Execution Identity

Prediction feedback depends on identity. The checker distinguishes:

  • same_execution: the result controls the execution it observes before that execution ends.
  • future_execution: the result controls a later execution.
  • resumed_execution: the result controls a continuation of the same logical execution.
  • retried_execution: the result controls a retry with a new execution identity.
  • forked_execution: the result controls a parallel child execution.
  • unknown_identity: the artifact does not expose enough identity information.

Default policy:

  • same execution before end -> causal_paradox;
  • future execution -> valid_acyclic;
  • post-run audit -> valid_acyclic;
  • unknown identity -> insufficient_info.

Adapters should preserve original IDs (trace_id, span_id, parent_id, run IDs, task IDs) so a human can audit the boundary.