"Most labs choose epoch counts by convention. Our genome case study trains one DNA model on genomes spanning three orders of magnitude to find where re-reading stops paying."
"One biological model is one opinion. Abacus compares several on traumatic brain injury data and uses where they disagree, plus what they encode inside, to catch prediction errors."
"Capability evaluation benchmarks say what a model can do, not how it behaves. Ethos records behavior and internals on the same run, so drift and silent updates show up before your users notice."
