Skip to content

task_icons__named_path__path_distance_value

  • domain: icons
  • scene_id: named_path
  • task: path_distance_value
  • module: src/trace_tasks/tasks/icons/named_path/path_distance_value.py

Program Contract

Program: count.path_stops_between_named_icons(scene=named_path, scope=ordered_path_stops, endpoints=two_distinct_icon_types, output=integer)

Candidate set: visible path stops along the single START-to-END icon path. Operands: two prompt-bound named icon types, each appearing exactly once in the rendered path. Operation: find the two endpoint icon occurrences and count the path stops strictly between them by START-to-END order. Output binding: answer uses the integer schema; generation binds a unique count by construction. Annotation witnesses: annotation uses the bbox_set schema for every counted path-stop icon strictly between the two prompt-bound endpoint icons. Query ids: single.

Reasoning Operations

Families: filtering, counting, topology

Contract

  1. The image shows a single continuous open path marked from START to END.
  2. The two prompt-bound named icon types each appear exactly once.
  3. Other path stops use non-endpoint icon types.
  4. The answer is the number of path stops strictly between the two named icons.
  5. answer_gt.type = integer.
  6. annotation_gt.type = bbox_set with one bbox for each counted intermediate path-stop icon.

Query IDs

  • single

Generation

Default answer support is 1..8. The generator adds 2..6 extra stops outside the two endpoint icons so the answer cannot be inferred from total stop count alone.

The endpoint icon types are sampled from the procedural named-icon vocabulary. Endpoint order along the path is an internal generation axis and does not change the public query contract.

Trace

The trace records: - ordered path points in pixel coordinates, - every icon stop with path position, shape, color, fill style, bbox, and role, - the two endpoint icon positions and names, - all positions strictly between the endpoints, - render-style metadata for path and text legibility, - answer, endpoint-shape, extra-stop, and fill-style probability metadata.