Run camp-full_map-nano-002/breaking_point

Run camp-full_map-nano-002/breaking_point: 91% pass rate on belief queries.

Of 46 scored trials (belief query kind), the model passed 91% and failed 9%.

10 additional trial(s) ended in ERROR (infra failures, cost cap, or cancellation) and are excluded from the pass rate above.

4 failure(s) recorded; failures not yet verified -- use "Verify failures" on this page (or scripts/verify_past_failures.py).

Diffuse weakness (below 90% but no outright flip) on: temporal (pass rate 89% at the high end).

In 2 of 4 classifiable belief failures the model reported where the object ACTUALLY was, not where the queried entity BELIEVED it was -- it collapsed their perspective into reality.

Notable failures

• record 93fd41ae — asked "Where does Quinn believe the bright basket is? Answer with only the location name (or 'unknown' if they have no belief)."
  expected: the basement
  actual: photo studio

  the model reported the object's TRUE final location, not what the queried entity actually believed -- it collapsed belief into reality

• record 928137f8 — asked "Where does David believe the red folder is? Answer with only the location name (or 'unknown' if they have no belief)."
  expected: the basement
  actual: supply closet

  the model reported the object's TRUE final location, not what the queried entity actually believed -- it collapsed belief into reality

• record 7d4fafff — asked "Where does Elliot believe the sturdy basket is? Answer with only the location name (or 'unknown' if they have no belief)."
  expected: the photo studio
  actual: library

  the model's answer matches the 'query_agent_own_last_move' shortcut, not the verified belief