Data to …… trials
I · Narrative

Belief tracking in deliberately misleading stories

gpt-oss-120b

Finding:HOLDS· grade B(research conclusion)

98% correct (41/42)

Tested up to chain length 10, scramble 0.8, distractor events 2

Pass ratesingle rate, no difficulty dial
98%

41/42 scored trials passed · 95% interval 88%–100%

breaks ≤ 30%holds if interval reaches 95%0%50%100%
What the few failures have in common

The model answers where the object really is, not where the person believes it is.

in 100% of failures (1 failures)

Example failure
Trial result: FAIL (one model answer)
FAIL
workshop
Open this trial →Open the run it came from →
Evidencewhat this finding rests on
  • Ground truthDeterministic simulation: the right answer is computed, not judged
  • StatisticsWilson 95% interval 88%–100%Grade B: moderate evidence
  • Sample size42 scored trials in total
  • Fresh-seed confirmationNot run yet
  • ReplicationNot run yet
Raw evidenceruns and trials

Back to the capability map →