Run camp-breaking_point-gemma4qat-001

Run camp-breaking_point-gemma4qat-001: 43% pass rate on belief queries.

Of 68 scored trials (belief query kind), the model passed 43% and failed 57%.

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

In 19 of 39 classifiable belief failures the model's wrong answer matches the 'last_mentioned_location' shortcut -- a plausible but ungrounded guess rather than tracking what that entity actually witnessed.

Notable failures

• record 68f6abf7 — asked "Where does Sofia believe the blue tablet is? Answer with only the location name (or 'unknown' if they have no belief)."
  expected: the workshop
  actual: rooftop terrace

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

• record e67dad0b — asked "Where does Priya believe the dusty box is? Answer with only the location name (or 'unknown' if they have no belief)."
  expected: the kitchen
  actual: wine cellar

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

• record 318073c8 — asked "Where does Finn believe the narrow folder is? Answer with only the location name (or 'unknown' if they have no belief)."
  expected: the loading dock
  actual: reception area

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