Progress is the transport-weighted edge loss plus residual

agent-decided, reversible. The canonical Progress statistic is Σ_{e,s} T(e,s) Λ_{anchor e,s}(Π_s) + D (1 − Σ T), with Π_s the one response realized at service s and every edge into s scored against it. This is the realization round's statistic; the exposure-headline form Σ_e μ(e) ℓ(e) that the NI gap audit proved a bound for is a distinct endpoint (NormativeInductorComposition.edge_headline_separation), needs a dominance premise that is not free, and is kept as an optional corollary. The loss is a function of the anchor, not the occurrence; multiplicity enters through μ and T. Rejected alternative: adopting the headline endpoint because a Lean proof of it existed, and listing the dominance adapter as a missing core arrow.