entity-aware-rri

name	entity-aware-rri
description	Use when ARIA-NBV work touches target/entity selection, GT OBB crops, target-specific RRI labels, target-conditioned VIN fields, or entity-aware diagnostics.
metadata	{"mode":"implementation","not_when":["scene-level RRI or geometry work with no target/entity contract","rollout planning where target-RRI semantics are already fixed","docs-only glossary wording without RRI, target, or actor-visibility impact"],"handoff_to":["diagnose-aria for concrete target-RRI failures or suspicious metrics","nbv-geometry-contracts for pose, camera, crop-frame, or unit issues","counterfactual-rollout-planner for multi-step target-gain evaluation","plan-grill for advisor-facing target-protocol decisions"],"evidence_required":["V0 versus V1 actor-visibility path and OBS-SEL / PRED-Q / GT-EVAL status","target crop support, invalid-target reasons, and target/scene RRI comparison","focused RRI/data-handling test or explicit missing seam"],"applies_to":["aria_nbv/aria_nbv/rri_metrics/**","aria_nbv/aria_nbv/data_handling/**","aria_nbv/aria_nbv/vin/**","docs/contents/thesis/**"],"triggers":["target RRI","entity-aware","GT OBB","OBS-SEL"],"must_read":["docs/contents/thesis/roadmap.qmd#roadmap-m3","docs/contents/thesis/questions.qmd#rq3-representation","aria_nbv/aria_nbv/rri_metrics/AGENTS.md","aria_nbv/aria_nbv/data_handling/AGENTS.md"],"canonical_sources":["docs/contents/theory/rri_theory.qmd#target-specific-error-gain","docs/contents/theory/rri_theory.qmd#scene-rri-role","docs/contents/theory/rri_theory.qmd#nbv-objective","docs/contents/theory/candidate_sampling_target_selection.qmd#actor-visible-target-selection","docs/contents/theory/candidate_sampling_target_selection.qmd#target-rri-qh-connection","docs/contents/thesis/roadmap.qmd#roadmap-m3","docs/contents/thesis/questions.qmd#rq3-representation","aria_nbv/aria_nbv/rri_metrics/AGENTS.md","aria_nbv/aria_nbv/data_handling/AGENTS.md"],"literature_refs":["quality-driven-rri","egocentric-aria-substrate","VIN-NBV-frahm2025"],"tool_refs":["mcp__code_index.search_code_advanced","mcp__code_index.get_symbol_body"],"verification":["cd aria_nbv && uv run pytest tests/rri_metrics","cd aria_nbv && uv run pytest tests/data_handling/test_vin_offline_store.py"]}

Entity-Aware RRI

OMX Integration

OMX owns the surrounding plan or execution loop; this skill supplies the target/entity evidence contract. Return actor-visible versus oracle-only boundaries, target validity evidence, and the focused RRI verification loop.

When To Use

Use this skill for:

• target/entity eligibility and selection policies
• GT OBB cropped RRI, target mesh/point support, and invalid target reasons
• target-conditioned VIN batch fields, labels, and diagnostics
• scene-level versus target-level RRI comparisons

Do not use it for generic scene-level RRI unless an entity/target contract is involved.

Read First

1. docs/contents/theory/rri_theory.qmd for target-specific RRI definitions.
2. docs/contents/theory/candidate_sampling_target_selection.qmd for actor-visible target selection and GT-label boundaries.
3. docs/contents/thesis/roadmap.qmd section M3 and docs/contents/thesis/questions.qmd section RQ3 for current thesis scope.
4. aria_nbv/aria_nbv/rri_metrics/AGENTS.md
5. aria_nbv/aria_nbv/data_handling/AGENTS.md

Rules

• Treat the Quarto theory pages and package AGENTS.md files listed in metadata.canonical_sources as the source of truth for formulas, V0/V1 semantics, and target-visibility policy.
• Keep actor-visible inputs, oracle-only labels, and GT evaluation crops separate in evidence and claims.
• Keep unsupported targets explicit with invalid reasons; do not turn them into low-RRI valid samples or scene-level fallback labels.
• Log or inspect target and scene RRI separately when comparing behavior.
• Hand off pose, camera, crop-frame, or unit uncertainty to nbv-geometry-contracts before changing labels.

Verification

• cd aria_nbv && uv run pytest tests/rri_metrics
• cd aria_nbv && uv run pytest tests/data_handling/test_vin_offline_store.py
• cd docs && quarto render contents/thesis/roadmap.qmd contents/thesis/questions.qmd when target-aware claims change
• make check-agent-memory for memory or skill edits