Define what the system must know to answer
A relationship question may need entity identity, direction, relationship kind, scope and period. A capability question may need a source, description and operating conditions. Requirements differ by semantic target. Stating them before evaluation prevents presenting choices that the available state cannot discriminate.
Separate missing information from incompatibility
A missing location, date or contract does not prove that an option is false. Deterministic filters may reject only when known inputs demonstrate incompatibility within defined scope. UNKNOWN, STALE and incomplete data lead to research or abstention; they should not become FALSE to make a decision simpler.
Use typed choices without forcing reality into them
An evaluator compares state with a declared Choice Space that needs definitions, boundaries, exclusivity and options such as UNKNOWN, OTHER or UNRESOLVED where appropriate. A selected option does not show that the space covers reality. Retaining the distribution and version makes it possible to review which question was asked and what choices were available.
Gather targeted context when a requirement is missing
An unanswerable question can identify which observation is missing: an independent source, resolved identity or current date. That gap directs concrete research instead of a cheaper but irrelevant model call. When new evidence arrives, the state fingerprint changes and only dependent dimensions need reevaluation.
Conceptual example
Hypothetical: the question is whether a manufacturer can serve a region. Its profile gives a headquarters but no source about service coverage. Location cannot distinguish “yes” from “no.” AXIGNAL leaves the question UNKNOWN or seeks specific evidence instead of inferring reach from the registered address.
Scope and limits
An answerability check does not guarantee that later evaluation is correct; it tests whether the state meets defined requirements for that question. Requirements, Choice Space and policy need versioning and traceability. A model output cannot fill missing context or authorize canonical truth. A missing requirement should be named precisely so the next observation can address it. Otherwise a new evaluation may repeat the same unsupported choice while appearing to add confidence. The question contract, not a convenient model response, determines what evidence is sufficient.
Editorial basis
This page explains product doctrine and boundaries. It does not demonstrate source coverage or observed outcomes.
Read the product modelExplore how AXIGNAL reasons
Explore AXIGNAL