commit 58c112b941608fed3c655c638c0e4c17daa5bb19
parent 67bd49c6e259f78aade3caeae40c3418dedf8071
Author: Brian Graham <brian@buildingbetterteams.de>
Date: Thu, 9 Apr 2026 20:15:12 +0200
Add test #25 rendering_clean, update calibration data
New test detects rendering trail bugs where falling pieces leave old
positions still colored. Checks filled cell growth vs pieces placed
during competitive play (threshold: 8x = trail bug).
Updated calibration: 1d08ee76 (broken rotation, no soft drop),
4949d521 (trail rendering bug, lines never clear).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Diffstat:
4 files changed, 95 insertions(+), 39 deletions(-)
diff --git a/tasks/tetris/eval/gameplay-bot/calibration/1d08ee76.json b/tasks/tetris/eval/gameplay-bot/calibration/1d08ee76.json
@@ -2,32 +2,32 @@
"run_id": "tetris_arch=none_ctx=none_noise=clean_dsgn=none_eff=high_echk=none_hlang=en_lang=js_lint=on_budget=low_model=haiku45_pw=avail_prompt=simple_rndr=none_strat=usub_tst=none_tedit=on_tglob=on_tgrep=on_tread=on_twrite=on_web=on_run3",
"short_id": "1d08ee76",
"label": "DOM game (haiku-4.5, en)",
- "notes": "",
- "human_tested_at": "",
+ "notes": "Super weird play. L-shape rotation is broken: flips and rotates simultaneously (wrong rotation matrix). Down arrow is instant drop (no soft drop, acts as hard drop). Scores only increase on line clear, not on drops. Multi-line clear works.",
+ "human_tested_at": "2026-04-09",
"human_tests": {
- "game_loads": null,
- "game_starts": null,
- "auto_drop": null,
- "move_left": null,
- "move_right": null,
- "move_down": null,
+ "game_loads": true,
+ "game_starts": true,
+ "auto_drop": true,
+ "move_left": true,
+ "move_right": true,
+ "move_down": false,
"rotate": null,
- "hard_drop": null,
+ "hard_drop": true,
"all_pieces_rotate": null,
- "piece_locks": null,
- "new_piece_spawns": null,
- "multiple_pieces": null,
- "line_clear": null,
- "score_changes": null,
- "game_over": null,
- "playable_30s": null,
- "multi_line_clear": null,
+ "piece_locks": true,
+ "new_piece_spawns": true,
+ "multiple_pieces": true,
+ "line_clear": true,
+ "score_changes": true,
+ "game_over": true,
+ "playable_30s": true,
+ "multi_line_clear": true,
"score_scaling": null,
"level_progression": null,
"speed_progression": null,
"next_piece_preview": null,
"game_over_display": null,
"counter_clockwise_rotation": null,
- "soft_drop_distinct": null
+ "soft_drop_distinct": false
}
-}
-\ No newline at end of file
+}
diff --git a/tasks/tetris/eval/gameplay-bot/calibration/4949d521.json b/tasks/tetris/eval/gameplay-bot/calibration/4949d521.json
@@ -2,25 +2,25 @@
"run_id": "tetris_arch=none_ctx=none_noise=clean_dsgn=none_eff=high_echk=none_hlang=en_lang=ts_lint=on_budget=low_model=haiku45_pw=avail_prompt=simple_rndr=none_strat=usub_tst=none_tedit=on_tglob=off_tgrep=on_tread=on_twrite=on_web=on_run1",
"short_id": "4949d521",
"label": "DOM game (haiku-4.5, en)",
- "notes": "",
- "human_tested_at": "",
+ "notes": "Rendering bug: falling piece leaves a trail behind it (previous position never repainted to black). Lines never clear despite filling a row. Score does increase though. Very confusing visually.",
+ "human_tested_at": "2026-04-09",
"human_tests": {
- "game_loads": null,
- "game_starts": null,
- "auto_drop": null,
- "move_left": null,
- "move_right": null,
- "move_down": null,
- "rotate": null,
- "hard_drop": null,
+ "game_loads": true,
+ "game_starts": true,
+ "auto_drop": true,
+ "move_left": true,
+ "move_right": true,
+ "move_down": true,
+ "rotate": true,
+ "hard_drop": true,
"all_pieces_rotate": null,
- "piece_locks": null,
- "new_piece_spawns": null,
- "multiple_pieces": null,
- "line_clear": null,
- "score_changes": null,
- "game_over": null,
- "playable_30s": null,
+ "piece_locks": true,
+ "new_piece_spawns": true,
+ "multiple_pieces": true,
+ "line_clear": false,
+ "score_changes": true,
+ "game_over": true,
+ "playable_30s": true,
"multi_line_clear": null,
"score_scaling": null,
"level_progression": null,
@@ -30,4 +30,4 @@
"counter_clockwise_rotation": null,
"soft_drop_distinct": null
}
-}
-\ No newline at end of file
+}
diff --git a/tasks/tetris/eval/gameplay-bot/tests.ts b/tasks/tetris/eval/gameplay-bot/tests.ts
@@ -768,6 +768,10 @@ async function runCompetitivePlayPhase(
let softDropTestDone = false;
let softDropDistinct: boolean | null = null;
+ // Rendering trail detection: track filled cell growth vs pieces placed
+ let filledCellSamples: number[] = [];
+ let trailCheckPieceMark = 0;
+
while (Date.now() - start < maxDuration) {
try {
const grid = await readGrid(page, cal);
@@ -882,6 +886,15 @@ async function runCompetitivePlayPhase(
}
}
+ // Rendering trail sampling: every ~10 pieces, snapshot filled count
+ if (result.pieces_placed > 0 && result.pieces_placed % 10 === 0 && result.pieces_placed !== trailCheckPieceMark) {
+ trailCheckPieceMark = result.pieces_placed;
+ const sampleGrid = await readGrid(page, cal);
+ if (sampleGrid) {
+ filledCellSamples.push(countFilled(sampleGrid));
+ }
+ }
+
// Execute the AI placement
await page.keyboard.press(cal.controls.drop);
await page.waitForTimeout(100);
@@ -991,6 +1004,33 @@ async function runCompetitivePlayPhase(
}
}
+ // Rendering trail detection: if filled cells grow much faster than pieces placed,
+ // the renderer is leaving trails (old piece positions not cleared)
+ if (result.pieces_placed >= 10 && filledCellSamples.length >= 2) {
+ // In a normal game, filled cells = locked cells - cleared cells.
+ // Each piece adds 4 cells, each line clear removes 10.
+ // With trails, filled cells grow unchecked because old positions stay colored.
+ // Heuristic: if filled count exceeds pieces_placed * 8, trails are likely.
+ // (Normal max without any clears would be pieces * 4; * 8 gives 2x headroom.)
+ const maxFilled = Math.max(...filledCellSamples);
+ if (maxFilled > result.pieces_placed * 8) {
+ result.rendering_trail_detected = true;
+ result.bugs_detected.push("rendering_trail");
+ } else {
+ // Also check if filled cells only ever increase across samples (never decrease
+ // from line clears) AND the latest sample is unreasonably high
+ const onlyIncreasing = filledCellSamples.every((v, i) =>
+ i === 0 || v >= filledCellSamples[i - 1]
+ );
+ if (onlyIncreasing && filledCellSamples.length >= 3 && maxFilled > result.pieces_placed * 6) {
+ result.rendering_trail_detected = true;
+ result.bugs_detected.push("rendering_trail");
+ } else {
+ result.rendering_trail_detected = false;
+ }
+ }
+ }
+
// Store CCW and soft drop results for test derivation
(result as any)._ccwResult = ccwResult;
(result as any)._ccwTestDone = ccwTestDone;
@@ -1035,6 +1075,8 @@ const ALL_TEST_NAMES = [
"game_over_display",
"counter_clockwise_rotation",
"soft_drop_distinct",
+ // Phase 8 continued: rendering quality (test 25)
+ "rendering_clean",
];
interface PhaseState {
@@ -1523,6 +1565,21 @@ function deriveTestResults(
}
}
+ // 25. rendering_clean
+ if (!phaseState.gameplayWorks || !competitivePlay) {
+ results.push(skipResult("rendering_clean", "competitive play phase did not run"));
+ } else if (competitivePlay.rendering_trail_detected === undefined) {
+ results.push(skipResult("rendering_clean", "not enough data to assess rendering trails"));
+ } else {
+ results.push({
+ name: "rendering_clean",
+ pass: !competitivePlay.rendering_trail_detected,
+ detail: competitivePlay.rendering_trail_detected
+ ? "rendering trail bug: falling piece leaves old cells colored after moving"
+ : "piece movement clears old cells correctly",
+ });
+ }
+
return results;
}
diff --git a/tasks/tetris/eval/gameplay-bot/types.ts b/tasks/tetris/eval/gameplay-bot/types.ts
@@ -63,6 +63,7 @@ export interface CompetitivePlayResult {
next_piece_visible: boolean;
speed_increased: boolean;
bugs_detected: string[];
+ rendering_trail_detected?: boolean;
}
/** Result of the calibration phase. */
