registry/compare
Two tier check 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=140 public cases each
two-tier-check
v1 by @nils87.9
81.4 to 92.3 · holdout 86.7
role-then-format
v2 by @ilya91.7
86.5 to 95.0 · holdout 90.3
Intervals overlap. The 3.8 point gap is not significant at this sample size.
Metrictwo-tier-checkrole-then-formatBetter
Public score87.991.7right
Holdout score86.790.3right
Interval width10.98.5right
Cost per run$1.685$1.850left
Median latency2.8s2.4sright
Sample size140168right
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 \ two-tier-check role-then-format \ --model claude-fable-5