registry/compare
Direct answer 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=126 public cases each
direct-answer
v1 by @system69.0
60.5 to 76.5 · holdout 68.5
role-then-format
v2 by @ilya91.7
86.5 to 95.0 · holdout 90.3
Intervals do not overlap. The 22.7 point gap is significant at this sample size.role-then-format wins
Metricdirect-answerrole-then-formatBetter
Public score69.091.7right
Holdout score68.590.3right
Interval width16.08.5right
Cost per run$1.021$1.850left
Median latency2.0s2.4sleft
Sample size126168right
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 \ direct-answer role-then-format \ --model claude-fable-5