Skip to content
docs
Arxo ↗

Judgment channel: boundaries

For LLMs3 sections

What the construct does NOT do: limits, reserved values, neighbouring constructs, and the selection rule.

  • Not decide for the judge. The channel names the wait, never closes it: before the answer the result is NEITHER, a power’s effect never materializes (a validity condition gives no support) — only the explanation of the missing effect changes.
  • Not enumerate witnesses without support. A goal-unbound variable is bound by a store witness: only a positive literal conjunct with a TRUE_ONLY/BOTH pair, arguments as variables, entity_ref, value. A computed argument, a constant, _, negation, not_known/refuted/unknown give no witnesses; the cap is eight continuations per rule. An unbound variable on a non-judgment conjunct leaves it out of analysis (COMPUTED).
  • Not read the quantifier. Per-element quantifier reading stays atomless for now (a known remainder, not an exception).
  • Not pass the judge through BOTH. A consumer conjunct with a BOTH pair is dead: a priority judge never passes through it.
  • Not masked by a term error. INTERPRETATION_REQUIRED, UNRESOLVED_NORMATIVE_CONFLICT, REQUIRES_JUDGMENT outrank term computation errors: “the court decides” shows even where “not computed”.
  • Not compute in the explanation. The explanation walk never counts: head compatibility is by form, witnesses by store reading; a candidate rule with a term error is found by the “by predicate” walk, which never looks for judges.
  • authority is an institution identifier, not a term; metadata issuer and jurisdiction are not terms either.
  • An answer with judgment but no request_schema on the declaration is JUDGMENT_PAYLOAD_MISMATCH: nobody can check the answer’s form.
  • An answer from the wrong organ is JUDGMENT_AUTHORITY_MISMATCH (error), the support rejected, REQUIRES_JUDGMENT stays.
  • Query order is the canonical order_requests, no repeats.
  • The judgment channel vs a court fact vs a reading: the outcome unknown — judgment; the outcome given — a fact; the text twofold — a reading. The choice goes in a comment on the feature.
  • REQUIRES_JUDGMENT vs COMPUTED at NEITHER: the first is “the named organ decides”, the second is “facts missing”. A test asserting only truth_status never tells them apart.
  • adjudicated vs case_input: the organ’s answer vs a party’s allegation.
  • A live vs a dead rule: a query only where the answer changes the outcome.
  • A condition’s judge vs a consequence’s judge: a rule reading a transition fact inherits the transition’s judges; whoever asked about the consequence learns that the tribunal decides — in earlier behavior this stayed silent.

Documentation for Arxo. Writings — blog.arxo.io.

Anonymous visit counts on stats.arxo.io, no cookies.