Data to …… trials
II · Tool planning

Notices when a task is impossible

Gemma 4 31B QAT

Finding:DEGRADES· grade B(research conclusion)

Declines 18 of 22 impossible tasks; writes a plan for the other 4

Pass ratesingle rate, no difficulty dial
82%

18/22 scored trials passed · 95% interval 62%–93%

breaks ≤ 30%holds if interval reaches 95%0%50%100%
Why it slips

The model writes a plan for an impossible goal instead of declining.

in 100% of failures (4 failures)

Example failure
Trial result: FAIL (one model answer)
FAILfabricated_plan_on_unsolvable:2
[ { "tool": "advance_0", "args": { "service": "svc_0" } }, { "tool": "advance_1", "args": { "service": "svc_0" } } ]
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 62%–93%Grade B: moderate evidence
  • Sample size22 scored trials in total
  • Fresh-seed confirmationNot run yet
  • ReplicationNot run yet
Raw evidenceruns and trials

Back to the capability map →