task_icons__reference_canvas__anchor_position_count¶
Identity¶
- domain:
icons - scene_id:
reference_canvas - module:
src/trace_tasks/tasks/icons/reference_canvas/anchor_position_count.py - prompt bundle:
icons_reference_canvas_v1
Program Contract¶
Program: count.reference_icon_anchor_spatial_predicate(scene=reference_canvas, scope=scene_panel_candidate_icons, reference=left_panel_reference_icon, anchor=marked_scene_icon, predicates=left_of_anchor|right_of_anchor|above_anchor|below_anchor, output=count)
Candidate set: the visible icon instances, icon attributes, fields, grids, paths, panels, reference items, and labeled option cards inside the scene_panel_candidate_icons objective scope.
Operands: visible scene state and prompt-bound operands named by reference_canvas, scene_panel_candidate_icons, reference, left_panel_reference_icon, anchor, marked_scene_icon, predicates, left_of_anchor, right_of_anchor, above_anchor, below_anchor.
Operation: evaluate count.reference_icon_anchor_spatial_predicate over the candidate set using the visible icon attributes, positions, relationships, transforms, counts, comparisons, or option-selection constraints encoded in the program expression; generation enforces a unique final answer.
Output binding: answer uses the integer schema; generation binds a unique final answer.
Annotation witnesses: annotation uses the bbox_set schema; the prompt/annotation contract defines the minimal visual witnesses.
Annotation schema: bbox_set.
Query ids: single.
Reasoning Operations¶
Families: filtering, counting, spatial_relations
Notes¶
The Reference and Anchor boxes are retained in trace metadata but are not part of user-facing annotation. The Anchor icon itself is not part of the counted candidate set.