task_icons__named_field__single_attribute_membership_count

Identity

• domain: icons
• scene_id: named_field
• task: named_shape_count
• module: src/trace_tasks/tasks/icons/named_field/single_attribute_membership_count.py
• prompt bundle: src/trace_tasks/resources/prompts/icons/named_field/icons_named_field_v1.json

Program Contract

Program: count.single_attribute_membership(scene=named_field, scope=all_icons, attribute=shape, output=count)

Candidate set: the visible icon instances, icon attributes, fields, grids, paths, panels, reference items, and labeled option cards inside the all_icons objective scope. Operands: visible scene state and prompt-bound operands named by named_field, all_icons, attribute, shape. Operation: evaluate count.single_attribute_membership 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. Query ids: single.

Reasoning Operations

Families: filtering, counting

Scene And Query

The task renders one panel containing procedurally generated named shape icons. The prompt names one shape, such as star, crescent, plus sign, or lightning bolt, and asks how many icons of that shape are present.

The target shape support is the full procedural named-icon vocabulary in src/trace_tasks/tasks/icons/shared/procedural_named_icons.py. Icons also sample a non-semantic fill_style visual attribute from solid, striped, and dotted; this varies rendering but is not part of this task's answer.

The scene samples several internal arrangement modes while keeping the same public task id and answer contract: jittered/ordered grids, shelf rows, collision-free scatter, clusters by shape, shape stacks, mixed stacks, and a target-stack-with-oddballs mode. Stack modes may use moderately larger target counts because the visual strategy is structural row/column counting rather than pure visual search. Stack-family renderers place same-group icons as compact row/column blocks with minimal internal gaps so a queried icon type can be counted from its stack. The target_stack_with_oddballs arrangement renders only one stack: the target shape icons plus exactly one odd non-target icon in that same stack, with no additional distractor stacks.

Answer Contract

• answer_gt.type = integer
• value is the count of icons whose shape_id equals the target shape
• default answer support is layout-dependent: non-stack layouts use smaller counts, and stack layouts are capped at moderate row/column counts

Annotation Contract

• annotation_gt.type = bbox_set
• one [x0, y0, x1, y1] pixel bounding box for each counted target-shape icon
• annotation boxes are sorted by the witnesses in reading order

Trace Contract

• scene_ir.entities contains one entity for each rendered procedural icon.
• Each entity records shape_id, prompt-facing shape_name, bbox, size, rotation, tint, fill_style, placement group, row/column layout coordinates, and icon noise metadata.
• query_spec.params.arrangement_mode records the sampled arrangement mode.
• scene_ir.relations.arrangement_details records stack/oddball details when applicable.
• execution_trace.shape_counts[target_shape_id] equals the integer answer.
• render_map.counted_instance_ids, witness_symbolic.counted_instance_ids, and projected_annotation.bbox_set are derived from the same rendered instances.

Prompt Contract

• scene_key = single_scene_counting
• task_key = counting_query
• query_key = named_shape_count
• prompts ask for the count of one named procedural shape
• Both output modes include contract-valid JSON examples