registry/compare
Role then format against Role then format
Both scored on the same cases against claude-fable-5, same judge. The intervals tell you whether the gap is real or noise.
suite reasoning-chainmodel claude-fable-5n=168 public cases each
role-then-format
v2 by @ilya91.7
86.5 to 95.0 · holdout 90.3
role-then-format
v2 by @ilya91.7
86.5 to 95.0 · holdout 90.3
Intervals overlap. The 0.0 point gap is not significant at this sample size.
Metricrole-then-formatrole-then-formatBetter
Public score91.791.7same
Holdout score90.390.3same
Interval width8.58.5same
Cost per run$1.850$1.850same
Median latency2.4s2.4ssame
Sample size168168same
Reading this verdict
- Overlap
- Two entries are only meaningfully apart when their confidence intervals do not overlap. A higher number with an overlapping interval is a coin flip, not a win.
- Holdout
- The public score can be gamed by tuning to visible cases. A gap between public and holdout is the tell.
- Reproduce
- evalness compare \ role-then-format role-then-format \ --model claude-fable-5