Exact guards and finite evidence

The present edge guard is exactly

A tolerance rule, such as declaring an edge only when ${\Delta }_e(a,b)>\eta$, defines a different graph and hence potentially a different quotient order. It is not substituted into ${\mathfrak{E}}_m$ without an explicit revision.

If an implemented receiver law is within $\epsilon$ in total variation of its comparison law for each intervention, the triangle inequality bounds the change of an intervention-pair witness by $2\epsilon$. A positive allowance does not establish exact absence of signalling CheungXXIIB (prop:tolerance).

Finite records require coverage, preparation and calibration assumptions, complete attempted-trial scoring, and any transfer allowances actually used. Cheung's Oai265-B supplies a documentary example of keeping such assumptions separate from its channel conclusion CheungOai265B (ass:statistics, thm:main). Its certificate is not an exact process-matrix zero test or an event-cone selection theorem.