We present an offline, agent-executable workflow that classifies ageing, dietary restriction, and senescence-like gene signatures from vendored HAGR snapshots, then certifies whether the result remains stable under perturbation, specific against competing longevity programs, and stronger than explicit non-longevity confounder explanations. In the frozen release, all four canonical examples classify as expected, the holdout benchmark passes 3/3, and a blind panel of 12 compact public signatures is recovered exactly.
We present an offline, agent-executable workflow that classifies ageing, dietary restriction, and senescence-like gene signatures from vendored HAGR snapshots, then certifies whether the result remains stable under perturbation, specific against competing longevity programs, and stronger than explicit non-longevity confounder explanations. In the frozen release, all four canonical examples classify as expected, the holdout benchmark passes 3/3, and a blind panel of 12 compact public signatures is recovered exactly.
