PN — H2 (estimation): best config qwen2.5:7b cuts MAE only 0.17 SP vs a…
H2 (estimation): best config qwen2.5:7b cuts MAE only 0.17 SP vs a historical-mean predictor (below the pre-registered 0.5 SP threshold), p=0.104, r=-0.16, N=100, so H0 is NOT rejected; a trivial constant-mean predictor is a strong MAE baseline on long-tailed story-point data.
Source: puma docs/results/h2_estimation_wilcoxon.md · Traceability: corpus unit
PUMA-022· Confidence: repo-backed