task_illustrations__construction_site__missing_patch_label¶
Summary¶
- Domain:
illustrations - Scene id:
construction_site - Implementation scene:
construction_site - Implementation source:
src/trace_tasks/tasks/illustrations/construction_site/missing_patch_label.py
Contract¶
- Domain:
illustrations - Scene id:
construction_site - Public task id:
task_illustrations__construction_site__missing_patch_label - Supported
query_idvalues:single - Query ids:
single - Answer schema:
option_letter - Annotation schema:
bbox_map - Program schema:
select_option(match_patch(source_image, missing_region, options)); scene=construction_site; scope=missing_patch_label
Program Contract¶
Program: select_option(match_patch(source_image, missing_region, options)); scene=construction_site; scope=missing_patch_label
Candidate set: the visible illustrated scene objects, people, regions, tiles, patches, labels, and option panels inside the missing_patch_label objective scope.
Operands: visible scene state and prompt-bound operands named by match_patch, source_image, missing_region, construction_site, missing_patch_label.
Operation: evaluate select_option over the candidate set using the visible illustrated objects, regions, layout relationships, counts, patch/tile transforms, or option-selection constraints encoded in the program expression; generation enforces a unique final answer.
Output binding: answer uses the option_letter schema; one of the visible option letters.
Annotation witnesses: annotation uses the bbox_map schema; the prompt/annotation contract defines the minimal visual witnesses.
Query ids: single.
Reasoning Operations¶
Families: transformation, matching
Task Contract¶
Renders a construction-site source panel with one missing visual region and four or six same-size lettered patch options. The model selects the option letter that restores the missing region.
This public task id is a stable scene-owned visual-option contract. The single public query uses exact patch matching only.
Query Branches¶
| Query id | Program schema |
|---|---|
single |
select_option(match_patch(source_image, missing_region, options, transform=none)); scene=construction_site; scope=missing_patch_label |
Program Metadata¶
- Program signatures:
selection.option_match - Base program contract:
select_option(match_patch(source_image, missing_region, options)); scene=construction_site; scope=missing_patch_label - Parameter axes:
none - Arguments:
source_image: semantic_role; allowedconstruction_site_source_panel; sourceprogram_schema_concretemissing_region: semantic_role; allowedmasked_source_region; sourceprogram_schema_concreteoptions: semantic_role; allowedlettered_patch_options; sourceprogram_schema_concrete- Argument metadata status:
curated - Supported query ids:
single
Answer Contract¶
- Answer schema:
option_letter - Generator
answer_gt.type:option_letter - The answer value is one of the visible option letters.
Annotation Contract¶
- Annotation schema:
bbox_map - Generator
annotation_gt.type:bbox_map - Annotation keys are
missing_regionandselected_option. - Annotation boxes are final-image pixel boxes around the missing source region and the selected patch option. Do not include all options, labels, or context-only source objects.
Prompt And Trace Requirements¶
- Prompt text must come from the
illustrations_construction_site_v1scene prompt bundle, with scene/task/output layers selected deterministically and recorded in metadata. - Render randomness, sampled fonts/styles, option order, crop box, and verifier payloads must be explicit in the instance trace.
- Option count is sampled from
4or6; all option bboxes use the same pixel width and height as the missing region. - Missing-region size is sampled relative to the resolved source image: width
15%-30%, height15%-26%, area at most6.5%. - Source-zone text labels are suppressed in this reconstruction view so the task is patch matching rather than text matching.
- The selected option bbox, answer label, and missing-region bbox must all come from the same
compose_patch_optionsexecution trace.