```
+
+- Second identical call is served from the cache (`~/petal/.../data/tts/*.mp3`).
+- A language with no configured instance returns 404 → client uses Web Speech.
+- Browser check via the uitest harness (`~/petal/uitest`): tap a word in the
+  WordCard / select a sentence and hit speak — expect the natural Piper voice.
+
+## Chinese voice (live)
+
+Each Piper HTTP server loads ONE model, so Chinese runs as a **second instance**:
+`piper-zh.service` on :5006 with `zh_CN-huayan-medium`. Deployed via:
+
+```bash
+scp deploy/piper-zh.service 192.168.1.212:~/.config/systemd/user/
+ssh 192.168.1.212 'export XDG_RUNTIME_DIR=/run/user/$(id -u)
+  ~/piper/venv/bin/python -m piper.download_voices zh_CN-huayan-medium --data-dir ~/piper/voices
+  systemctl --user daemon-reload && systemctl --user enable --now piper-zh.service'
+```
+
+Then in petal's `start.sh`: `TTS_ENDPOINT_ZH=http://127.0.0.1:5006` and
+`TTS_VOICE_ZH=zh_CN-huayan-medium`. The handler maps language → instance from config,
+so adding more languages is just another instance + env pair (no code change).

deploy/piper-zh.service

@@ -0,0 +1,17 @@
+[Unit]
+Description=Piper TTS HTTP server — Chinese voice (read-aloud backend for petal)
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+# Bound to loopback only — petal proxies to it; never exposed off-box.
+ExecStart=%h/piper/venv/bin/python -m piper.http_server \
+  -m zh_CN-huayan-medium \
+  --data-dir %h/piper/voices \
+  --host 127.0.0.1 --port 5006
+Restart=on-failure
+RestartSec=3
+
+[Install]
+WantedBy=default.target

deploy/piper.service

@@ -0,0 +1,17 @@
+[Unit]
+Description=Piper TTS HTTP server (read-aloud backend for petal)
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+# Bound to loopback only — petal proxies to it; never exposed off-box.
+ExecStart=%h/piper/venv/bin/python -m piper.http_server \
+  -m en_US-amy-medium \
+  --data-dir %h/piper/voices \
+  --host 127.0.0.1 --port 5005
+Restart=on-failure
+RestartSec=3
+
+[Install]
+WantedBy=default.target

deploy/setup-piper.sh

@@ -0,0 +1,43 @@
+#!/usr/bin/env bash
+# One-time Piper TTS setup on millenia as a *user* systemd service (no sudo).
+# Idempotent — safe to re-run. Run on the box:
+#   bash setup-piper.sh
+# To survive logout/reboot, also (once, needs root):
+#   sudo loginctl enable-linger "$USER"
+set -euo pipefail
+
+PIPER_HOME="$HOME/piper"
+VENV="$PIPER_HOME/venv"
+VOICES="$PIPER_HOME/voices"
+VOICE=en_US-amy-medium
+UNIT_DIR="$HOME/.config/systemd/user"
+export XDG_RUNTIME_DIR="/run/user/$(id -u)"
+
+echo ">> ffmpeg check (petal transcodes Piper WAV -> mp3 with it):"
+command -v ffmpeg >/dev/null || { echo "ffmpeg not found on PATH"; exit 1; }
+
+echo ">> venv + piper-tts[http]"
+[ -d "$VENV" ] || python3 -m venv "$VENV"
+"$VENV/bin/pip" install --upgrade pip >/dev/null
+"$VENV/bin/pip" install "piper-tts[http]"
+
+echo ">> download voice $VOICE"
+mkdir -p "$VOICES"
+"$VENV/bin/python" -m piper.download_voices "$VOICE" --data-dir "$VOICES"
+
+echo ">> install user service"
+mkdir -p "$UNIT_DIR"
+install -m 0644 "$(dirname "$0")/piper.service" "$UNIT_DIR/piper.service"
+systemctl --user daemon-reload
+systemctl --user enable --now piper.service
+
+echo ">> wait + status"
+sleep 2
+systemctl --user --no-pager --full status piper.service | head -n 6 || true
+
+echo ">> smoke test"
+curl -sf -X POST localhost:5005/ \
+  -H 'Content-Type: application/json' \
+  -d "{\"text\":\"hello there\",\"voice\":\"$VOICE\"}" -o /tmp/piper-test.wav \
+  && echo "OK: /tmp/piper-test.wav ($(stat -c%s /tmp/piper-test.wav) bytes)" \
+  || { echo "smoke test FAILED"; journalctl --user -u piper.service --no-pager | tail -n 20; exit 1; }

internal/config/config.go

@@ -20,6 +20,17 @@ type Config struct {
 	LLMChatModel string // Ask Petal model; falls back to LLMModel if empty
 	LLMTimeout   time.Duration
 
+	// TTS (read-aloud). Off unless TTSEndpoint is set — when empty, the /api/tts
+	// route isn't mounted and the frontend falls back to the browser's Web Speech
+	// API. Endpoint points at a local Piper HTTP server.
+	TTSEndpoint   string // Piper instance serving the English voice
+	TTSEndpointZH string // Piper instance serving the Chinese voice; empty = zh falls back to Web Speech
+	TTSVoiceEN    string // Piper voice id for English (e.g. en_US-amy-medium)
+	TTSVoiceZH    string // Piper voice id for Chinese (e.g. zh_CN-huayan-medium)
+	TTSCacheDir string // on-disk store for synthesized clips (content-addressed)
+	TTSTimeout  time.Duration
+	TTSFormat   string // mp3 | opus | wav — mp3/opus transcode Piper's WAV via ffmpeg
+
 	// Auth (deferred — not wired in the local-dev build, kept for later)
 	AuthentikURL          string
 	AuthentikClientID     string
@@ -41,6 +52,14 @@ func Load() *Config {
 		LLMChatModel: env("LLM_CHAT_MODEL", ""),
 		LLMTimeout:   envDuration("LLM_TIMEOUT", 30*time.Second),
 
+		TTSEndpoint:   env("TTS_ENDPOINT", ""),
+		TTSEndpointZH: env("TTS_ENDPOINT_ZH", ""),
+		TTSVoiceEN:    env("TTS_VOICE_EN", "en_US-amy-medium"),
+		TTSVoiceZH:    env("TTS_VOICE_ZH", "zh_CN-huayan-medium"),
+		TTSCacheDir: env("TTS_CACHE_DIR", "./data/tts"),
+		TTSTimeout:  envDuration("TTS_TIMEOUT", 15*time.Second),
+		TTSFormat:   env("TTS_AUDIO_FORMAT", "mp3"),
+
 		AuthentikURL:          env("AUTHENTIK_URL", ""),
 		AuthentikClientID:     env("AUTHENTIK_CLIENT_ID", ""),
 		AuthentikClientSecret: env("AUTHENTIK_CLIENT_SECRET", ""),

internal/llm/checkpoint.go

@@ -14,6 +14,16 @@ import (
 // that protects the inference endpoint from rapid repeat checks.
 const CheckpointInterval = 30 * time.Second
 
+// checkpointMaxTokens caps the checkpoint's generated output. Each suggestion
+// echoes its original sentence, a replacement, and a friendly explanation, so
+// the full JSON for the prompt's "up to 5 issues" runs ~2,000 tokens on a long
+// document (measured ~2,007 for a 380-word doc). The old 1,024 cap truncated
+// the JSON mid-array — done_reason "length" — leaving an object that never
+// closed, so ParseCheckpoint found no parseable JSON and the pass 502'd on
+// every long doc. This is a ceiling, not a target: the model stops at its JSON
+// close well before it, so the headroom costs nothing on shorter docs.
+const checkpointMaxTokens = 4096
+
 // RawSuggestion is one item as the model emits it. Positions are resolved
 // server-side from Original; the frontend re-anchors by string at render time.
 type RawSuggestion struct {
@@ -34,7 +44,7 @@ type checkpointResponse struct {
 func RunCheckpoint(ctx context.Context, client LLMClient, contentText, tone string) ([]RawSuggestion, error) {
 	raw, err := client.Complete(ctx, CompletionRequest{
 		Messages:          CheckpointMessages(TruncateDoc(contentText), tone),
-		MaxTokens:         1024,
+		MaxTokens:         checkpointMaxTokens,
 		Temperature:       0.3,
 		RepetitionPenalty: 1.15,
 		TopP:              0.9,
@@ -48,20 +58,32 @@ func RunCheckpoint(ctx context.Context, client LLMClient, contentText, tone stri
 
 // ParseCheckpoint extracts the suggestions array from a model response. Smaller
 // models sometimes wrap JSON in prose or markdown fences despite instructions,
-// so we salvage the outermost {...} object before decoding.
+// so we salvage the outermost {...} object before decoding. When the model
+// truncates mid-array (it ran out of output budget) the outer object never
+// closes and won't parse — we then recover the suggestion objects that *did*
+// complete, so a long document still gets feedback instead of a hard failure.
 func ParseCheckpoint(raw string) ([]RawSuggestion, error) {
-	jsonText := extractJSONObject(raw)
-	if jsonText == "" {
-		return nil, fmt.Errorf("checkpoint: no JSON object in model output: %q", truncateForError(raw))
+	var suggestions []RawSuggestion
+	parsed := false
+	if jsonText := extractJSONObject(raw); jsonText != "" {
+		var resp checkpointResponse
+		if err := json.Unmarshal([]byte(jsonText), &resp); err == nil {
+			suggestions = resp.Suggestions
+			parsed = true
+		}
 	}
-	var parsed checkpointResponse
-	if err := json.Unmarshal([]byte(jsonText), &parsed); err != nil {
-		return nil, fmt.Errorf("checkpoint: parse JSON: %w (got %q)", err, truncateForError(jsonText))
+	if !parsed {
+		// No closeable object, or it failed to decode (most often a truncated
+		// array). Salvage the individual suggestion objects that completed.
+		suggestions = salvageSuggestions(raw)
+		if suggestions == nil {
+			return nil, fmt.Errorf("checkpoint: no JSON object in model output: %q", truncateForError(raw))
+		}
 	}
 	// Drop items the model returned with an empty original — they can't be
 	// anchored — and normalize whitespace the model may have echoed.
-	out := parsed.Suggestions[:0]
-	for _, s := range parsed.Suggestions {
+	out := suggestions[:0]
+	for _, s := range suggestions {
 		s.Original = strings.TrimSpace(s.Original)
 		s.Replacement = strings.TrimSpace(s.Replacement)
 		if s.Original == "" {
@@ -72,12 +94,55 @@ func ParseCheckpoint(raw string) ([]RawSuggestion, error) {
 	return out, nil
 }
 
+// salvageSuggestions recovers as many complete suggestion objects as it can from
+// a response whose top-level JSON didn't parse — typically one truncated mid
+// "suggestions" array. It scans the array region for brace-balanced {...}
+// objects and keeps each that decodes into a RawSuggestion with a non-empty
+// original; the final, cut-off object simply never balances and is dropped.
+// Returns nil when there is no salvageable array (so the caller can error).
+func salvageSuggestions(raw string) []RawSuggestion {
+	key := strings.Index(raw, `"suggestions"`)
+	if key < 0 {
+		return nil
+	}
+	lb := strings.IndexByte(raw[key:], '[')
+	if lb < 0 {
+		return nil
+	}
+	rest := raw[key+lb+1:]
+	out := []RawSuggestion{}
+	for {
+		obj, end := firstBalancedObject(rest)
+		if end < 0 {
+			break
+		}
+		var s RawSuggestion
+		if err := json.Unmarshal([]byte(obj), &s); err == nil && strings.TrimSpace(s.Original) != "" {
+			out = append(out, s)
+		}
+		rest = rest[end:]
+	}
+	if len(out) == 0 {
+		return nil
+	}
+	return out
+}
+
 // extractJSONObject returns the substring from the first '{' to its matching
 // closing '}', or "" if none. Tolerates fences/preamble around the object.
 func extractJSONObject(s string) string {
+	obj, _ := firstBalancedObject(s)
+	return obj
+}
+
+// firstBalancedObject returns the first brace-balanced {...} substring in s and
+// the index in s just past its closing brace, or ("", -1) if no object closes
+// (none present, or the only one is truncated). String contents — including
+// braces inside quoted values — are skipped so they can't unbalance the count.
+func firstBalancedObject(s string) (string, int) {
 	start := strings.IndexByte(s, '{')
 	if start < 0 {
-		return ""
+		return "", -1
 	}
 	depth := 0
 	inString := false
@@ -98,11 +163,11 @@ func extractJSONObject(s string) string {
 		case c == '}':
 			depth--
 			if depth == 0 {
-				return s[start : i+1]
+				return s[start : i+1], i + 1
 			}
 		}
 	}
-	return ""
+	return "", -1
 }
 
 func truncateForError(s string) string {
@@ -127,17 +192,33 @@ func NewRateLimiter(interval time.Duration) *RateLimiter {
 }
 
 // Allow reports whether a checkpoint may run for docID now. When allowed it
-// records the time and returns (true, 0); when throttled it returns
-// (false, retryAfter) where retryAfter is the wait until the next allowed run.
-func (rl *RateLimiter) Allow(docID string) (bool, time.Duration) {
+// records the time and returns (true, 0, recordedAt); when throttled it returns
+// (false, retryAfter, zero) where retryAfter is the wait until the next allowed
+// run. recordedAt lets a caller whose pass fails hand the exact timestamp to
+// Release so it rolls back only its own slot.
+func (rl *RateLimiter) Allow(docID string) (bool, time.Duration, time.Time) {
 	rl.mu.Lock()
 	defer rl.mu.Unlock()
 	now := time.Now()
 	if last, ok := rl.last[docID]; ok {
 		if elapsed := now.Sub(last); elapsed < rl.interval {
-			return false, rl.interval - elapsed
+			return false, rl.interval - elapsed, time.Time{}
 		}
 	}
 	rl.last[docID] = now
-	return true, 0
+	return true, 0, now
+}
+
+// Release rolls back the slot a prior Allow recorded for docID, but only if no
+// newer Allow has since claimed it. Called when an LLM pass fails: Allow runs
+// before the model call, so without this a failed checkpoint would hold the
+// per-document slot for the full interval and a retry (or the frontend's
+// auto-retry) would hit the throttle path and get the stale/empty set back
+// instead of re-running. `at` is the time the failed Allow returned.
+func (rl *RateLimiter) Release(docID string, at time.Time) {
+	rl.mu.Lock()
+	defer rl.mu.Unlock()
+	if last, ok := rl.last[docID]; ok && last.Equal(at) {
+		delete(rl.last, docID)
+	}
 }

internal/llm/checkpoint_test.go

@@ -42,6 +42,21 @@ func TestParseCheckpoint(t *testing.T) {
 			raw:  `{"suggestions":[{"original":"use {x}","replacement":"use x","explanation":"drop the braces","type":"clarity"}]}`,
 			want: 1,
 		},
+		{
+			// The bug that 502'd every long doc: the model hit num_predict and the
+			// JSON never closed. Two suggestions completed; the third is cut off.
+			// Salvage keeps the two that completed instead of failing outright.
+			name: "truncated mid-array salvages completed objects",
+			raw:  `{"suggestions":[{"original":"I has","replacement":"I have","explanation":"subject-verb agreement","type":"grammar"},{"original":"a apple","replacement":"an apple","explanation":"use an before a vowel","type":"grammar"},{"original":"she relised","replacement":"she real`,
+			want: 2,
+		},
+		{
+			// A truncated array where even the first object is incomplete has
+			// nothing to salvage → still an error (the caller releases + retries).
+			name:    "truncated before any object closes",
+			raw:     `{"suggestions":[{"original":"It was a thursdy morning when Clara`,
+			wantErr: true,
+		},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
@@ -65,16 +80,31 @@ func TestParseCheckpoint(t *testing.T) {
 func TestRateLimiter(t *testing.T) {
 	rl := NewRateLimiter(30 * time.Second)
 
-	if ok, _ := rl.Allow("doc1"); !ok {
+	ok, _, at := rl.Allow("doc1")
+	if !ok {
 		t.Fatal("first call should be allowed")
 	}
-	if ok, retry := rl.Allow("doc1"); ok || retry <= 0 {
+	if ok, retry, _ := rl.Allow("doc1"); ok || retry <= 0 {
 		t.Fatalf("immediate second call should be throttled, got ok=%v retry=%v", ok, retry)
 	}
 	// A different document is independent.
-	if ok, _ := rl.Allow("doc2"); !ok {
+	if ok, _, _ := rl.Allow("doc2"); !ok {
 		t.Fatal("different doc should be allowed")
 	}
+
+	// Releasing doc1's slot (as a failed pass does) lets the next check re-run
+	// immediately instead of waiting out the interval.
+	rl.Release("doc1", at)
+	if ok, _, _ := rl.Allow("doc1"); !ok {
+		t.Fatal("after Release the next call should be allowed again")
+	}
+	// Release only rolls back its own slot: a stale timestamp must not evict the
+	// slot a newer Allow holds (so a late failure can't cancel a fresh in-flight
+	// check). doc2 still holds its slot from above; a zero-time Release is a no-op.
+	rl.Release("doc2", time.Time{})
+	if ok, _, _ := rl.Allow("doc2"); ok {
+		t.Fatal("stale Release must not free doc2's current slot")
+	}
 }
 
 func TestTruncateDoc(t *testing.T) {

internal/llm/prompts.go

@@ -165,3 +165,24 @@ func RewriteMessages(text, style string) []Message {
 		{Role: "user", Content: text},
 	}
 }
+
+// translateSystemPrompt drives the explanation translator: it renders a
+// suggestion's English explanation into Simplified Chinese so an ESL reader sees
+// the "why" in her first language. Strict about returning ONLY the translation
+// (no quotes, no pinyin, no English echo) so it can drop straight into the chat
+// bubble. Kept warm and plain — these are short, friendly one-liners.
+const translateSystemPrompt = `You are Petal, a warm writing assistant. Translate the English text the user ` +
+	`sends into natural, friendly Simplified Chinese (Mandarin). It is a short explanation of a writing ` +
+	`suggestion, written for a native Chinese speaker learning English.
+
+Respond with ONLY the Simplified Chinese translation. No quotation marks, no pinyin, no English, no preamble — ` +
+	`just the translated sentence.`
+
+// TranslateMessages builds the message array for translating one short English
+// explanation into Simplified Chinese.
+func TranslateMessages(text string) []Message {
+	return []Message{
+		{Role: "system", Content: translateSystemPrompt},
+		{Role: "user", Content: text},
+	}
+}

internal/llm/translate.go

@@ -0,0 +1,23 @@
+package llm
+
+import (
+	"context"
+)
+
+// RunTranslate renders a short English explanation into Simplified Chinese. It
+// is a one-shot Complete (the result seeds the Ask Petal bubble), kept at a low
+// temperature so the translation is faithful rather than creative. Output is
+// trimmed of any stray surrounding quotes the model may add.
+func RunTranslate(ctx context.Context, client LLMClient, text string) (string, error) {
+	out, err := client.Complete(ctx, CompletionRequest{
+		Messages:          TranslateMessages(text),
+		MaxTokens:         512,
+		Temperature:       0.2,
+		TopP:              0.9,
+		RepetitionPenalty: 1.1,
+	})
+	if err != nil {
+		return "", err
+	}
+	return cleanRewrite(out), nil
+}

internal/suggestions/handlers.go

@@ -55,6 +55,7 @@ func (h *Handler) Routes() chi.Router {
 	r.Post("/{id}/accept", h.accept)
 	r.Post("/{id}/dismiss", h.dismiss)
 	r.Post("/{id}/chat", h.chat)
+	r.Post("/{id}/translate", h.translate)
 	return r
 }
 
@@ -101,7 +102,8 @@ func (h *Handler) runPass(w http.ResponseWriter, r *http.Request, limiter *llm.R
 		return
 	}
 
-	if ok, _ := limiter.Allow(docID); !ok {
+	ok, _, slotAt := limiter.Allow(docID)
+	if !ok {
 		// Throttled: return the existing pending set unchanged rather than an
 		// error, so the frontend keeps showing current suggestions.
 		existing, err := h.fetchPending(docID)
@@ -115,6 +117,10 @@ func (h *Handler) runPass(w http.ResponseWriter, r *http.Request, limiter *llm.R
 
 	raw, err := run(r.Context(), h.Client, contentText, tone)
 	if err != nil {
+		// Allow ran before the model call, so a failed pass would otherwise hold
+		// the per-document slot for the full interval — stranding the frontend's
+		// auto-retry on the throttle path. Release it so a retry can re-run.
+		limiter.Release(docID, slotAt)
 		errorJSON(w, http.StatusBadGateway, "llm pass failed: "+err.Error())
 		return
 	}

internal/suggestions/translate.go

@@ -0,0 +1,57 @@
+package suggestions
+
+import (
+	"database/sql"
+	"errors"
+	"net/http"
+	"strings"
+
+	"github.com/go-chi/chi/v5"
+
+	"gitea.parodia.dev/drwily/petal/internal/db"
+	"gitea.parodia.dev/drwily/petal/internal/llm"
+)
+
+type translateResponse struct {
+	Translation string `json:"translation"`
+}
+
+// translate renders a suggestion's English explanation into Simplified Chinese
+// for the Ask Petal bubble, so the ESL reader sees the "why" in her first
+// language instead of a second copy of the same English text. The explanation is
+// loaded server-side from the suggestion id (scoped to the local user) and never
+// trusted from the client, mirroring chat (spec Note #10).
+func (h *Handler) translate(w http.ResponseWriter, r *http.Request) {
+	sugID := chi.URLParam(r, "id")
+
+	var explanation string
+	err := h.DB.QueryRow(
+		`SELECT s.explanation
+		   FROM suggestions s
+		   JOIN documents d ON d.id = s.doc_id
+		  WHERE s.id = ? AND d.user_id = ?`,
+		sugID, db.LocalUserID,
+	).Scan(&explanation)
+	if errors.Is(err, sql.ErrNoRows) {
+		errorJSON(w, http.StatusNotFound, "suggestion not found")
+		return
+	}
+	if err != nil {
+		serverError(w, err)
+		return
+	}
+
+	explanation = strings.TrimSpace(explanation)
+	if explanation == "" {
+		writeJSON(w, http.StatusOK, translateResponse{Translation: ""})
+		return
+	}
+
+	out, err := llm.RunTranslate(r.Context(), h.Client, explanation)
+	if err != nil {
+		errorJSON(w, http.StatusBadGateway, "translate failed: "+err.Error())
+		return
+	}
+
+	writeJSON(w, http.StatusOK, translateResponse{Translation: out})
+}

internal/tts/handler.go

@@ -0,0 +1,264 @@
+// Package tts implements read-aloud: it proxies short passages to a local Piper
+// HTTP server (a fast, offline neural TTS), optionally transcodes Piper's WAV to
+// mp3/opus with ffmpeg, and serves the audio back to the editor. Synthesized
+// clips are content-addressed on disk so tapping the same word twice is instant
+// and never re-synthesizes. The feature is entirely optional: when TTS_ENDPOINT
+// is unset the route isn't mounted and the frontend falls back to the browser's
+// Web Speech API.
+package tts
+
+import (
+	"bytes"
+	"context"
+	"crypto/sha256"
+	"encoding/hex"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"time"
+	"unicode/utf8"
+
+	"github.com/go-chi/chi/v5"
+
+	"gitea.parodia.dev/drwily/petal/internal/config"
+)
+
+// maxTextBytes bounds a single synthesis request. Read-aloud is for a word or a
+// sentence or two, not whole documents; this keeps one tap from pinning Piper.
+const maxTextBytes = 4000
+
+// lengthScale slows Piper slightly below natural pace — a learner following along
+// with the words, mirroring the old utterance.rate = 0.95. Higher = slower.
+const lengthScale = 1.1
+
+// audioFormat describes one output encoding: the cache-file extension, the
+// response Content-Type, and the ffmpeg args that turn Piper's WAV (on stdin)
+// into this format (on stdout). A nil ffmpegArgs means "serve the WAV as-is".
+type audioFormat struct {
+	ext         string
+	contentType string
+	ffmpegArgs  []string
+}
+
+// formats maps the configured TTS_AUDIO_FORMAT to its encoding. mp3 is the safe
+// default (every browser plays it); opus is ~half the size for speech but has
+// patchier Safari support; wav skips ffmpeg entirely.
+var formats = map[string]audioFormat{
+	"wav": {ext: ".wav", contentType: "audio/wav"},
+	"mp3": {ext: ".mp3", contentType: "audio/mpeg",
+		ffmpegArgs: []string{"-f", "mp3", "-c:a", "libmp3lame", "-b:a", "64k", "-ac", "1"}},
+	"opus": {ext: ".opus", contentType: "audio/ogg",
+		ffmpegArgs: []string{"-f", "ogg", "-c:a", "libopus", "-b:a", "32k", "-ac", "1"}},
+}
+
+// route is the Piper instance and voice id serving one language. Each Piper
+// HTTP server loads exactly one model, so distinct languages mean distinct
+// endpoints (e.g. English on :5005, Chinese on :5006).
+type route struct {
+	endpoint string // Piper HTTP base URL (no trailing slash)
+	voice    string // Piper voice id (sent in the request; also part of the cache key)
+}
+
+// Handler proxies synthesis to Piper and caches the result on disk.
+type Handler struct {
+	routes   map[string]route // base language (e.g. "en", "zh") -> Piper instance
+	cacheDir string
+	format   audioFormat
+	client   *http.Client
+}
+
+// New builds a Handler from config. It returns (nil, false) when TTS_ENDPOINT is
+// unset, signalling main to skip mounting the route. An unknown TTS_AUDIO_FORMAT
+// falls back to mp3 rather than failing the whole server.
+func New(cfg *config.Config) (*Handler, bool) {
+	if strings.TrimSpace(cfg.TTSEndpoint) == "" {
+		return nil, false
+	}
+
+	format, ok := formats[strings.ToLower(strings.TrimSpace(cfg.TTSFormat))]
+	if !ok {
+		format = formats["mp3"]
+	}
+
+	// Map by base language so en-US, en-GB, etc. all resolve to the English
+	// instance (the client sends BCP-47 tags like the old Web Speech path did).
+	// A language is only routable when both its endpoint and voice are set;
+	// otherwise the client falls back to Web Speech for that language.
+	routes := map[string]route{}
+	if cfg.TTSVoiceEN != "" {
+		routes["en"] = route{strings.TrimRight(cfg.TTSEndpoint, "/"), cfg.TTSVoiceEN}
+	}
+	if cfg.TTSEndpointZH != "" && cfg.TTSVoiceZH != "" {
+		routes["zh"] = route{strings.TrimRight(cfg.TTSEndpointZH, "/"), cfg.TTSVoiceZH}
+	}
+
+	if err := os.MkdirAll(cfg.TTSCacheDir, 0o755); err != nil {
+		// A missing cache dir isn't fatal — synthesis still works, it just won't
+		// cache. Disable the feature only on endpoint absence, not this.
+		fmt.Fprintf(os.Stderr, "tts: cache dir %q: %v\n", cfg.TTSCacheDir, err)
+	}
+
+	return &Handler{
+		routes:   routes,
+		cacheDir: cfg.TTSCacheDir,
+		format:   format,
+		client:   &http.Client{Timeout: cfg.TTSTimeout},
+	}, true
+}
+
+// Routes mounts the synthesis endpoint. Mount under "/tts" so the full path is
+// POST /api/tts.
+func (h *Handler) Routes() chi.Router {
+	r := chi.NewRouter()
+	r.Post("/", h.synth)
+	return r
+}
+
+// synthRequest is the body the editor posts: a passage and the BCP-47 language
+// tag it's written in (e.g. "en-US", "zh-CN").
+type synthRequest struct {
+	Text string `json:"text"`
+	Lang string `json:"lang"`
+}
+
+// synth resolves a voice for the requested language, returns cached audio when
+// present, and otherwise asks Piper to synthesize, transcodes if configured,
+// caches, and serves. An unconfigured language yields 404 so the client can fall
+// back to Web Speech without treating it as an error.
+func (h *Handler) synth(w http.ResponseWriter, r *http.Request) {
+	var req synthRequest
+	// Cap the raw body generously above maxTextBytes to leave room for JSON
+	// framing and the lang field; the text itself is truncated after decoding.
+	if err := json.NewDecoder(io.LimitReader(r.Body, maxTextBytes+1024)).Decode(&req); err != nil {
+		http.Error(w, "invalid request body", http.StatusBadRequest)
+		return
+	}
+	text := strings.TrimSpace(req.Text)
+	if text == "" {
+		http.Error(w, "empty text", http.StatusBadRequest)
+		return
+	}
+	if len(text) > maxTextBytes {
+		// Trim back to the last valid rune boundary so we never hand Piper a
+		// half-encoded multibyte character (common with Chinese, 3 bytes/char).
+		text = text[:maxTextBytes]
+		for len(text) > 0 && !utf8.ValidString(text) {
+			text = text[:len(text)-1]
+		}
+	}
+
+	rt, ok := h.routes[baseLang(req.Lang)]
+	if !ok {
+		// No voice for this language — let the client fall back to Web Speech.
+		http.Error(w, "no voice for language", http.StatusNotFound)
+		return
+	}
+
+	// Content-addressed: identical (voice, text) → identical clip. The format
+	// extension keeps encodings from colliding in the same dir.
+	sum := sha256.Sum256([]byte(rt.voice + "\n" + text))
+	name := hex.EncodeToString(sum[:])[:32] + h.format.ext
+	path := filepath.Join(h.cacheDir, name)
+
+	if _, err := os.Stat(path); err == nil {
+		h.serve(w, r, path)
+		return
+	}
+
+	audio, err := h.synthesize(r.Context(), rt, text)
+	if err != nil {
+		http.Error(w, "synthesis failed", http.StatusBadGateway)
+		fmt.Fprintf(os.Stderr, "tts: synthesize: %v\n", err)
+		return
+	}
+
+	// Best-effort cache write via a temp file + rename so a concurrent reader
+	// never sees a half-written clip. A failed write just means no caching.
+	if h.cacheDir != "" {
+		tmp := path + ".tmp"
+		if err := os.WriteFile(tmp, audio, 0o644); err == nil {
+			_ = os.Rename(tmp, path)
+		}
+	}
+
+	w.Header().Set("Content-Type", h.format.contentType)
+	w.Header().Set("Cache-Control", "public, max-age=31536000, immutable")
+	_, _ = w.Write(audio)
+}
+
+// serve streams a cached clip with a long-lived immutable cache header (the URL
+// is content-addressed, so the bytes never change for a given request).
+func (h *Handler) serve(w http.ResponseWriter, r *http.Request, path string) {
+	w.Header().Set("Content-Type", h.format.contentType)
+	w.Header().Set("Cache-Control", "public, max-age=31536000, immutable")
+	http.ServeFile(w, r, path)
+}
+
+// synthesize POSTs to the route's Piper instance, then transcodes the returned
+// WAV when the configured format calls for it.
+func (h *Handler) synthesize(ctx context.Context, rt route, text string) ([]byte, error) {
+	body, _ := json.Marshal(map[string]any{
+		"text":         text,
+		"voice":        rt.voice,
+		"length_scale": lengthScale,
+	})
+	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, rt.endpoint+"/", bytes.NewReader(body))
+	if err != nil {
+		return nil, err
+	}
+	httpReq.Header.Set("Content-Type", "application/json")
+
+	resp, err := h.client.Do(httpReq)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		snippet, _ := io.ReadAll(io.LimitReader(resp.Body, 256))
+		return nil, fmt.Errorf("piper %d: %s", resp.StatusCode, strings.TrimSpace(string(snippet)))
+	}
+	wav, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, err
+	}
+
+	if h.format.ffmpegArgs == nil {
+		return wav, nil
+	}
+	return transcode(ctx, wav, h.format.ffmpegArgs)
+}
+
+// transcode pipes WAV bytes through ffmpeg (stdin → stdout) into the target
+// encoding. ffmpeg is assumed on PATH; an error here surfaces as a 502.
+func transcode(ctx context.Context, wav []byte, args []string) ([]byte, error) {
+	// Guard ffmpeg against a hang independent of the HTTP client timeout.
+	ctx, cancel := context.WithTimeout(ctx, 30*time.Second)
+	defer cancel()
+
+	full := append([]string{"-hide_banner", "-loglevel", "error", "-i", "pipe:0"}, args...)
+	full = append(full, "pipe:1")
+	cmd := exec.CommandContext(ctx, "ffmpeg", full...)
+	cmd.Stdin = bytes.NewReader(wav)
+	var out, errBuf bytes.Buffer
+	cmd.Stdout = &out
+	cmd.Stderr = &errBuf
+	if err := cmd.Run(); err != nil {
+		return nil, fmt.Errorf("ffmpeg: %v: %s", err, strings.TrimSpace(errBuf.String()))
+	}
+	return out.Bytes(), nil
+}
+
+// baseLang reduces a BCP-47 tag to its primary subtag, lowercased: "en-US" → "en",
+// "zh-CN" → "zh", "" → "". Mirrors the client's pickVoice base-language matching.
+func baseLang(tag string) string {
+	tag = strings.ToLower(strings.TrimSpace(tag))
+	if i := strings.IndexAny(tag, "-_"); i >= 0 {
+		return tag[:i]
+	}
+	return tag
+}

internal/tts/handler_test.go

@@ -0,0 +1,170 @@
+package tts
+
+import (
+	"bytes"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
+	"testing"
+)
+
+// newStubPiper returns a fake Piper server that echoes a fixed WAV body and
+// records how many times it was called and the last request payload.
+func newStubPiper(t *testing.T, body []byte) (*httptest.Server, *int32, *synthEcho) {
+	t.Helper()
+	var calls int32
+	last := &synthEcho{}
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		atomic.AddInt32(&calls, 1)
+		var req map[string]any
+		_ = json.NewDecoder(r.Body).Decode(&req)
+		last.voice, _ = req["voice"].(string)
+		last.text, _ = req["text"].(string)
+		w.Header().Set("Content-Type", "audio/wav")
+		_, _ = w.Write(body)
+	}))
+	t.Cleanup(srv.Close)
+	return srv, &calls, last
+}
+
+type synthEcho struct{ voice, text string }
+
+// newHandler builds a wav-format handler (no ffmpeg) pointed at a stub server.
+func newHandler(t *testing.T, endpoint string) *Handler {
+	t.Helper()
+	return &Handler{
+		routes:   map[string]route{"en": {strings.TrimRight(endpoint, "/"), "en_US-amy-medium"}},
+		cacheDir: t.TempDir(),
+		format:   formats["wav"],
+		client:   http.DefaultClient,
+	}
+}
+
+func post(t *testing.T, h *Handler, text, lang string) *httptest.ResponseRecorder {
+	t.Helper()
+	b, _ := json.Marshal(synthRequest{Text: text, Lang: lang})
+	req := httptest.NewRequest(http.MethodPost, "/", bytes.NewReader(b))
+	rr := httptest.NewRecorder()
+	h.synth(rr, req)
+	return rr
+}
+
+func TestSynthSuccessAndVoiceSelection(t *testing.T) {
+	wav := []byte("RIFF....fake-wav")
+	srv, calls, last := newStubPiper(t, wav)
+	h := newHandler(t, srv.URL)
+
+	rr := post(t, h, "hello there", "en-US")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("status = %d, want 200", rr.Code)
+	}
+	if got := rr.Body.Bytes(); !bytes.Equal(got, wav) {
+		t.Fatalf("body = %q, want the piper wav", got)
+	}
+	if ct := rr.Header().Get("Content-Type"); ct != "audio/wav" {
+		t.Fatalf("content-type = %q, want audio/wav", ct)
+	}
+	if last.voice != "en_US-amy-medium" {
+		t.Fatalf("piper voice = %q, want en_US-amy-medium", last.voice)
+	}
+	if *calls != 1 {
+		t.Fatalf("piper calls = %d, want 1", *calls)
+	}
+}
+
+func TestRoutesByLanguageToSeparateInstances(t *testing.T) {
+	enSrv, enCalls, _ := newStubPiper(t, []byte("EN-wav"))
+	zhSrv, zhCalls, zhLast := newStubPiper(t, []byte("ZH-wav"))
+	h := &Handler{
+		routes: map[string]route{
+			"en": {strings.TrimRight(enSrv.URL, "/"), "en_US-amy-medium"},
+			"zh": {strings.TrimRight(zhSrv.URL, "/"), "zh_CN-huayan-medium"},
+		},
+		cacheDir: t.TempDir(),
+		format:   formats["wav"],
+		client:   http.DefaultClient,
+	}
+
+	if rr := post(t, h, "你好世界", "zh-CN"); rr.Code != http.StatusOK {
+		t.Fatalf("zh status = %d, want 200", rr.Code)
+	}
+	if *zhCalls != 1 || *enCalls != 0 {
+		t.Fatalf("calls en=%d zh=%d, want en=0 zh=1 (zh routed to zh instance)", *enCalls, *zhCalls)
+	}
+	if zhLast.voice != "zh_CN-huayan-medium" {
+		t.Fatalf("zh voice = %q, want zh_CN-huayan-medium", zhLast.voice)
+	}
+}
+
+func TestUnknownLanguageReturns404(t *testing.T) {
+	srv, calls, _ := newStubPiper(t, []byte("x"))
+	h := newHandler(t, srv.URL)
+
+	rr := post(t, h, "你好", "zh-CN") // only "en" is configured
+	if rr.Code != http.StatusNotFound {
+		t.Fatalf("status = %d, want 404", rr.Code)
+	}
+	if *calls != 0 {
+		t.Fatalf("piper calls = %d, want 0 (should not synthesize)", *calls)
+	}
+}
+
+func TestCacheHitSkipsPiper(t *testing.T) {
+	srv, calls, _ := newStubPiper(t, []byte("RIFF....fake-wav"))
+	h := newHandler(t, srv.URL)
+
+	if rr := post(t, h, "same words", "en-US"); rr.Code != http.StatusOK {
+		t.Fatalf("first status = %d, want 200", rr.Code)
+	}
+	if rr := post(t, h, "same words", "en-US"); rr.Code != http.StatusOK {
+		t.Fatalf("second status = %d, want 200", rr.Code)
+	}
+	if *calls != 1 {
+		t.Fatalf("piper calls = %d, want 1 (second served from cache)", *calls)
+	}
+}
+
+func TestEmptyTextReturns400(t *testing.T) {
+	srv, _, _ := newStubPiper(t, []byte("x"))
+	h := newHandler(t, srv.URL)
+
+	if rr := post(t, h, "   ", "en-US"); rr.Code != http.StatusBadRequest {
+		t.Fatalf("status = %d, want 400", rr.Code)
+	}
+}
+
+func TestTextIsCapped(t *testing.T) {
+	srv, _, last := newStubPiper(t, []byte("x"))
+	h := newHandler(t, srv.URL)
+
+	long := strings.Repeat("a", maxTextBytes+500)
+	if rr := post(t, h, long, "en-US"); rr.Code != http.StatusOK {
+		t.Fatalf("status = %d, want 200", rr.Code)
+	}
+	if len(last.text) != maxTextBytes {
+		t.Fatalf("piper received %d chars, want capped to %d", len(last.text), maxTextBytes)
+	}
+}
+
+func TestBaseLang(t *testing.T) {
+	cases := map[string]string{"en-US": "en", "EN_gb": "en", "zh-CN": "zh", "en": "en", "": ""}
+	for in, want := range cases {
+		if got := baseLang(in); got != want {
+			t.Errorf("baseLang(%q) = %q, want %q", in, got, want)
+		}
+	}
+}
+
+// ensure the stub's body is fully consumable (guards against the LimitReader cap
+// accidentally truncating a normal request body in synth()).
+func TestRequestBodyNotTruncated(t *testing.T) {
+	srv, _, last := newStubPiper(t, []byte("x"))
+	h := newHandler(t, srv.URL)
+	text := strings.Repeat("word ", 200) // ~1000 bytes, under the cap
+	post(t, h, text, "en-US")
+	if last.text != strings.TrimSpace(text) {
+		t.Fatalf("piper text length = %d, want %d", len(last.text), len(strings.TrimSpace(text)))
+	}
+}

web/src/App.tsx

@@ -14,10 +14,15 @@ import { PetalCompanion } from './components/Companion/PetalCompanion'
 import { UpdateBanner } from './components/UpdateBanner/UpdateBanner'
 import { useVersionWatch } from './hooks/useVersionWatch'
 import { PetalFall } from './effects/PetalFall'
+import { useNightMode } from './hooks/useNightMode'
 import { playSuggestionSound } from './audio/sounds'
 
 export default function App() {
   const updateAvailable = useVersionWatch()
+  // Late-night calm mode: dark theme + falling stars instead of petals. The hook
+  // toggles the `petal-night` class on <html>; we pass the flag to the ambient
+  // layer so the petals become stars.
+  const night = useNightMode()
   const [docs, setDocs] = useState<DocSummary[]>([])
   const [currentDoc, setCurrentDoc] = useState<Document | null>(null)
   const [title, setTitle] = useState('')
@@ -370,7 +375,7 @@ export default function App() {
 
   return (
     <div className="flex h-full flex-col">
-      <PetalFall />
+      <PetalFall night={night} />
       <header
         onMouseDown={handleChromeDown}
         className="petal-no-print flex h-12 shrink-0 items-center gap-2 px-5"

web/src/api/client.ts

@@ -147,6 +147,10 @@ export const api = {
     req<void>(`/suggestions/${id}/accept`, { method: 'POST' }),
   dismissSuggestion: (id: string) =>
     req<void>(`/suggestions/${id}/dismiss`, { method: 'POST' }),
+  // Simplified-Chinese rendering of a suggestion's explanation, for the Ask Petal
+  // opening bubble (the explanation itself stays English in the card body).
+  translateSuggestion: (id: string) =>
+    req<{ translation: string }>(`/suggestions/${id}/translate`, { method: 'POST' }),
 
   // Version history. listVersions returns metadata only (no bodies); getVersion
   // loads one full snapshot for preview; snapshotDoc takes an explicit restore

web/src/audio/speech.ts

@@ -1,12 +1,42 @@
-// Read-aloud via the browser's Web Speech API — an offline pronunciation aid for
-// an ESL writer: hear how an English word or passage sounds. No model, no
-// network, no extra weight in the binary. Feature-detected so the buttons that
-// use it can hide where speech isn't available.
+// Read-aloud for an ESL writer: hear how an English word or passage sounds.
+//
+// Primary path is server-side neural TTS (Piper, proxied at /api/tts) — natural,
+// consistent across devices, and far better than the browser's built-in voices
+// (which fall back to robotic espeak on Chromium). If the server route is absent
+// (TTS disabled) or unreachable, we fall back to the browser's Web Speech API so
+// the buttons still do something. No model or network is strictly required.
 
+// speechSupported reports whether read-aloud can do anything at all. Audio
+// playback is universal, so as long as we can construct an Audio element OR the
+// Web Speech API exists, the buttons should show. The server path is tried at
+// call time and degrades on its own.
 export function speechSupported(): boolean {
+  if (typeof window === 'undefined') return false
+  return typeof window.Audio === 'function' || 'speechSynthesis' in window
+}
+
+// webSpeechSupported gates the fallback path specifically.
+function webSpeechSupported(): boolean {
   return typeof window !== 'undefined' && 'speechSynthesis' in window && 'SpeechSynthesisUtterance' in window
 }
 
+// current holds the in-flight server-audio element and its object URL so a rapid
+// re-tap can stop and replace it (mirroring speechSynthesis.cancel()).
+let current: { audio: HTMLAudioElement; url: string } | null = null
+
+// requestSeq increments on every speak() call so a slow fetch that resolves after
+// a newer tap can detect it's stale and bow out instead of double-playing.
+let requestSeq = 0
+
+function stopCurrent(): void {
+  if (current) {
+    current.audio.pause()
+    URL.revokeObjectURL(current.url)
+    current = null
+  }
+  if (webSpeechSupported()) window.speechSynthesis.cancel()
+}
+
 // pickVoice prefers an exact locale match (e.g. en-US), then any voice in the
 // same language family (en-*). Returns undefined to let the engine default.
 function pickVoice(lang: string): SpeechSynthesisVoice | undefined {
@@ -15,12 +45,10 @@ function pickVoice(lang: string): SpeechSynthesisVoice | undefined {
   return voices.find((v) => v.lang === lang) || voices.find((v) => v.lang?.startsWith(base))
 }
 
-// speak reads `text` aloud, cancelling anything already in flight so rapid taps
-// don't queue up. `lang` defaults to US English (the language she's learning);
-// pass a zh locale to hear a Chinese passage. A touch slower than default so
-// learners can follow along.
-export function speak(text: string, lang = 'en-US'): void {
-  if (!speechSupported() || !text.trim()) return
+// speakWebSpeech is the fallback: the browser's built-in synthesizer. A touch
+// slower than default so learners can follow along.
+function speakWebSpeech(text: string, lang: string): void {
+  if (!webSpeechSupported()) return
   const synth = window.speechSynthesis
   synth.cancel()
   const utterance = new SpeechSynthesisUtterance(text)
@@ -30,3 +58,58 @@ export function speak(text: string, lang = 'en-US'): void {
   utterance.rate = 0.95
   synth.speak(utterance)
 }
+
+// detectLang guesses a BCP-47 locale from the text so the right Piper voice is
+// chosen. Any Han/kana character routes to Chinese (zh-CN) — a selection is
+// essentially never mixed at the level that matters, and reading a Chinese
+// passage with the English voice spells each character out ("Chinese letter…").
+// Everything else defaults to US English, the language she's learning.
+const CJK = /[぀-ヿ㐀-䶿一-鿿豈-﫿ｦ-ﾟ]/
+export function detectLang(text: string): string {
+  return CJK.test(text) ? 'zh-CN' : 'en-US'
+}
+
+// speak reads `text` aloud, cancelling anything already in flight so rapid taps
+// don't queue up. `lang` defaults to a guess from the text (Chinese vs English)
+// so callers can just pass the selection; pass an explicit locale to override.
+// It tries the server's neural voice first and silently falls back to the browser
+// voice if that's unavailable (route off, network error, or a 404 for a language
+// with no configured voice).
+export function speak(text: string, lang = detectLang(text)): void {
+  if (!text.trim()) return
+  stopCurrent()
+  const seq = ++requestSeq
+
+  fetch('/api/tts', {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({ text, lang }),
+  })
+    .then((res) => {
+      if (!res.ok) throw new Error(`tts ${res.status}`)
+      return res.blob()
+    })
+    .then((blob) => {
+      // A newer tap superseded this one while the fetch was in flight — drop it.
+      if (seq !== requestSeq) return
+      const url = URL.createObjectURL(blob)
+      const audio = new Audio(url)
+      current = { audio, url }
+      // Release the blob once playback finishes (or errors) to avoid leaking.
+      const cleanup = () => {
+        if (current?.audio === audio) {
+          URL.revokeObjectURL(url)
+          current = null
+        }
+      }
+      audio.addEventListener('ended', cleanup)
+      audio.addEventListener('error', cleanup)
+      return audio.play()
+    })
+    .catch(() => {
+      // Server TTS unavailable for this request — use the browser voice instead,
+      // unless a newer tap has already superseded this one.
+      if (seq !== requestSeq) return
+      speakWebSpeech(text, lang)
+    })
+}

web/src/components/Companion/tips.ts

@@ -15,6 +15,11 @@ export const ENCOURAGEMENTS: Line[] = [
   { zh: '我很喜欢这个改法 💕', en: 'I really like that change.' },
   { zh: '继续保持，加油！', en: 'Keep going — you’ve got this!' },
   { zh: '嗯嗯，这样清楚多了 👍', en: 'Mm, that’s much clearer.' },
+  { zh: '这个词用得真好 🌷', en: 'That’s such a good word choice.' },
+  { zh: '哇，这一段读起来真舒服 ☁️', en: 'Ooh, that paragraph flows so nicely.' },
+  { zh: '看你越写越有信心，真好 💛', en: 'I love watching you write with more confidence.' },
+  { zh: '一点点进步，都是了不起的进步 🌱', en: 'Every little bit of progress counts.' },
+  { zh: '今天的你，文字闪闪发光 ✨', en: 'Your words are sparkling today.' },
 ]
 
 // Gentle, generic writing/ESL tips — the fallback when the rules-based prose
@@ -36,6 +41,23 @@ export const BREAKS: Line[] = [
   { zh: '看看远方，放松一下眼睛 🌿', en: 'Look into the distance for a moment — give your eyes a break. 🌿' },
 ]
 
+// Shown when she's still writing late at night (≥11pm). Caring, a little
+// playful — the kitten is always asleep, so "you should be too" lands as a gag,
+// never a scold. zh stays gentle; the English subtitle carries the wink.
+export const BEDTIME: Line[] = [
+  { zh: '你的床在想你了哦 🛏️', en: 'I bet your bed is missing you right now.' },
+  { zh: '太累可写不出好文字呀，早点歇着吧 🌙', en: 'A tired writer is a bad writer — get some rest.' },
+  { zh: '好好睡一觉，灵感自己会来 ✨', en: 'Sleep is a wondrous enabler.' },
+  { zh: '听见了吗？没有吧——大家都睡了，你也该睡啦 😴', en: "Hear that? No… you don't, because everyone is sleeping and you should be too." },
+  // A few old Chinese proverbs on sleep (zh = a faithful rendering of the
+  // English sense — the verified classical 原文 isn't recoverable; swap in the
+  // exact source text if you have it). They read a touch wittier/wiser than the
+  // gentle lines above, which suits a late-night nudge.
+  { zh: '一夜不眠，十日不安。', en: "The loss of one night's sleep is followed by ten days of inconvenience." },
+  { zh: '前半夜醒着想自己的过错，后半夜睡着才想别人的不是。', en: 'Think of your own faults the first part of the night when you are awake, and of the faults of others the latter part of the night when you are asleep.' },
+  { zh: '黄昏与人相骂，夜半独自难眠。', en: 'Curse your spouse at evening, sleep alone at night.' },
+]
+
 // First hello when the app opens.
 export const GREETING: Line = {
   zh: '嗨～我在这儿陪你写作哦 🐱',

web/src/components/Companion/useCompanion.ts

@@ -1,6 +1,7 @@
 import { useCallback, useEffect, useRef, useState } from 'react'
 import type { SaveStatus } from '../../hooks/useAutoSave'
 import {
+  BEDTIME,
   BREAKS,
   ENCOURAGEMENTS,
   ERRORS,
@@ -14,12 +15,13 @@ import {
 } from './tips'
 import { analyzeProse } from './prose'
 import { playPop, playSound, type SoundName } from '../../audio/sounds'
+import { isBedtime } from '../../lib/night'
 
 // The kitten's expression. Maps to a Lottie animation when assets are present,
 // otherwise to an emoji placeholder (see PetalCompanion).
 export type Mood = 'idle' | 'happy' | 'talking' | 'sleeping' | 'celebrate'
 
-export type BubbleTone = 'cheer' | 'tip' | 'break' | 'error'
+export type BubbleTone = 'cheer' | 'tip' | 'break' | 'error' | 'bedtime'
 export interface Bubble extends Line {
   tone: BubbleTone
 }
@@ -44,6 +46,9 @@ const IDLE_MS = 75_000 // no edits → kitten naps
 const BREAK_MS = 25 * 60_000 // continuous writing → suggest a break
 const TIP_MIN_GAP = 4 * 60_000 // at most one spontaneous tip per this window
 const PROACTIVE_GAP = 40_000 // floor between any two unsolicited bubbles
+const BEDTIME_GAP = 30 * 60_000 // at most one "go to bed" nudge per this window
+// The late-night window itself (isBedtime) lives in ../../lib/night so the
+// companion nag and the night-mode theme/starfall share one definition.
 // How long a bubble lingers. These are *floors* — readBubbleMs extends them by
 // how much there is to read, since she reads both the Mandarin and the English
 // (and tips now quote a slice of her own sentence, so they run longer).
@@ -58,7 +63,7 @@ const now = () => Date.now()
 // length of the Mandarin + English lines so denser advice stays up long enough
 // to actually finish reading.
 function readBubbleMs(b: Bubble): number {
-  const base = b.tone === 'cheer' ? CHEER_MS : BUBBLE_MS
+  const base = b.tone === 'cheer' ? CHEER_MS : b.tone === 'bedtime' ? BUBBLE_MS + 4_000 : BUBBLE_MS
   const chars = b.zh.length + b.en.length
   return Math.min(MAX_BUBBLE_MS, base + chars * READ_MS_PER_CHAR)
 }
@@ -75,6 +80,7 @@ export function useCompanion({ wordCount, saveStatus, llmDown, editTick, acceptT
   const lastProactive = useRef(0)
   const lastTip = useRef(0)
   const lastBreak = useRef(0)
+  const lastBedtime = useRef(0)
   const nextMilestone = useRef(0) // index into MILESTONES
   const sleeping = useRef(false)
 
@@ -262,6 +268,15 @@ export function useCompanion({ wordCount, saveStatus, llmDown, editTick, acceptT
         return
       }
 
+      // Burning the midnight oil? Gently suggest bed — caring, low-frequency,
+      // and only while she's actually still at it (the idle branch above already
+      // returned if she's away/napping).
+      if (isBedtime() && t - lastBedtime.current > BEDTIME_GAP) {
+        lastBedtime.current = t
+        say({ ...pick(BEDTIME), tone: 'bedtime' }, { proactive: true })
+        return
+      }
+
       // Otherwise an occasional tip — context-aware when her text gives us
       // something concrete to gently point at, generic warmth otherwise.
       if (t - lastTip.current > TIP_MIN_GAP) {

web/src/components/Editor/AskPetal.tsx

@@ -1,9 +1,11 @@
 import { useEffect, useRef, useState } from 'react'
-import { streamSuggestionChat, type ChatMessage } from '../../api/client'
+import { api, streamSuggestionChat, type ChatMessage } from '../../api/client'
 
 interface Props {
   suggestionId: string
-  // Pre-populates Petal's first bubble so the conversation opens with context.
+  // The English explanation (shown in the card body). Petal's opening bubble is
+  // its Simplified-Chinese translation, fetched on open — so the panel doesn't
+  // just repeat the same English text twice. Falls back to this on failure.
   explanation: string
 }
 
@@ -17,9 +19,10 @@ const CHAT_FONT = "'Nunito', 'PingFang SC', 'Microsoft YaHei', 'Noto Sans CJK SC
 // the card (unmounting) clears it. Each send streams Petal's reply token-by-
 // token into the latest assistant bubble.
 export function AskPetal({ suggestionId, explanation }: Props) {
-  const [messages, setMessages] = useState<ChatMessage[]>([
-    { role: 'assistant', content: explanation },
-  ])
+  // Opening bubble starts empty (caret-only) and fills with the Mandarin
+  // translation once it lands; `seeding` drives that loading caret.
+  const [messages, setMessages] = useState<ChatMessage[]>([{ role: 'assistant', content: '' }])
+  const [seeding, setSeeding] = useState(true)
   const [input, setInput] = useState('')
   const [streaming, setStreaming] = useState(false)
   const scrollRef = useRef<HTMLDivElement>(null)
@@ -36,6 +39,31 @@ export function AskPetal({ suggestionId, explanation }: Props) {
     inputRef.current?.focus()
   }, [])
 
+  // Fetch the Chinese translation of the explanation to seed the first bubble.
+  // Only replaces the seed bubble if the user hasn't started chatting yet (the
+  // conversation always opens with this one assistant turn). Falls back to the
+  // English explanation if the translation can't be fetched.
+  useEffect(() => {
+    let cancelled = false
+    api
+      .translateSuggestion(suggestionId)
+      .then((res) => {
+        if (cancelled) return
+        const text = res.translation.trim() || explanation
+        setMessages((prev) => (prev.length === 1 ? [{ role: 'assistant', content: text }] : prev))
+      })
+      .catch(() => {
+        if (cancelled) return
+        setMessages((prev) => (prev.length === 1 ? [{ role: 'assistant', content: explanation }] : prev))
+      })
+      .finally(() => {
+        if (!cancelled) setSeeding(false)
+      })
+    return () => {
+      cancelled = true
+    }
+  }, [suggestionId, explanation])
+
   async function send() {
     const text = input.trim()
     if (!text || streaming) return
@@ -86,7 +114,12 @@ export function AskPetal({ suggestionId, explanation }: Props) {
         style={{ maxHeight: 220 }}
       >
         {messages.map((m, i) => (
-          <Bubble key={i} role={m.role} content={m.content} streaming={streaming && i === messages.length - 1} />
+          <Bubble
+            key={i}
+            role={m.role}
+            content={m.content}
+            streaming={(streaming && i === messages.length - 1) || (seeding && i === 0)}
+          />
         ))}
       </div>
 

web/src/components/Editor/EditorCore.tsx

@@ -18,7 +18,8 @@ import type { EditorView } from '@tiptap/pm/view'
 import { useCallback, useEffect, useRef, useState } from 'react'
 import { Toolbar } from '../Toolbar/Toolbar'
 import { SuggestionCard } from './SuggestionCard'
-import { SuggestionHighlight, setSuggestions, findRange } from './SuggestionHighlight'
+import { SuggestionRail, type RailItem } from './SuggestionRail'
+import { SuggestionHighlight, setSuggestions, setActiveSuggestion, findRange } from './SuggestionHighlight'
 import { SpellCheck, setSpellChecker, wordAt } from './SpellCheck'
 import { MisspellCard } from './MisspellCard'
 import { WordCard } from './WordCard'
@@ -219,6 +220,18 @@ export function EditorCore({
   const rewriteReqRef = useRef(0)
   // The in-document Find & Replace bar (Ctrl/Cmd+F).
   const [findOpen, setFindOpen] = useState(false)
+  // The right-margin comment rail. `railItems` carries each anchored suggestion's
+  // vertical offset; `railEnabled` is true only when the viewport has room for the
+  // column beside the editor (otherwise we fall back to the inline hover cards).
+  // `railExpandedId` is the card showing its full explanation + Ask Petal, and
+  // `activeId` is the suggestion currently emphasized (hovered text or card).
+  const [railItems, setRailItems] = useState<RailItem[]>([])
+  const [railEnabled, setRailEnabled] = useState(false)
+  const [railExpandedId, setRailExpandedId] = useState<string | null>(null)
+  const [activeId, setActiveId] = useState<string | null>(null)
+  // A stable handle to the latest recompute so the editor's onUpdate (captured
+  // once at construction) can trigger a re-measure without stale closures.
+  const recomputeRailRef = useRef<() => void>(() => {})
 
   const editor = useEditor({
     extensions: [
@@ -282,6 +295,8 @@ export function EditorCore({
         content_text: editor.getText(),
         word_count: editor.storage.characterCount.words(),
       })
+      // Edits reflow the text, so the rail anchors need re-measuring.
+      recomputeRailRef.current()
     },
     onSelectionUpdate: ({ editor }) => {
       // A selection supersedes the hover gloss; an empty one clears the bubble.
@@ -333,8 +348,87 @@ export function EditorCore({
     setSuggestions(editor.state, editor.view.dispatch, suggestions)
     // Close the card if its suggestion is gone.
     setHover((h) => (h && suggestions.some((s) => s.id === h.suggestion.id) ? h : null))
+    // Drop any rail expand/emphasis that points at a now-removed suggestion.
+    setRailExpandedId((id) => (id && suggestions.some((s) => s.id === id) ? id : null))
+    setActiveId((id) => (id && suggestions.some((s) => s.id === id) ? id : null))
   }, [editor, suggestions])
 
+  // Re-measure the margin rail: whether there's room for the column beside the
+  // editor, and where each suggestion's highlight sits vertically. Anchors are
+  // taken from the live decoration DOM (keyed by data-suggestion-id) relative to
+  // the wrapper — stable under scroll since text and wrapper scroll together.
+  // Deferred a frame so decoration DOM and reflow have settled.
+  const recomputeRail = useCallback(() => {
+    requestAnimationFrame(() => {
+      const wrapper = wrapperRef.current
+      if (!wrapper) return
+      const wrapRect = wrapper.getBoundingClientRect()
+      // Need room for the 300px column + its 32px gutter (see .petal-rail), plus
+      // a little breathing space to the viewport edge.
+      setRailEnabled(window.innerWidth - wrapRect.right >= 348)
+      const seen = new Set<string>()
+      const items: RailItem[] = []
+      wrapper.querySelectorAll<HTMLElement>('.petal-suggestion[data-suggestion-id]').forEach((el) => {
+        const id = el.getAttribute('data-suggestion-id')
+        if (!id || seen.has(id)) return
+        const s = suggestions.find((x) => x.id === id)
+        if (!s) return
+        seen.add(id)
+        items.push({ suggestion: s, anchorTop: el.getBoundingClientRect().top - wrapRect.top })
+      })
+      setRailItems(items)
+    })
+  }, [suggestions])
+
+  useEffect(() => {
+    recomputeRailRef.current = recomputeRail
+  }, [recomputeRail])
+
+  // Re-anchor when the suggestion set changes (after the decorations repaint),
+  // and keep the rail in sync with viewport/editor width changes (room + reflow).
+  useEffect(() => {
+    recomputeRail()
+    const wrapper = wrapperRef.current
+    const ro = wrapper ? new ResizeObserver(() => recomputeRail()) : null
+    if (wrapper && ro) ro.observe(wrapper)
+    window.addEventListener('resize', recomputeRail)
+    return () => {
+      ro?.disconnect()
+      window.removeEventListener('resize', recomputeRail)
+    }
+  }, [recomputeRail])
+
+  // Emphasize the flagged text for the active suggestion, mirroring the rail
+  // card ↔ text link both ways. Driven through the decoration plugin (not an
+  // imperative DOM class) so it survives the repaints that fire on every edit.
+  useEffect(() => {
+    if (!editor) return
+    setActiveSuggestion(editor.state, editor.view.dispatch, railEnabled ? activeId : null)
+  }, [editor, activeId, railEnabled])
+
+  // Scroll a suggestion's highlight to the middle of the viewport (clicking its
+  // rail card jumps the editor to the flagged text).
+  const scrollToSuggestion = useCallback((id: string) => {
+    const el = wrapperRef.current?.querySelector(
+      `.petal-suggestion[data-suggestion-id="${CSS.escape(id)}"]`,
+    )
+    el?.scrollIntoView({ block: 'center', behavior: 'smooth' })
+  }, [])
+
+  // Clicking a rail card's body jumps to the text and toggles its detail panel.
+  const activateRailCard = useCallback(
+    (id: string) => {
+      setActiveId(id)
+      scrollToSuggestion(id)
+      setRailExpandedId((cur) => (cur === id ? null : id))
+    },
+    [scrollToSuggestion],
+  )
+
+  const toggleRailExpand = useCallback((id: string) => {
+    setRailExpandedId((cur) => (cur === id ? null : id))
+  }, [])
+
   const openCardFor = useCallback(
     (id: string, el: HTMLElement) => {
       const wrapper = wrapperRef.current
@@ -366,10 +460,17 @@ export function EditorCore({
       if (!target) return
       const id = target.getAttribute('data-suggestion-id')
       if (!id) return
+      // With the rail open the card already lives in the margin — hovering the
+      // text just emphasizes its card (and the highlight) rather than popping a
+      // second, redundant floating card.
+      if (railEnabled) {
+        setActiveId(id)
+        return
+      }
       clearTimeout(closeTimer.current)
       openCardFor(id, target)
     },
-    [openCardFor],
+    [openCardFor, railEnabled],
   )
 
   const scheduleClose = useCallback(() => {
@@ -389,35 +490,53 @@ export function EditorCore({
   // pointer can bridge the small gap from text to card.
   const handleMouseOut = useCallback(
     (e: React.MouseEvent) => {
-      if ((e.target as HTMLElement).closest('.petal-suggestion')) scheduleClose()
+      if (!(e.target as HTMLElement).closest('.petal-suggestion')) return
+      if (railEnabled) {
+        setActiveId(null)
+        return
+      }
+      scheduleClose()
     },
-    [scheduleClose],
+    [scheduleClose, railEnabled],
   )
 
   const keepOpen = useCallback(() => clearTimeout(closeTimer.current), [])
 
   // Accept applies the replacement to the document, plays a little confetti
-  // burst where the card sat, then notifies the parent.
+  // burst over the flagged text, then notifies the parent. The confetti is
+  // anchored to the highlight itself (captured before the replacement removes it),
+  // so it fires in the right spot whether the accept came from the hover card or
+  // the margin rail.
   const handleAccept = useCallback(
     (s: Suggestion) => {
+      const wrapper = wrapperRef.current
+      const el = wrapper?.querySelector(
+        `.petal-suggestion[data-suggestion-id="${CSS.escape(s.id)}"]`,
+      ) as HTMLElement | null
+      let burst: { top: number; left: number } | null = null
+      if (wrapper && el) {
+        const wrapRect = wrapper.getBoundingClientRect()
+        const elRect = el.getBoundingClientRect()
+        burst = { top: elRect.top - wrapRect.top, left: elRect.right - wrapRect.left }
+      }
       if (editor && s.replacement.trim() !== '') {
         const range = findRange(editor.state.doc, s.original)
         if (range) {
           editor.chain().focus().insertContentAt(range, s.replacement).run()
         }
       }
-      setHover((h) => {
-        if (h) {
-          setConfetti({ top: h.top, left: h.left + 16 })
-          clearTimeout(confettiTimer.current)
-          confettiTimer.current = setTimeout(() => setConfetti(null), 720)
-        }
-        return h
-      })
+      // Fall back to the hover card's position if the highlight wasn't found.
+      if (!burst && hover) burst = { top: hover.top, left: hover.left + 16 }
+      if (burst) {
+        setConfetti(burst)
+        clearTimeout(confettiTimer.current)
+        confettiTimer.current = setTimeout(() => setConfetti(null), 720)
+      }
       closeCard()
+      setRailExpandedId(null)
       onAccept(s)
     },
-    [editor, onAccept, closeCard],
+    [editor, onAccept, closeCard, hover],
   )
 
   const handleDismiss = useCallback(
@@ -439,7 +558,16 @@ export function EditorCore({
       const suggestionEl = (e.target as HTMLElement).closest('.petal-suggestion') as HTMLElement | null
       if (suggestionEl) {
         const id = suggestionEl.getAttribute('data-suggestion-id')
-        if (id) openCardFor(id, suggestionEl)
+        if (id) {
+          // With the rail open, a tap emphasizes and expands its margin card
+          // instead of opening a floating one.
+          if (railEnabled) {
+            setActiveId(id)
+            setRailExpandedId(id)
+          } else {
+            openCardFor(id, suggestionEl)
+          }
+        }
         return
       }
       if (!editor || !spellChecker) return
@@ -461,7 +589,7 @@ export function EditorCore({
       setWordInfo(null)
       setMisspell({ ...range, suggestions: spellChecker.suggest(range.word), top, left })
     },
-    [editor, spellChecker, closeCard, openCardFor],
+    [editor, spellChecker, closeCard, openCardFor, railEnabled],
   )
 
   const replaceMisspelling = useCallback(
@@ -854,7 +982,7 @@ export function EditorCore({
             onAdd={addMisspellingToDict}
           />
         )}
-        {hover && (
+        {hover && !railEnabled && (
           <SuggestionCard
             suggestion={hover.suggestion}
             style={{ top: hover.top, left: hover.left }}
@@ -865,6 +993,18 @@ export function EditorCore({
             onExpandChange={setPinned}
           />
         )}
+        {railEnabled && railItems.length > 0 && (
+          <SuggestionRail
+            items={railItems}
+            activeId={activeId}
+            expandedId={railExpandedId}
+            onAccept={handleAccept}
+            onDismiss={handleDismiss}
+            onHover={setActiveId}
+            onActivate={activateRailCard}
+            onToggleExpand={toggleRailExpand}
+          />
+        )}
       </div>
     </div>
   )

web/src/components/Editor/SuggestionCard.tsx

@@ -1,15 +1,7 @@
 import { useState } from 'react'
-import type { Suggestion, SuggestionType } from '../../api/client'
+import type { Suggestion } from '../../api/client'
 import { AskPetal } from './AskPetal'
-
-// Per-type accent color + human label, mirroring the design tokens.
-const TYPE_META: Record<SuggestionType, { color: string; label: string }> = {
-  grammar: { color: 'var(--color-mint)', label: 'Grammar' },
-  phrasing: { color: 'var(--color-peach)', label: 'Phrasing' },
-  idiom: { color: 'var(--color-lavender)', label: 'Idiom' },
-  clarity: { color: 'var(--color-sky)', label: 'Clarity' },
-  voice: { color: 'var(--color-honey)', label: 'Voice' },
-}
+import { TYPE_META } from './suggestionMeta'
 
 interface Props {
   suggestion: Suggestion

web/src/components/Editor/SuggestionHighlight.ts

@@ -16,9 +16,17 @@ export const suggestionPluginKey = new PluginKey<PluginState>('petalSuggestions'
 
 interface PluginState {
   suggestions: Suggestion[]
+  // The suggestion currently emphasized from its margin card / a hover, or null.
+  // Carried in plugin state (not an imperative DOM class) so it survives the
+  // decoration repaints that fire on every document change.
+  activeId: string | null
   decorations: DecorationSet
 }
 
+// Meta carried on a transaction to update the plugin: either a fresh suggestion
+// list or a change to which suggestion is emphasized.
+type SuggestionMeta = { suggestions: Suggestion[] } | { activeId: string | null }
+
 // mapOffset converts a character offset within a textblock's flattened text into
 // an absolute ProseMirror position, accounting for inline atoms (e.g. hard
 // breaks) that occupy a position but contribute no text.
@@ -130,14 +138,15 @@ export function findRange(doc: PMNode, search: string): { from: number; to: numb
   return result
 }
 
-function buildDecorations(doc: PMNode, suggestions: Suggestion[]): DecorationSet {
+function buildDecorations(doc: PMNode, suggestions: Suggestion[], activeId: string | null): DecorationSet {
   const decos: Decoration[] = []
   for (const s of suggestions) {
     const range = findRange(doc, s.original)
     if (!range) continue
+    const active = s.id === activeId ? ' petal-suggestion-active' : ''
     decos.push(
       Decoration.inline(range.from, range.to, {
-        class: `petal-suggestion petal-suggestion-${s.type}`,
+        class: `petal-suggestion petal-suggestion-${s.type}${active}`,
         'data-suggestion-id': s.id,
       }),
     )
@@ -148,10 +157,22 @@ function buildDecorations(doc: PMNode, suggestions: Suggestion[]): DecorationSet
 // setSuggestions pushes a new suggestion list into the plugin. Decorations are
 // rebuilt against the current document immediately.
 export function setSuggestions(state: EditorState, dispatch: (tr: Transaction) => void, suggestions: Suggestion[]) {
-  const tr = state.tr.setMeta(suggestionPluginKey, suggestions)
+  const tr = state.tr.setMeta(suggestionPluginKey, { suggestions } satisfies SuggestionMeta)
   dispatch(tr)
 }
 
+// setActiveSuggestion marks one suggestion (or none) as emphasized, rebuilding
+// the decorations so the flagged text gets the active wash. Driven by the margin
+// rail's hover/selection so the card↔text link reads both ways.
+export function setActiveSuggestion(
+  state: EditorState,
+  dispatch: (tr: Transaction) => void,
+  activeId: string | null,
+) {
+  if (suggestionPluginKey.getState(state)?.activeId === activeId) return
+  dispatch(state.tr.setMeta(suggestionPluginKey, { activeId } satisfies SuggestionMeta))
+}
+
 export const SuggestionHighlight = Extension.create({
   name: 'suggestionHighlight',
 
@@ -160,17 +181,29 @@ export const SuggestionHighlight = Extension.create({
       new Plugin<PluginState>({
         key: suggestionPluginKey,
         state: {
-          init: () => ({ suggestions: [], decorations: DecorationSet.empty }),
+          init: () => ({ suggestions: [], activeId: null, decorations: DecorationSet.empty }),
           apply(tr, value, _oldState, newState) {
-            const meta = tr.getMeta(suggestionPluginKey) as Suggestion[] | undefined
-            if (meta) {
-              return { suggestions: meta, decorations: buildDecorations(newState.doc, meta) }
+            const meta = tr.getMeta(suggestionPluginKey) as SuggestionMeta | undefined
+            if (meta && 'suggestions' in meta) {
+              return {
+                suggestions: meta.suggestions,
+                activeId: value.activeId,
+                decorations: buildDecorations(newState.doc, meta.suggestions, value.activeId),
+              }
+            }
+            if (meta && 'activeId' in meta) {
+              return {
+                suggestions: value.suggestions,
+                activeId: meta.activeId,
+                decorations: buildDecorations(newState.doc, value.suggestions, meta.activeId),
+              }
             }
             // On any document change, re-anchor by string against the new doc.
             if (tr.docChanged) {
               return {
                 suggestions: value.suggestions,
-                decorations: buildDecorations(newState.doc, value.suggestions),
+                activeId: value.activeId,
+                decorations: buildDecorations(newState.doc, value.suggestions, value.activeId),
               }
             }
             return value

web/src/components/Editor/SuggestionRail.tsx

@@ -0,0 +1,211 @@
+import { forwardRef, useLayoutEffect, useRef, useState } from 'react'
+import type { Suggestion } from '../../api/client'
+import { AskPetal } from './AskPetal'
+import { TYPE_META } from './suggestionMeta'
+
+// Vertical breathing room kept between stacked cards when their natural anchors
+// would otherwise collide.
+const CARD_GAP = 12
+
+export interface RailItem {
+  suggestion: Suggestion
+  // Vertical offset (px) of the suggestion's highlight, relative to the editor
+  // wrapper — the position the card wants to sit beside.
+  anchorTop: number
+}
+
+interface Props {
+  items: RailItem[]
+  // The suggestion currently emphasized (its highlight hovered, or this card
+  // hovered) — gets a lifted, ring-accented treatment so the text↔card link reads.
+  activeId: string | null
+  // The card expanded to show the full explanation + Ask Petal, or null.
+  expandedId: string | null
+  onAccept: (s: Suggestion) => void
+  onDismiss: (s: Suggestion) => void
+  // Pointer entering/leaving a card, so the matching highlight can light up.
+  onHover: (id: string | null) => void
+  // A card's body was clicked — scroll its highlight into view and toggle expand.
+  onActivate: (id: string) => void
+  onToggleExpand: (id: string) => void
+}
+
+// SuggestionRail is the right-margin "comment column": every outstanding
+// suggestion as a card, vertically aligned to the text it flags, so the writer
+// sees the whole queue at once instead of hunting highlight by highlight. Cards
+// are anchored to their highlight's vertical position and pushed down only as far
+// as needed to avoid overlapping their neighbor above (a la doc-editor comments).
+export function SuggestionRail({
+  items,
+  activeId,
+  expandedId,
+  onAccept,
+  onDismiss,
+  onHover,
+  onActivate,
+  onToggleExpand,
+}: Props) {
+  // Measured resolved tops keyed by suggestion id (after collision avoidance).
+  const [tops, setTops] = useState<Record<string, number>>({})
+  const cardRefs = useRef<Map<string, HTMLDivElement>>(new Map())
+  // Bumped whenever a card's own height changes (Ask Petal streaming in, wrapping
+  // text) so the stack re-packs and cards below don't get overlapped.
+  const [measureTick, setMeasureTick] = useState(0)
+  const observerRef = useRef<ResizeObserver | null>(null)
+  if (!observerRef.current && typeof ResizeObserver !== 'undefined') {
+    observerRef.current = new ResizeObserver(() => setMeasureTick((t) => t + 1))
+  }
+  useLayoutEffect(() => () => observerRef.current?.disconnect(), [])
+
+  // Sort by anchor, then greedily stack: each card sits at its anchor unless that
+  // would overlap the previous card, in which case it slides down just enough.
+  // Re-runs whenever the anchors or the expanded card (which changes a height)
+  // shift. Reads live offsetHeight so variable-length explanations pack tightly.
+  const ordered = [...items].sort((a, b) => a.anchorTop - b.anchorTop)
+  const layoutKey = ordered.map((i) => `${i.suggestion.id}:${Math.round(i.anchorTop)}`).join('|')
+  useLayoutEffect(() => {
+    let cursor = -Infinity
+    const next: Record<string, number> = {}
+    for (const { suggestion, anchorTop } of ordered) {
+      const h = cardRefs.current.get(suggestion.id)?.offsetHeight ?? 96
+      const top = Math.max(anchorTop, cursor)
+      next[suggestion.id] = top
+      cursor = top + h + CARD_GAP
+    }
+    setTops((prev) => {
+      const ids = Object.keys(next)
+      if (ids.length === Object.keys(prev).length && ids.every((id) => prev[id] === next[id])) return prev
+      return next
+    })
+    // layoutKey captures anchor positions; expandedId/measureTick capture height
+    // changes (toggling detail, Ask Petal streaming in).
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [layoutKey, expandedId, measureTick])
+
+  return (
+    <div className="petal-rail petal-no-print" aria-label="Suggestions">
+      {ordered.map(({ suggestion }) => (
+        <RailCard
+          key={suggestion.id}
+          ref={(el) => {
+            const prev = cardRefs.current.get(suggestion.id)
+            if (prev && prev !== el) observerRef.current?.unobserve(prev)
+            if (el) {
+              cardRefs.current.set(suggestion.id, el)
+              observerRef.current?.observe(el)
+            } else {
+              cardRefs.current.delete(suggestion.id)
+            }
+          }}
+          suggestion={suggestion}
+          top={tops[suggestion.id] ?? 0}
+          active={activeId === suggestion.id}
+          expanded={expandedId === suggestion.id}
+          onAccept={onAccept}
+          onDismiss={onDismiss}
+          onHover={onHover}
+          onActivate={onActivate}
+          onToggleExpand={onToggleExpand}
+        />
+      ))}
+    </div>
+  )
+}
+
+interface CardProps {
+  suggestion: Suggestion
+  top: number
+  active: boolean
+  expanded: boolean
+  onAccept: (s: Suggestion) => void
+  onDismiss: (s: Suggestion) => void
+  onHover: (id: string | null) => void
+  onActivate: (id: string) => void
+  onToggleExpand: (id: string) => void
+}
+
+const RailCard = forwardRef<HTMLDivElement, CardProps>(function RailCard(
+  { suggestion, top, active, expanded, onAccept, onDismiss, onHover, onActivate, onToggleExpand },
+  ref,
+) {
+  const meta = TYPE_META[suggestion.type]
+  const hasReplacement = suggestion.replacement.trim() !== ''
+
+  return (
+    <div
+      ref={ref}
+      role="group"
+      aria-label={`${meta.label} suggestion`}
+      onMouseEnter={() => onHover(suggestion.id)}
+      onMouseLeave={() => onHover(null)}
+      className={`petal-rail-card${active ? ' petal-rail-card-active' : ''}`}
+      style={{ top, borderLeftColor: meta.color }}
+    >
+      <div className="flex items-center justify-between">
+        <span
+          className="inline-flex items-center rounded-full px-2 py-0.5 text-[0.68rem] font-bold"
+          style={{ background: meta.color, color: 'var(--color-plum)' }}
+        >
+          {meta.label}
+        </span>
+        <button
+          type="button"
+          aria-label="Dismiss suggestion"
+          onClick={() => onDismiss(suggestion)}
+          className="petal-rail-x"
+        >
+          ✕
+        </button>
+      </div>
+
+      {/* The body toggles the expanded explanation + Ask Petal and points the
+          editor at the flagged text. */}
+      <button type="button" onClick={() => onActivate(suggestion.id)} className="mt-2 block w-full text-left">
+        {hasReplacement && (
+          <span className="flex flex-col gap-0.5" style={{ fontFamily: 'var(--font-body)' }}>
+            <span className="truncate text-[0.9rem] line-through" style={{ color: 'var(--color-muted)' }}>
+              {suggestion.original}
+            </span>
+            <span className="truncate text-[0.9rem] font-medium" style={{ color: 'var(--color-plum)' }}>
+              {suggestion.replacement}
+            </span>
+          </span>
+        )}
+        <span
+          className={`mt-1.5 block text-[0.82rem] leading-snug${expanded ? '' : ' line-clamp-2'}`}
+          style={{ color: 'var(--color-plum)' }}
+        >
+          {suggestion.explanation}
+        </span>
+      </button>
+
+      {expanded && <AskPetal suggestionId={suggestion.id} explanation={suggestion.explanation} />}
+
+      <div className="mt-2.5 flex items-center gap-2">
+        {hasReplacement && (
+          <button
+            type="button"
+            onClick={() => onAccept(suggestion)}
+            className="rounded-full px-3 py-1 text-xs font-bold text-white"
+            style={{ background: 'var(--color-accent)' }}
+            onMouseEnter={(e) => (e.currentTarget.style.background = 'var(--color-accent-hover)')}
+            onMouseLeave={(e) => (e.currentTarget.style.background = 'var(--color-accent)')}
+          >
+            Accept
+          </button>
+        )}
+        <button
+          type="button"
+          onClick={() => onToggleExpand(suggestion.id)}
+          className="rounded-full px-2.5 py-1 text-xs font-bold transition-colors"
+          style={{
+            background: expanded ? 'var(--color-surface-alt)' : 'transparent',
+            color: 'var(--color-accent-hover)',
+          }}
+        >
+          {expanded ? 'Hide Petal' : 'Ask Petal ✨'}
+        </button>
+      </div>
+    </div>
+  )
+})

web/src/components/Editor/suggestionMeta.ts

@@ -0,0 +1,12 @@
+import type { SuggestionType } from '../../api/client'
+
+// Per-type accent color + human label, mirroring the design tokens. Shared by the
+// inline hover SuggestionCard and the margin SuggestionRail so a given suggestion
+// type reads the same color/name wherever it surfaces.
+export const TYPE_META: Record<SuggestionType, { color: string; label: string }> = {
+  grammar: { color: 'var(--color-mint)', label: 'Grammar' },
+  phrasing: { color: 'var(--color-peach)', label: 'Phrasing' },
+  idiom: { color: 'var(--color-lavender)', label: 'Idiom' },
+  clarity: { color: 'var(--color-sky)', label: 'Clarity' },
+  voice: { color: 'var(--color-honey)', label: 'Voice' },
+}

web/src/effects/PetalFall.tsx

@@ -17,6 +17,23 @@ const PALETTE: { r: number; g: number; b: number }[] = [
   { r: 255, g: 122, b: 172 }, // deep rose
 ]
 
+// Night-mode sparkle palette: soft starlight tones for the small twinkle accents
+// scattered between the chunky cartoon stars (moonlit white, warm gold).
+const STAR_PALETTE: { r: number; g: number; b: number }[] = [
+  { r: 255, g: 255, b: 255 }, // white
+  { r: 255, g: 240, b: 200 }, // warm gold
+]
+
+// Chunky cartoon-star colors — bold, candy-bright fills à la a Mario power star
+// or a Kirby warp star. Each is drawn with a glossy gradient + a puffy outline.
+const CARTOON_COLORS: { r: number; g: number; b: number }[] = [
+  { r: 255, g: 201, b: 56 }, // golden power-star
+  { r: 255, g: 150, b: 190 }, // bubblegum pink
+  { r: 130, g: 195, b: 255 }, // sky blue
+  { r: 150, g: 225, b: 190 }, // mint
+  { r: 200, g: 175, b: 255 }, // lavender
+]
+
 const SPRITE_PX = 160 // off-screen render resolution per petal sprite
 
 function rgba(c: { r: number; g: number; b: number }, a: number): string {
@@ -67,6 +84,118 @@ function makeSprite(color: { r: number; g: number; b: number }): HTMLCanvasEleme
   return cv
 }
 
+// Draw one star sprite: a soft radial glow with a crisp four-point sparkle on
+// top, filled brightest at the core so it reads as a twinkling star.
+function makeStarSprite(color: { r: number; g: number; b: number }): HTMLCanvasElement {
+  const s = SPRITE_PX
+  const cv = document.createElement('canvas')
+  cv.width = s
+  cv.height = s
+  const g = cv.getContext('2d')!
+  g.translate(s / 2, s / 2)
+
+  // Soft halo behind the sparkle.
+  const glow = g.createRadialGradient(0, 0, 0, 0, 0, s * 0.5)
+  glow.addColorStop(0, rgba(mixWhite(color, 0.5), 0.5))
+  glow.addColorStop(0.35, rgba(color, 0.16))
+  glow.addColorStop(1, rgba(color, 0))
+  g.fillStyle = glow
+  g.beginPath()
+  g.arc(0, 0, s * 0.5, 0, Math.PI * 2)
+  g.fill()
+
+  // Four-point sparkle: alternate a long outer radius with a small inner one so
+  // the points pinch in sharply.
+  const R = s * 0.46
+  const r = s * 0.08
+  g.beginPath()
+  for (let i = 0; i < 8; i++) {
+    const ang = (i * Math.PI) / 4 - Math.PI / 2
+    const rad = i % 2 === 0 ? R : r
+    const x = Math.cos(ang) * rad
+    const y = Math.sin(ang) * rad
+    if (i === 0) g.moveTo(x, y)
+    else g.lineTo(x, y)
+  }
+  g.closePath()
+  const core = g.createRadialGradient(0, 0, 0, 0, 0, R)
+  core.addColorStop(0, rgba(mixWhite(color, 0.7), 0.98))
+  core.addColorStop(0.5, rgba(color, 0.9))
+  core.addColorStop(1, rgba(color, 0.2))
+  g.fillStyle = core
+  g.fill()
+
+  return cv
+}
+
+// Draw one chunky cartoon star — a fat five-point star with a glossy radial
+// fill, a soft outer glow so it pops off the dark sky, a puffy rounded outline
+// (round line joins → Kirby-ish soft points), and a little corner shine. Reads
+// like a game power-up rather than a delicate sparkle.
+function makeCartoonStar(base: { r: number; g: number; b: number }): HTMLCanvasElement {
+  const s = SPRITE_PX
+  const cv = document.createElement('canvas')
+  cv.width = s
+  cv.height = s
+  const g = cv.getContext('2d')!
+  g.translate(s / 2, s / 2)
+
+  const R = s * 0.4 // outer point radius
+  const r = R * 0.46 // inner radius — small ⇒ chunky, defined points
+
+  // Soft outer glow so the star separates from the night background.
+  const glow = g.createRadialGradient(0, 0, R * 0.3, 0, 0, R * 1.35)
+  glow.addColorStop(0, rgba(base, 0.35))
+  glow.addColorStop(1, rgba(base, 0))
+  g.fillStyle = glow
+  g.beginPath()
+  g.arc(0, 0, R * 1.35, 0, Math.PI * 2)
+  g.fill()
+
+  // Five-point star path (start at the top point).
+  const star = () => {
+    g.beginPath()
+    for (let i = 0; i < 10; i++) {
+      const ang = -Math.PI / 2 + (i * Math.PI) / 5
+      const rad = i % 2 === 0 ? R : r
+      const x = Math.cos(ang) * rad
+      const y = Math.sin(ang) * rad
+      if (i === 0) g.moveTo(x, y)
+      else g.lineTo(x, y)
+    }
+    g.closePath()
+  }
+
+  // Candy fill: bright near the top-left, deepening toward the lower edge.
+  const fill = g.createRadialGradient(-R * 0.25, -R * 0.3, R * 0.1, 0, 0, R * 1.1)
+  fill.addColorStop(0, rgba(mixWhite(base, 0.7), 1))
+  fill.addColorStop(0.5, rgba(base, 1))
+  fill.addColorStop(1, rgba(scale(base, 0.82), 1))
+  star()
+  g.fillStyle = fill
+  g.fill()
+
+  // Puffy outline in a deeper shade of the same hue (round joins soften the
+  // points so it feels hand-drawn, not spiky).
+  g.lineJoin = 'round'
+  g.lineCap = 'round'
+  g.lineWidth = s * 0.05
+  g.strokeStyle = rgba(scale(base, 0.5), 0.95)
+  star()
+  g.stroke()
+
+  // Glossy shine: a small soft white blob up in the top-left lobe.
+  const shine = g.createRadialGradient(-R * 0.22, -R * 0.32, 0, -R * 0.22, -R * 0.32, R * 0.3)
+  shine.addColorStop(0, 'rgba(255,255,255,0.85)')
+  shine.addColorStop(1, 'rgba(255,255,255,0)')
+  g.fillStyle = shine
+  g.beginPath()
+  g.ellipse(-R * 0.22, -R * 0.32, R * 0.26, R * 0.18, -0.5, 0, Math.PI * 2)
+  g.fill()
+
+  return cv
+}
+
 interface Petal {
   baseX: number
   y: number
@@ -81,7 +210,7 @@ interface Petal {
   sprite: number
 }
 
-export function PetalFall() {
+export function PetalFall({ night = false }: { night?: boolean }) {
   const canvasRef = useRef<HTMLCanvasElement>(null)
 
   useEffect(() => {
@@ -98,7 +227,12 @@ export function PetalFall() {
     const canvas = el
     const ctx = c2d
 
-    const sprites = PALETTE.map(makeSprite)
+    // Night sky = chunky cartoon stars (the stars of the show) with a couple of
+    // small twinkle sparkles mixed in for depth. Random sprite pick weights it
+    // ~70% chunky stars.
+    const sprites = night
+      ? [...CARTOON_COLORS.map(makeCartoonStar), ...STAR_PALETTE.map(makeStarSprite)]
+      : PALETTE.map(makeSprite)
     const dpr = Math.min(window.devicePixelRatio || 1, 2)
     let W = window.innerWidth
     let H = window.innerHeight
@@ -107,19 +241,39 @@ export function PetalFall() {
     const count = Math.round(Math.min(46, Math.max(16, (W * H) / 46000)))
     const rnd = (a: number, b: number) => a + Math.random() * (b - a)
 
-    const spawn = (initial: boolean): Petal => ({
-      baseX: rnd(0, W),
-      y: initial ? rnd(-H, H) : rnd(-60, -20),
-      size: rnd(16, 38),
-      vy: rnd(22, 52),
-      angle: rnd(0, Math.PI * 2),
-      spin: rnd(-0.9, 0.9),
-      swayAmp: rnd(14, 42),
-      swayFreq: rnd(0.3, 0.85),
-      swayPhase: rnd(0, Math.PI * 2),
-      alpha: rnd(0.5, 0.82),
-      sprite: Math.floor(Math.random() * sprites.length),
-    })
+    // Stars are smaller, drift down a touch slower, sway less, and barely spin
+    // (they twinkle in place instead of tumbling like petals).
+    const spawn = (initial: boolean): Petal =>
+      night
+        ? {
+            baseX: rnd(0, W),
+            y: initial ? rnd(-H, H) : rnd(-60, -20),
+            size: rnd(13, 32),
+            vy: rnd(14, 36),
+            angle: rnd(0, Math.PI * 2),
+            // Every star visibly spins — random direction, and a guaranteed
+            // minimum speed (a 5-point star is symmetric every 72°, so a slow
+            // rate reads as motionless). Varied so no two turn quite alike.
+            spin: (Math.random() < 0.5 ? -1 : 1) * rnd(0.5, 2.2),
+            swayAmp: 0, // stars fall straight down — no petal-like drift
+            swayFreq: rnd(0.4, 1.1), // still drives the twinkle shimmer below
+            swayPhase: rnd(0, Math.PI * 2),
+            alpha: rnd(0.55, 0.9),
+            sprite: Math.floor(Math.random() * sprites.length),
+          }
+        : {
+            baseX: rnd(0, W),
+            y: initial ? rnd(-H, H) : rnd(-60, -20),
+            size: rnd(16, 38),
+            vy: rnd(22, 52),
+            angle: rnd(0, Math.PI * 2),
+            spin: rnd(-0.9, 0.9),
+            swayAmp: rnd(14, 42),
+            swayFreq: rnd(0.3, 0.85),
+            swayPhase: rnd(0, Math.PI * 2),
+            alpha: rnd(0.5, 0.82),
+            sprite: Math.floor(Math.random() * sprites.length),
+          }
 
     let petals = Array.from({ length: count }, () => spawn(true))
 
@@ -157,10 +311,16 @@ export function PetalFall() {
           continue
         }
 
+        // Stars shimmer gently (shallow pulse so the chunky ones glow rather
+        // than blink); petals hold a steady alpha.
+        const alpha = night
+          ? p.alpha * (0.8 + 0.2 * Math.sin(t * p.swayFreq * 2.2 + p.swayPhase))
+          : p.alpha
+
         ctx.save()
         ctx.translate(x, p.y)
         ctx.rotate(p.angle)
-        ctx.globalAlpha = p.alpha
+        ctx.globalAlpha = alpha
         ctx.drawImage(sprites[p.sprite], -p.size / 2, -p.size / 2, p.size, p.size)
         ctx.restore()
       }
@@ -187,7 +347,7 @@ export function PetalFall() {
       document.removeEventListener('visibilitychange', onVisibility)
       petals = []
     }
-  }, [])
+  }, [night])
 
   return (
     <canvas

web/src/hooks/useCheckpoint.ts

@@ -20,17 +20,29 @@ export function useCheckpoint(docId: string | null) {
   const [llmDown, setLlmDown] = useState(false)
 
   const debounceRef = useRef<ReturnType<typeof setTimeout>>(undefined)
+  // Pending auto-retry timer for a failed checkpoint (see runCheck).
+  const retryRef = useRef<ReturnType<typeof setTimeout>>(undefined)
   const docIdRef = useRef(docId)
   docIdRef.current = docId
 
   // Token to discard responses from a doc we've since navigated away from.
   const runRef = useRef(0)
 
-  const runCheck = useCallback(async () => {
+  // Backoff schedule for re-running a checkpoint that failed. A paste fires
+  // exactly ONE checkpoint, and nothing re-fires until the next keystroke — so a
+  // single transient failure (a momentary stall on the shared inference box)
+  // would otherwise strand the user on "resting" indefinitely after a paste.
+  // These delays clear the server's 30s per-doc floor by the later attempts, and
+  // a failed pass now releases its slot server-side so a retry can truly re-run.
+  const RETRY_DELAYS_MS = [3000, 12000, 35000]
+
+  const runCheck = useCallback(async (attempt = 0) => {
     const id = docIdRef.current
     if (!id) return
+    clearTimeout(retryRef.current)
     const run = ++runRef.current
     setChecking(true)
+    let retrying = false
     try {
       const fresh = await api.checkDoc(id)
       if (run === runRef.current && id === docIdRef.current) {
@@ -39,9 +51,18 @@ export function useCheckpoint(docId: string | null) {
       }
     } catch (err) {
       console.error('checkpoint failed', err)
-      if (run === runRef.current) setLlmDown(true)
+      if (run !== runRef.current) return
+      if (attempt < RETRY_DELAYS_MS.length) {
+        // Keep trying quietly — don't flag "resting" until retries are exhausted.
+        retrying = true
+        retryRef.current = setTimeout(() => void runCheck(attempt + 1), RETRY_DELAYS_MS[attempt])
+      } else {
+        setLlmDown(true)
+      }
     } finally {
-      if (run === runRef.current) setChecking(false)
+      // Stay in the "checking" state while a retry is queued so the breathing dot
+      // keeps reassuring rather than flickering off between attempts.
+      if (run === runRef.current && !retrying) setChecking(false)
     }
   }, [])
 
@@ -51,6 +72,7 @@ export function useCheckpoint(docId: string | null) {
   const runVoice = useCallback(async () => {
     const id = docIdRef.current
     if (!id) return
+    clearTimeout(retryRef.current) // a voice pass supersedes a queued grammar retry
     const run = ++runRef.current
     setVoicing(true)
     try {
@@ -70,13 +92,15 @@ export function useCheckpoint(docId: string | null) {
   // Call on every edit; schedules a check 4s after typing settles.
   const schedule = useCallback(() => {
     clearTimeout(debounceRef.current)
-    debounceRef.current = setTimeout(runCheck, DEBOUNCE_MS)
+    clearTimeout(retryRef.current) // a fresh edit supersedes any queued retry
+    debounceRef.current = setTimeout(() => void runCheck(), DEBOUNCE_MS)
   }, [runCheck])
 
   // Load existing pending suggestions whenever the document changes, and cancel
   // any in-flight debounce from the previous doc.
   useEffect(() => {
     clearTimeout(debounceRef.current)
+    clearTimeout(retryRef.current)
     runRef.current++
     setSuggestions([])
     setChecking(false)
@@ -97,7 +121,13 @@ export function useCheckpoint(docId: string | null) {
     }
   }, [docId])
 
-  useEffect(() => () => clearTimeout(debounceRef.current), [])
+  useEffect(
+    () => () => {
+      clearTimeout(debounceRef.current)
+      clearTimeout(retryRef.current)
+    },
+    [],
+  )
 
   // Drop one suggestion locally (after accept/dismiss) without a refetch.
   const removeSuggestion = useCallback((id: string) => {

web/src/hooks/useNightMode.ts

@@ -0,0 +1,25 @@
+import { useEffect, useState } from 'react'
+import { isBedtime } from '../lib/night'
+
+// Tracks whether it's late enough to shift Petal into its calm night mode (dark
+// theme + falling stars instead of petals). Re-checks once a minute so the swap
+// happens on its own as the clock rolls past 11pm — no reload needed. Returns
+// the boolean and also toggles a `petal-night` class on <html> so the CSS token
+// overrides cascade to everything (body background, panels, portals included).
+export function useNightMode(): boolean {
+  const [night, setNight] = useState(() => isBedtime())
+
+  useEffect(() => {
+    const tick = () => setNight(isBedtime())
+    tick()
+    const id = setInterval(tick, 60_000)
+    return () => clearInterval(id)
+  }, [])
+
+  useEffect(() => {
+    document.documentElement.classList.toggle('petal-night', night)
+    return () => document.documentElement.classList.remove('petal-night')
+  }, [night])
+
+  return night
+}

web/src/index.css

@@ -37,6 +37,23 @@
   --shadow-soft: 0 4px 20px rgba(180, 130, 160, 0.12);
 }
 
+/* Night mode — toggled by adding `petal-night` to <html> (useNightMode) once the
+   clock passes ~11pm. We only re-point the palette tokens; every Tailwind color
+   utility reads them via var(), so the whole UI shifts to a calm, dim, moonlit
+   theme without touching component markup. Pastel accent/type colors are kept —
+   they read beautifully on the dark plum ground. Falling petals become stars. */
+html.petal-night {
+  --color-bg: #14111E;          /* deep night plum-indigo */
+  --color-surface: #211C30;     /* raised panels */
+  --color-surface-alt: #2A2440; /* alt surface / hover wash */
+  --color-border: #38304E;      /* dim lavender border */
+  --color-plum: #ECE3F2;        /* ink → soft moonlit lavender-white */
+  --color-muted: #9F93B8;       /* muted lavender-grey, lifted for contrast */
+  --color-accent: #E8A0BF;      /* rose still primary — pops on dark */
+  --color-accent-hover: #F2B7D2;/* lighter on hover against the dark */
+  --shadow-soft: 0 6px 28px rgba(0, 0, 0, 0.45);
+}
+
 html, body, #root {
   height: 100%;
 }
@@ -47,6 +64,8 @@ body {
   color: var(--color-plum);
   font-family: var(--font-ui);
   -webkit-font-smoothing: antialiased;
+  /* Gentle dusk/dawn fade when night mode flips. */
+  transition: background 600ms ease, color 600ms ease;
 }
 
 /* Gentle, consistent motion across interactive elements */
@@ -223,6 +242,67 @@ button, a, input {
   animation: petal-suggestion-in 200ms ease both;
 }
 
+/* Text emphasized from its margin card (or a hover) — a soft wash so the
+   card↔text link reads at a glance, both directions. */
+.petal-suggestion-active {
+  background: var(--color-surface-alt);
+  box-shadow: 0 0 0 3px var(--color-surface-alt);
+  border-radius: 3px;
+}
+
+/* --- Suggestion margin rail -------------------------------------------------
+   The right-hand "comment column": every outstanding suggestion as a card,
+   vertically aligned to the text it flags, so the whole queue is visible at a
+   glance instead of one-highlight-at-a-time on hover. It floats in the whitespace
+   just right of the editor column and is only mounted when there's room for it
+   (EditorCore measures the gap). Cards are absolutely positioned by a resolved
+   top (anchor + collision avoidance), so the container only anchors the column. */
+.petal-rail {
+  position: absolute;
+  top: 0;
+  left: 100%;
+  margin-left: 32px;
+  width: 300px;
+  /* Below the companion mascot (z-40): where a stacked card reaches the
+     bottom-right corner it simply tucks behind the sleeping cat. */
+  z-index: 10;
+}
+.petal-rail-card {
+  position: absolute;
+  left: 0;
+  right: 0;
+  background: var(--color-surface);
+  border: 1px solid var(--color-border);
+  border-left: 3px solid var(--color-border); /* type color set inline */
+  border-radius: var(--radius-card);
+  box-shadow: var(--shadow-soft);
+  padding: 0.7rem 0.85rem;
+  /* `top` eases so re-stacking (accept/dismiss/edit) glides instead of jumping.
+     The entrance uses fill `backwards` so its translateY doesn't linger and fight
+     the active-state transform once it's done. */
+  transition: top 240ms cubic-bezier(0.2, 0.7, 0.3, 1), box-shadow 200ms ease,
+    transform 200ms ease, border-color 200ms ease;
+  animation: petal-suggestion-in 220ms ease backwards;
+}
+.petal-rail-card-active {
+  transform: translateX(-5px);
+  box-shadow: 0 8px 26px rgba(180, 130, 160, 0.24);
+  border-top-color: var(--color-accent);
+  border-right-color: var(--color-accent);
+  border-bottom-color: var(--color-accent);
+}
+.petal-rail-x {
+  color: var(--color-muted);
+  font-size: 0.72rem;
+  line-height: 1;
+  padding: 2px 5px;
+  border-radius: var(--radius-pill);
+}
+.petal-rail-x:hover {
+  background: var(--color-surface-alt);
+  color: var(--color-plum);
+}
+
 /* --- Find & Replace ---------------------------------------------------------
    In-document search (Ctrl/Cmd+F). Every match gets a soft honey wash; the
    current match is brighter with a rose ring so it stands out as you step

web/src/lib/night.ts

@@ -0,0 +1,11 @@
+// Shared "is it late?" definition, used by both the companion's bedtime nag and
+// the night-mode theme/starfall switch so they always agree. Local clock — the
+// writer's own machine (millenia's timezone in deployment).
+
+export const BEDTIME_FROM = 23 // local hour the night window opens (11pm)…
+export const BEDTIME_TO = 4 //    …and closes (4am); past this we assume an early start
+
+export function isBedtime(d: Date = new Date()): boolean {
+  const h = d.getHours()
+  return h >= BEDTIME_FROM || h < BEDTIME_TO
+}