DEC — The canonical baseline uses N=200 (deep validation of one config for tight…
The canonical baseline uses N=200 (deep validation of one config for tight CIs) while the multi-model sweep uses N=100 per cell (broad exploration across 27 conditions = 2,700 predictions at viable compute); both exceed the N=30 minimum for the chosen test.
Source: PUMA project documentation · Traceability: corpus unit
MEM-029· Confidence: verified-at-primary-source