task_icons__named_grid__line_adjacency_pair_count¶
- domain:
icons - scene_id:
named_grid - task:
line_adjacency_pair_count - module:
src/trace_tasks/tasks/icons/named_grid/line_adjacency_pair_count.py
Program Contract¶
Program: count.adjacent_pair(scene=named_grid, scope=numbered_row_or_column, adjacency=orthogonal_consecutive, pair_order=unordered, attributes=shape_pair, output=count)
Candidate set: adjacent two-cell windows along the prompt-addressed numbered row or column.
Operands: visible scene state and prompt-bound operands named by named_grid, numbered_row_or_column, orthogonal_consecutive, unordered, and the two shape labels in shape_pair.
Operation: count every adjacent two-cell window in the selected row or column whose two shape ids match the requested unordered pair. Direction does not matter, so A-B and B-A both match. Overlapping windows count separately, so A-B-A contributes two matching adjacent pairs.
Output binding: answer uses the integer schema; generation binds a unique final answer.
Annotation witnesses: annotation uses the segment_set schema; each segment connects the centers of one counted adjacent pair.
Annotation schema: segment_set.
Query ids: row_unordered_adjacent_pair_count, column_unordered_adjacent_pair_count.
Reasoning Operations¶
Families: filtering, counting, spatial_relations
Query IDs¶
row_unordered_adjacent_pair_countcolumn_unordered_adjacent_pair_count
Generation¶
Default answer support is 1..5. Grid sizes are sampled from every
row/column combination in 4..6:
4 x 44 x 54 x 65 x 45 x 55 x 66 x 46 x 56 x 6
The selected grid line is constructed to contain exactly the sampled number of matching adjacent pair windows. Non-matching positions in that line use other icon types to prevent accidental extra matches. Other rows or columns may contain the same icon types as distractors because only the prompt-addressed line is in scope.
Fill style and color are rendered as non-semantic visual variation.
Trace¶
The trace records:
- every named icon with row/column indices, row/column numbers, cell bbox, and icon bbox,
- the two target shape ids/names,
- the queried axis and one-based row/column number,
- the counted adjacent cell pairs and their projected center-to-center segment annotations,
- render-style metadata for row/column number-label text legibility,
- query, answer, grid-size, line-index, shape, and fill-style probability metadata.