Merge pull request 'perf(ui): reuse hint result as move preview, skip redundant evaluate' (#85) from feature/hint-preview-reuse into development

A hint's MoveRecord words are upper-cased on decode (for the move-history view),
whereas an EvalResult keeps the backend's lower case. Seeding the move preview
from the hint verbatim flipped the score caption to upper case and nudged its
height. Lower-case the words in previewFromHint so the caption reads the same
as the evaluate path it replaces.

ui/src/game/Game.svelte

@@ -29,6 +29,7 @@
     newPlacement,
     place,
     placementFromHint,
+    previewFromHint,
     rackView,
     recallAt,
     recallToSlot,
@@ -751,7 +752,14 @@
         if (isCoarse()) zoomed = true;
         view = { ...view, hintsRemaining: h.hintsRemaining, walletBalance: h.walletBalance };
         syncWallet(h.walletBalance);
-        recompute();
+        // The hint is the engine's own top-ranked, fully scored legal move: reuse it as the
+        // preview instead of a redundant evaluate (same engine call, same placement). Cancel any
+        // pending evaluate so a stale one cannot clobber it; a later manual edit re-arms recompute().
+        if (previewTimer) {
+          clearTimeout(previewTimer);
+          previewTimer = null;
+        }
+        preview = previewFromHint(h.move);
       }
     } catch (e) {
       // The backend does not spend a hint when there is no move.

ui/src/lib/placement.test.ts

@@ -6,6 +6,7 @@ import {
   newPlacement,
   place,
   placementFromHint,
+  previewFromHint,
   rackView,
   recallAt,
   recallIndex,
@@ -105,6 +106,24 @@ describe('placementFromHint', () => {
   });
 });
 
+describe('previewFromHint', () => {
+  it('reuses the hint move as a legal preview, lower-casing words to match evaluate', () => {
+    const preview = previewFromHint({
+      player: 0,
+      action: 'play',
+      dir: 'V',
+      mainRow: 7,
+      mainCol: 7,
+      tiles: [{ row: 7, col: 7, letter: 'C', blank: false }],
+      words: ['CAT', 'AN'],
+      count: 0,
+      score: 18,
+      total: 42,
+    });
+    expect(preview).toEqual({ legal: true, score: 18, words: ['cat', 'an'], dir: 'V' });
+  });
+});
+
 describe('reorderIndices', () => {
   it('lifts an element and drops it at the given slot among the others', () => {
     expect(reorderIndices(4, 0, 2)).toEqual([1, 2, 0, 3]);

ui/src/lib/placement.ts

@@ -4,7 +4,7 @@
 // payload. It is board-agnostic (the gateway/engine does full legality validation at
 // submit), which keeps it trivially unit-testable.
 
-import type { Tile } from './model';
+import type { EvalResult, MoveRecord, Tile } from './model';
 import type { PlacedTile } from './client';
 
 export interface PendingTile {
@@ -55,6 +55,19 @@ export function placementFromHint(tiles: Tile[], rack: string[]): Placement {
   return { rack: [...rack], pending };
 }
 
+/**
+ * previewFromHint turns a hint's move into a move preview. A hint is the engine's own
+ * top-ranked, fully scored legal play, so its move already carries the score, words and
+ * direction an evaluate would compute for the same placement; reusing it lets the client
+ * skip a redundant evaluate round-trip after auto-placing the hint. The words are
+ * lower-cased to match the evaluate result the caption renders: a hint's MoveRecord words
+ * are upper-cased on decode (for the move-history view), whereas an EvalResult keeps the
+ * backend's lower case, so reusing them verbatim would otherwise change the caption's case.
+ */
+export function previewFromHint(move: MoveRecord): EvalResult {
+  return { legal: true, score: move.score, words: move.words.map((w) => w.toLowerCase()), dir: move.dir };
+}
+
 function usedIndexes(p: Placement): Set<number> {
   return new Set(p.pending.map((t) => t.rackIndex));
 }