task_three_d__object_scene__object_relation_label

Summary

• Domain: three_d
• Scene id: object_scene
• Package: src/trace_tasks/tasks/three_d/object_scene/
• Supported query_id: on_top_of_prop, under_prop, inside_prop
• Answer type: option_letter
• Annotation type: bbox
• Annotation schema: bbox

Program Contract

Program: select(label(candidate_objects, spatial_relation_to_reference = requested_relation)); scene=object_scene; scope=object_relation_label

Candidate set: the visible 3D objects, surfaces, room/street/warehouse structures, spatial anchors, markers, and labeled options inside the object_relation_label objective scope. Operands: visible scene state and prompt-bound operands named by label, candidate_objects, spatial_relation_to_reference, requested_relation, object_scene, object_relation_label plus the active query_id branch. Operation: evaluate select over the candidate set using the finalized 3D scene state, camera projection, object identities, spatial relations, counts, distances, or option-selection constraints encoded in the program expression; generation enforces a unique final answer. Output binding: answer uses the option_letter schema; generation binds a unique final answer. Annotation witnesses: annotation uses the bbox schema; the prompt/annotation contract defines the minimal visual witnesses. Query ids: on_top_of_prop, under_prop, inside_prop.

Reasoning Operations

Families: spatial_relations

Contract

The image uses the object_scene renderer: a perspective 3D floor, table, or platform scene with projected objects, markers, references, or paired views depending on the task. The public task id defines the stable objective contract; query ids are used only for genuine semantic operations within that contract. Render style, camera, canvas preset, object placement, labels, colors, and prompt wording variants are generation metadata, not public task axes.

The verifier computes the answer from finalized scene metadata and projection records, not from pixels. The prompt bundle is three_d_object_scene_v1 under src/trace_tasks/resources/prompts/three_d/object_scene/.

Candidate option objects are sampled from the curated object_scene-compatible named-object pool derived from the domain-wide THREE_D_NAMED_OBJECT_SHAPE_TYPES support. This excludes broad render-only small objects such as drum from named MCQ candidates. Candidate objects are rendered with a relation-task scale multiplier so they read as small props relative to the larger reference prop. The same candidate object pool is used for every relation query id; query ids change only the requested spatial relation and compatible reference prop. Piano is excluded from this task's extra context props. For under relations, the selected object is scaled smaller, placed in the open interior of a table or arch support, and accepted only when its projected bbox does not collide heavily with visible legs, posts, or the lintel/tabletop. Under objects use normal depth ordering rather than a foreground overpaint bias.

Annotation Contract

Annotation is a scalar bbox around the selected visible object. The selected object is the only visual witness; option text is not annotation.

Prompt And Trace

The trace records selected prompt keys, camera/projection data, object or marker records, rendered pixel witnesses, answer-support metadata, and the solver fields needed to recompute the answer and annotation from the same finalized scene.

Determinism

Generation is deterministic from instance_seed, explicit params, config defaults, prompt bundle, and code versions. Answers and annotation come from the same finalized 3D scene trace.