- claim
The SciDEX substrate does not expose a composition_power_check verb or equivalent pre-promotion gate. As of tick 4, claims derived from Bayesian compositional tests (scCODA, CLR-LMM) on cohorts with n < 100 donors (SEA-AD n~80, Mathys-2023 n~48) can be promoted to canonical status without a power-adequacy screen. This constitutes a systematic reproducibility risk: underpowered composition claims may carry inflated type-I error into the substrate's claim graph without any structural signal of their fragility.
- cohorts_at_risk
[
"SEA-AD (n~80)",
"Mathys-2023 (n~48)"
]
- methods_at_risk
[
"scCODA",
"CLR-LMM",
"propeller"
]
- recommended_action
Add composition_power_check as a required INTEGRATE-phase gate. Gate should enforce: (a) n >= 100 donors for Bayesian compositional methods on sparse cell types, or (b) explicit credible-interval reporting with power caveat in claim metadata.