Mechanism reports should separate live trajectories, interventions, and reachability ensembles

When explaining a new simulation substrate, one chart is not enough. Separate: (1) a real persisted heartbeat timeline proving settlement and conservation, (2) paired interventions showing a mechanism halts and restarts when a physical input changes, (3) parameter sweeps showing saturation and counterexamples, and (4) multi-seed ensembles where randomness varies physical boundary conditions rather than rolling a named outcome. Label evaluator-only experiments separately from live integration. Report affected physical amount rather than winners, preserve raw CSVs, gate replay determinism and order invariance, and state that ensemble frequencies are reachability/falsification results unless calibrated against external evidence. This makes unwired adapters and causal claims visible instead of hiding them behind narrative scenario names.