Phase 12 + 13: collocation coach + vocabulary garden

Phase 12 — collocation coach: a third suggestion family for gentle
"natives usually say…" hints on non-native word pairings, reusing the
existing runPass/pendingScope/rail machinery.
- llm/collocation.go (RunCollocation, 25s floor, reuses ParseCheckpoint)
  + collocationSystemPrompt/CollocationMessages (warm, Mandarin gloss,
  defers grammar/spelling to the grammar family)
- migration 0005 rebuilds the suggestions table to extend the type CHECK
  (SQLite can't ALTER a CHECK)
- collocationScope + CollocationLimit + POST /{id}/collocation
- fix: grammarScope was `type != 'voice'` and would wipe the new
  collocation flags; now `type NOT IN ('voice','collocation')`
- frontend: --color-blossom, "Make it sound natural 🌸" pill,
  collocating/runCollocation in useCheckpoint, StatusBar dot

Phase 13 — vocabulary garden: capture looked-up words and surface them
for gentle spaced repetition.
- new internal/vocab package: migration 0006 (vocab_words, SM-2-lite
  columns, doc_id ON DELETE SET NULL, UNIQUE(user_id,word)),
  scheduler.go (Leitner ladder 1/3/7/16/35 then geometric; gentle
  "again", no streak-shaming), handlers (capture-upsert/list/due/
  review/delete, owner-scoped, SQLite-side datetime math)
- auto-capture on word lookup (dictionary-known words only, captures
  the surrounding sentence + doc_id) + 🤍/💚 toggle on WordCard
- GardenPanel: blossom grid (bloom by reps), flashcard review (sentence
  blanked, flip, again/good/easy, recognition↔production), sleepy-kitten
  footer; opened from a global 🌷 header button

Tests: TestCollocationPassCoexists, vocab scheduler + handlers, db CHECK
extended. go build/vet/test + tsc + vite + vitest (51/51) clean;
migration verified against a copy of the live DB; live backend smoke
walked the full vocab lifecycle + the warm-502 collocation path.

Claude-Session: https://claude.ai/code/session_016Yr6jELuRc7hyzYLccQKZd
This commit is contained in:
prosolis
2026-06-26 15:50:25 -07:00
parent 20442d1356
commit 8aa437ec82
23 changed files with 1614 additions and 58 deletions

View File

@@ -0,0 +1,35 @@
package llm
import (
"context"
"time"
)
// CollocationInterval is the minimum gap between collocation passes for one
// document. Like the voice pass it runs on an explicit user action ("Make it
// sound natural") rather than a typing cadence, so this floor only guards the
// inference endpoint against the button being mashed.
const CollocationInterval = 25 * time.Second
// RunCollocation sends the WHOLE document for a collocation pass — gentle
// "natives usually say…" hints on non-native word pairings — and parses the JSON
// result with the checkpoint's tolerant parser. It is deliberately NOT
// TruncateDoc'd: like the voice pass it reads the whole document so the hints
// reflect the full piece. Each flag carries a native replacement to apply.
//
// The tone argument is accepted for a uniform pass signature and passed through
// to the prompt so a hint can prefer a register-appropriate pairing.
func RunCollocation(ctx context.Context, client LLMClient, contentText, tone string) ([]RawSuggestion, error) {
raw, err := client.Complete(ctx, CompletionRequest{
Messages: CollocationMessages(contentText, tone),
MaxTokens: 2048,
Temperature: 0.3,
RepetitionPenalty: 1.15,
TopP: 0.9,
Stop: []string{"```", "\n\n\n\n"},
})
if err != nil {
return nil, err
}
return ParseCheckpoint(raw)
}

View File

@@ -92,6 +92,54 @@ func VoiceMessages(contentText string) []Message {
}
}
// collocationSystemPrompt drives the collocation coach — the single most
// valuable polish for an ESL writer. It flags word PAIRINGS that are not wrong,
// just non-native ("do a decision" → "make a decision", "strong rain" → "heavy
// rain"), and explicitly DEFERS real grammar/spelling errors to the grammar
// checkpoint so the two families don't overlap. Every explanation is framed as a
// warm "natives usually say…" note with a short Mandarin gloss — never
// "error/wrong" — because these are stylistic, not mistakes. It is a distinct
// pass from the grammar checkpoint (do not bundle them). `replacement` carries
// the natural pairing the writer can accept in one tap.
const collocationSystemPrompt = `You are a warm, encouraging writing assistant helping someone who speaks English as a second language. ` +
`You are reviewing a COMPLETE document for COLLOCATIONS only — the natural word pairings native speakers use.
A collocation is a pair or short group of words that native speakers habitually say together. ESL writers ` +
`often choose words that are grammatically correct but sound non-native: "do a decision" instead of "make a decision", ` +
`"strong rain" instead of "heavy rain", "say a joke" instead of "tell a joke". These are NOT grammar mistakes — they ` +
`are just not what a native speaker would naturally say.
Identify up to 5 such non-native word pairings. For each, give the natural pairing a native speaker would use. ` +
`Be gentle and specific. Do NOT flag grammar errors, spelling mistakes, or unclear sentences — those are handled ` +
`elsewhere. Only flag word pairings that are correct but sound non-native.%s
Phrase every explanation warmly as "Natives usually say…" and include a brief Simplified Chinese gloss in parentheses. ` +
`Never use the words "error", "wrong", or "mistake" — these are friendly polish, not corrections.
Respond ONLY with valid JSON. No preamble, no markdown fences. Format:
{
"suggestions": [
{
"original": "exact word pairing from the document",
"replacement": "the natural native pairing",
"explanation": "friendly note, e.g. 'Natives usually say \"make a decision\" rather than \"do a decision\" (native usage / 地道说法).'",
"type": "collocation"
}
]
}
If every pairing already sounds natural, return: {"suggestions": []}`
// CollocationMessages builds the message array for a collocation pass over the
// WHOLE document (no truncation), gently steered toward the document's tone so a
// hint can prefer a register-appropriate pairing.
func CollocationMessages(contentText, tone string) []Message {
return []Message{
{Role: "system", Content: fmt.Sprintf(collocationSystemPrompt, toneGuidance(tone))},
{Role: "user", Content: contentText},
}
}
// askPetalSystemTemplate is the Ask Petal tutor prompt. The suggestion context
// is interpolated in; the user's own messages are appended after this system
// turn by the caller.