loop-benchmarking

commit 13710ed75e85f22717f6463b1b3bc0b822a0a66b
parent d1b5c77738368fcf645c325b912383d5c69f22ed
Author: Brian Graham <brian@buildingbetterteams.de>
Date:   Sat, 11 Apr 2026 09:59:29 +0200

V2: partial landmarks work (agent hit limit)

Diffstat:
M
tasks/tetris/eval/gameplay-bot-v2/types.ts
 | 
28
++++++++++++++++++++++++++++

1 file changed, 28 insertions(+), 0 deletions(-)
diff --git a/tasks/tetris/eval/gameplay-bot-v2/types.ts b/tasks/tetris/eval/gameplay-bot-v2/types.ts
@@ -123,6 +123,21 @@ export interface SurveyData {
   clickable_elements: number;
 }
 
+/**
+ * Landmarks detected on the loaded page to determine whether a game has
+ * actually rendered (as opposed to a blank 200 OK or an empty DOM). Used by
+ * the game_loads test. A game is considered "loaded" if the body has content
+ * and at least one game-shaped element is present.
+ */
+export interface GameLandmarks {
+  bodyHasContent: boolean;
+  hasCanvas: boolean;
+  hasDomGrid: boolean;
+  hasTetrisRatioElement: boolean;
+  hasManyCellsContainer: boolean;
+  landmarksFound: string[];
+}
+
 /** Configuration returned by calibration. */
 export interface DriverCalibration {
   renderer: RendererType;
@@ -181,6 +196,13 @@ export interface TetrisDriver {
   // -- Lifecycle --
   loadPage(url: string): Promise<{ loaded: boolean; detail: string; errorsOnLoad: number }>;
   surveyPage(): Promise<SurveyData>;
+  /**
+   * Detect high-level "game is present" landmarks on the loaded page. Used by
+   * the game_loads test to decide pass/fail based on what the user would see
+   * (canvas, DOM grid, tetris-ratio element, or a many-cells container) rather
+   * than on whether load-time console errors occurred.
+   */
+  detectGameLandmarks(): Promise<GameLandmarks>;
   calibrate(): Promise<DriverCalibration>;
   recalibrate(): Promise<DriverCalibration>;
   /**
@@ -342,6 +364,12 @@ export interface GameSession {
    * that multiple piece types can each rotate through 2+ distinct shapes.
    */
   rotationShapesByPiece: Map<string, Set<string>>;
+  /**
+   * Landmarks detected immediately after the initial page load. Used by the
+   * game_loads test to decide whether a game rendered at all, independent of
+   * any console errors emitted during load.
+   */
+  gameLoadLandmarks?: GameLandmarks;
 }
 
 /** An event observed during continuous grid scanning. */