Background Topology Planner — generic experiment (sample fixture: l05_rooftop_interior)

diagnostic-only input W1 산출은 evidence pack + EvidenceExtractionRuleSet 만. topology / shot intent / unit need / decision ledger 는 후속 wave 에서 생성.

run_id=20260524_1305_36b2a8 / plan_version=btp_w1 / acceptance_mode=sample_fixture / lexicon_hash=f29bbd1eac1f4c68…

§1. 흐름 요약

  1. source evidence → topology candidate → shot spatial intent → background unit need → spatial decision ledger
  2. 현재 wave = W1 = evidence pack only
  3. next: W2 = topology candidate generation (SetTopologyGraph + StructuralStateModel)

§2. source_kind 별 evidence count

source_kindcount
planning_doc60
episode_fulltext91
shot_description129
scene_summary172
entity_catalog63
location_catalog36
existing_artifact135

§3. evidence_type 별 count

evidence_typecount
boundary_hint21
camera_hint10
door_window_hint94
entity_layout_hint124
furniture_hint51
place_hint70
space_hint78
state_hint193
zone_hint45

§4. confidence_band 분포

confidence_bandcount
observed413
inferred_candidate273
ambiguous0

§5. ambiguous / unresolved evidence (≤ 50 rows)

evidence_idsource_kindsource_refquote

§6. sample fixture leakage guard

FORBIDDEN_LEAKAGE_LITERALS check: CLEAN

§7. acceptance (mode=sample_fixture)

hard_pass=True

다음 단계: W2 = topology candidate generation (PlaceContinuityGroup / SetTopologyGraph / StructuralStateModel). 본 wave 의 evidence_id (deterministic sha256) 가 W2 의 evidence_refs anchor 로 그대로 사용됨.