feat(planner): add entity-ID-keyed ScenarioBelief snapshot projection

docs/context/issue_2538_scenario_belief_planner_projection.md

@@ -29,23 +29,51 @@ Issue #2538 adds a planner-facing ScenarioBelief projection helper:
 - The stream-gap planner remains opt-in for uncertainty consumption. Missing or malformed
   uncertainty metadata still keeps deterministic pedestrian rows.
 
+The additive issue #8050 diagnostic seam also provides
+`project_belief_aware_planner_input(...)` for the currently supported projection target
+`BeliefGuidedLocalPlanner`. It retains every `ScenarioBelief.agents` entry in an immutable,
+entity-ID-keyed `tracks` mapping, including entries absent from the visible legacy rows, and
+reports distinct `no_belief`, `empty_belief`, `projection_target_not_supported`, `invalid_belief`,
+and `projected` statuses. Its diagnostics expose both retained and projected track counts.
+Serialization is versioned and deterministic; legacy observations and
+the existing stream-gap path are unchanged.
+
+`track_id` is the entity identifier supplied by one `ScenarioBelief` snapshot, not a
+visible-observation row number. The current representation does not expose retirement generations,
+so the diagnostic reports `identity_generation_available: false`,
+`identity_reuse_safe: false`, `retired_track_count: null`, and
+`stateful_identity_admitted: false`. Stateful consumers must reset at an externally supplied
+lifecycle boundary. No generation or continuity token is fabricated, and no retirement, reuse, or
+benchmark/safety benefit is inferred. Aggregate confidence and the radius covariance block are
+explicitly labelled as adapter-derived and unavailable-as-modelled in the diagnostics.
+
 ## Claim Boundary
 
-This proves only that ScenarioBelief uncertainty metadata can reach one planner-compatible local
-observation shape and can be consumed by the existing stream-gap uncertainty gate on a fixture. It
-does not prove better navigation, safety, SNQI, planner performance, perception calibration, or
-benchmark movement.
+Safe claim: this is a deterministic, entity-ID-keyed projection of one `ScenarioBelief` snapshot
+that prevents joining uncertainty metadata by visible-row position. It proves only that
+ScenarioBelief uncertainty metadata can reach one planner-compatible local observation shape and
+can be consumed by the existing stream-gap uncertainty gate on a fixture. It does not prove
+cross-lifecycle identity continuity, better navigation, safety, SNQI, planner performance,
+perception calibration, or benchmark movement.
 
 ## Validation
 
 ```bash
-uv run pytest tests/planner/test_stream_gap_planner.py -k "uncertainty or scenario_belief" -q
-uv run ruff check robot_sf/planner/scenario_belief_adapter.py tests/planner/test_stream_gap_planner.py
-uv run ruff format --check robot_sf/planner/scenario_belief_adapter.py tests/planner/test_stream_gap_planner.py
+uv run pytest tests/planner/test_scenario_belief_track_projection.py \
+  tests/planner/test_scenario_belief_uncertainty_gate.py \
+  tests/planner/test_stream_gap_planner.py \
+  tests/representation/test_scenario_belief.py -q
+uv run ruff check robot_sf/planner/scenario_belief_adapter.py \
+  tests/planner/test_scenario_belief_track_projection.py
+uv run ruff format --check robot_sf/planner/scenario_belief_adapter.py \
+  tests/planner/test_scenario_belief_track_projection.py
 ```
 
 ## Follow-Up
 
 The next useful step is a runtime observation-builder path that produces a ScenarioBelief during an
-environment step and routes this projection into a planner selection or smoke command. Until that
-exists, this remains a unit-level planner interface smoke.
+environment step and routes this projection into a planner selection or smoke command. A
+track-generation/retirement owner is required before a stateful planner can claim reuse-safe
+identity semantics. Until those gates exist, this remains a unit-level planner interface smoke;
+#8050 stays open with `implementation_admitted: false`, and no downstream planner work is part of
+this slice.