Run camp-ef-belief-gemma4qat-001/C

Run camp-ef-belief-gemma4qat-001/C: 90% pass rate on belief queries.

Of 30 scored trials (belief query kind), the model passed 90% and failed 10%.

3 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: coreference (pass rate 89% at the high end), evidence_span (pass rate 80% at the high end).

In 3 of 3 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 88e752b6 — asked "Where does Quentin believe the blue folder is? Answer with only the location name (or 'unknown' if they have no belief)."
  expected: the kitchen
  actual: south stairwell

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

• record d067ef31 — asked "Where does Beatrix believe the dusty wrench is? Answer with only the location name (or 'unknown' if they have no belief)."
  expected: the basement
  actual: library

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

• record c7aa9710 — asked "Where does Omar believe the worn box is? Answer with only the location name (or 'unknown' if they have no belief)."
  expected: the gym
  actual: west wing

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