task_games__sokoban__closest_box_goal_label

Contract

1. Domain: games
2. Scene id: sokoban
3. Public task id: task_games__sokoban__closest_box_goal_label
4. Supported query_id values: single
5. Answer schema: option_letter
6. Annotation schema: bbox
7. Program schema: select_closest_box_by_matching_goal(boxes, matching_goals, distance=manhattan); scene=sokoban; scope=closest_box_goal_label
8. Scalar annotation checked: true

Program Contract

Program: select_closest_box_by_matching_goal(boxes, matching_goals, distance=manhattan); scene=sokoban; scope=closest_box_goal_label

Candidate set: the visible game board, pieces, tokens, cards, tiles, marked state, legal-move cues, result panels, and labeled options inside the closest_box_goal_label objective scope. Operands: visible scene state and prompt-bound operands named by boxes, matching_goals, distance, manhattan, sokoban, closest_box_goal_label. Operation: evaluate select_closest_box_by_matching_goal over the candidate set using the visible game state, rules, legal moves, comparisons, counts, simulations, or option-selection constraints encoded in the program expression; generation enforces a unique final answer. Output binding: answer uses the option_letter schema; generation binds a unique final answer. Annotation witnesses: annotation uses the bbox schema; the prompt/annotation contract defines the minimal visual witnesses. Query ids: single.

Reasoning Operations

Families: ranking, spatial_relations, matching

Generation Notes

1. The board shows lettered colored boxes and matching colored goal dots.
2. No box starts on its matching goal.
3. The task asks which labeled box is closest to its matching colored goal dot.
4. Distance is Manhattan grid distance: row steps plus column steps, ignoring walls, boxes, the player, and other board objects.
5. The closest box is unique by construction.
6. Annotation is the scalar bbox of the selected box cell.
7. Prompt wording comes from src/trace_tasks/resources/prompts/games/sokoban/games_sokoban_v1.json.