Misuse Guide

Do not use Causal Halting as a general termination prover.

Bad conclusions:

  • “The checker returned valid_acyclic, so the program terminates.”
  • “The checker found causal_paradox, so the whole agent is unsafe.”
  • “The checker returned unproved, so the design is invalid.”

Correct conclusions:

  • valid_acyclic: no modeled prediction-feedback cycle was found, with validity scope no_modeled_prediction_feedback_only.
  • causal_paradox: a modeled observation/prediction result controls the same execution it observes.
  • unproved: the question is semantically hard or outside the proof strength used.
  • insufficient_info: the structured artifact does not expose enough identity/effect data.

CHC does not solve the classical Halting Problem and does not prove arbitrary termination.

Production-style analyzers must not return valid_acyclic when execution, result, channel, prediction target, or temporal identity is ambiguous. They should return insufficient_info instead.