Establish injectivity of a log-derived observation channel

Determine whether any channel derived from an event log is injective on distributions over labelled partial-order process variants, so that distinct process models can always be distinguished from the resulting observations.

Background

Trace-level event logs discard the partial-order information that distinguishes concurrency from balanced sequential interleavings, and the paper proves that every trace-observable comparison inherits this non-identifiability in the relevant setting. The authors leave unresolved whether some alternative channel derived from logged information could be injective outright. This question is explicitly relative to the scheduling law: under a tilted scheduling law, the relative frequencies of interleavings may carry additional identifying information.