Skip to main content
← Lab

calibration

null resultPython · CPUModel behavior & reasoningJuly 6, 2026

the question

Does a small model's multiple-choice confidence match its accuracy?

what came out

No -- systematically overconfident, measured from exact top-logprobs. A 2x2 (size x ARC difficulty) study: miscalibration compounds with both smaller size and harder task.

method & receipts

  • Result: null result — an honest negative, reported as it came out.
  • Reproducible one script re-runs the whole thing from scratch.
  • Independently reviewed an adversarial review pass, kept in the repo.
  • Machine-readable claims every reported number, in a checkable file.
  • Tested — a correctness/benchmark suite ships alongside the code.

→ read the code and re-run it

github.com/v-code01/calibration