Let the garden keep what she was given, not only what she sought

Two halves of the same idea, both read out of work Petal already
records.

Planting: an accepted collocation is a learnable chunk, so it becomes a
phrase card. The scheduler didn't need to know — a three-word chunk
climbs the ladder exactly like a looked-up word. What needed care was
deciding what *isn't* a chunk (single words are word choice; a
six-word-plus "collocation" is a rewritten sentence, and sentences make
miserable flashcards), and that the example must be the *corrected*
sentence — the stored draft still holds the phrasing she just left
behind. Re-accepting the same chunk leaves the existing card alone
rather than resetting a schedule it has been climbing. The whole thing
is best-effort: accepting an edit must never fail because a flashcard
couldn't be made.

The growth journal: kept this month beside kept the month before, the
phrasing that stuck, the patterns that faded. The queries were the easy
part; the honesty is the feature. "Stuck" needs the phrase in a *second*
document, because one document is just the edit where she left it.
"Faded" says nothing at all unless she has been writing lately —
otherwise a month away from Petal comes back to her as progress, which
is the one way this could lie. And a suggestion had to start recording
when she *decided* it, not when the model proposed it, so 0012 adds
resolved_at and backfills the old rows to their created_at.

It lives as a second tab in the garden, and it feeds the kitten: after
an accept she now sometimes hears something true of her alone, once per
line, half the time, never waited for.

Claude-Session: https://claude.ai/code/session_016y6gyuHkQXPiEuW8RGQyua

internal/db/db.go

@@ -459,6 +459,24 @@ CREATE TABLE personal_words (
     created_at DATETIME DEFAULT CURRENT_TIMESTAMP,
     PRIMARY KEY (user_id, lang, word)
 );
+`,
+		},
+		{
+			// The growth journal reads the suggestions table as a record of what the
+			// writer has been learning, and that reading only works if a row is dated
+			// by *her decision* rather than by the model's proposal. `created_at` is
+			// when a checkpoint offered the edit; a suggestion offered in April and
+			// accepted in June is June's growth, not April's.
+			//
+			// Existing rows are backfilled to created_at — which is exactly the
+			// approximation the journal would have had to make anyway, and is very
+			// nearly right in practice since edits are settled minutes after a
+			// checkpoint. Only pending rows keep a NULL: nothing has been decided.
+			name: "0012_suggestion_resolved_at",
+			stmt: `
+ALTER TABLE suggestions ADD COLUMN resolved_at DATETIME;
+UPDATE suggestions SET resolved_at = created_at WHERE status != 'pending';
+CREATE INDEX idx_suggestions_resolved ON suggestions(status, resolved_at);
 `,
 		},
 	}

internal/db/db_test.go

@@ -83,3 +83,70 @@ func TestOpenMigratesAndSeeds(t *testing.T) {
 		t.Errorf("expected exactly 1 local user after reopen, got %d", users)
 	}
 }
+
+// TestResolvedAtBackfill runs migration 0012 against a database that predates
+// it, which is the only shape that matters: on the live box the suggestions
+// table is years of settled edits with no resolved_at to their name. Backfilling
+// to created_at is exactly the approximation the growth journal would otherwise
+// have had to make, and a pending row must stay NULL — nothing has been decided.
+func TestResolvedAtBackfill(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "old.db")
+	d, err := Open(path)
+	if err != nil {
+		t.Fatalf("open: %v", err)
+	}
+
+	// Rewind to the state before 0012: drop the column and forget the migration.
+	if _, err := d.Exec(`DROP INDEX idx_suggestions_resolved`); err != nil {
+		t.Fatalf("rewind index: %v", err)
+	}
+	if _, err := d.Exec(`ALTER TABLE suggestions DROP COLUMN resolved_at`); err != nil {
+		t.Fatalf("rewind schema: %v", err)
+	}
+	if _, err := d.Exec(`DELETE FROM schema_migrations WHERE name = '0012_suggestion_resolved_at'`); err != nil {
+		t.Fatalf("rewind migration record: %v", err)
+	}
+	if _, err := d.Exec(`INSERT INTO documents (id, user_id) VALUES ('d1', ?)`, LocalUserID); err != nil {
+		t.Fatalf("insert document: %v", err)
+	}
+	for _, s := range []struct{ id, status string }{
+		{"s-old", "accepted"},
+		{"s-open", "pending"},
+	} {
+		if _, err := d.Exec(
+			`INSERT INTO suggestions (id, doc_id, from_pos, to_pos, original, replacement, explanation, type, status, created_at)
+			 VALUES (?, 'd1', 0, 3, 'teh', 'the', 'x', 'grammar', ?, '2026-01-02 03:04:05')`,
+			s.id, s.status,
+		); err != nil {
+			t.Fatalf("seed %s: %v", s.id, err)
+		}
+	}
+	d.Close()
+
+	d2, err := Open(path)
+	if err != nil {
+		t.Fatalf("reopen (migrate): %v", err)
+	}
+	defer d2.Close()
+
+	// Compared against created_at read back the same way: the driver renders a
+	// DATETIME column itself, so the assertion is "the same instant", not a
+	// particular text format.
+	var settled, created *string
+	if err := d2.QueryRow(
+		`SELECT resolved_at, created_at FROM suggestions WHERE id = 's-old'`,
+	).Scan(&settled, &created); err != nil {
+		t.Fatalf("read settled row: %v", err)
+	}
+	if settled == nil || created == nil || *settled != *created {
+		t.Errorf("resolved_at = %v, want it backfilled from created_at (%v)", settled, created)
+	}
+
+	var pending *string
+	if err := d2.QueryRow(`SELECT resolved_at FROM suggestions WHERE id = 's-open'`).Scan(&pending); err != nil {
+		t.Fatalf("read pending row: %v", err)
+	}
+	if pending != nil {
+		t.Errorf("pending row got resolved_at = %v, want NULL — nothing was decided", *pending)
+	}
+}