Glossary: inference, arbitration, and interpretive error space
This family groups the terms that explain how a system moves from incomplete evidence to a final answer. The central distinction is direct: a response may be fluent and coherent while being produced through unauthorized inference, weak arbitration or unexposed indeterminacy.
This layer does not replace proof, source hierarchy or response legitimacy. It explains the mechanics that make those layers necessary.
Canonical terms
- Interpretive error space names the range of plausible but unsafe readings produced by weak canon, weak perimeter or weak authority ordering.
- Free inference identifies meaning completion beyond the retrieved or authorized corpus.
- Default inference identifies the fallback answer produced when no boundary blocks completion.
- Arbitration governs how competing sources, meanings, versions or response paths are chosen, exposed, qualified or refused.
- Indeterminacy names the condition where the available material does not justify a clean answer.
- Interpretive fidelity measures whether the output preserves canonical meaning, not only isolated facts.
Operational sequence
- Identify the interpretive error space.
- Block free and default inference where the corpus does not authorize completion.
- Arbitrate conflicts explicitly.
- Expose indeterminacy instead of smoothing it.
- Test interpretive fidelity through proof of fidelity, canon-output gap and interpretation trace.
This sequence keeps a system from treating plausible completion as legitimate interpretation.
Related pages
- Inference prohibition
- Non-inference regime
- Answer legitimacy
- Proof of fidelity
- Unauthorized synthesis
- Surface coherence
How to read this lexical family
This family explains what happens when a system must fill gaps. Inference is not inherently wrong. The problem begins when the system infers without permission, arbitrates between competing meanings without exposing the basis of the choice or hides indeterminacy behind a fluent answer.
Interpretive error space is the field of possible wrong but plausible outputs. Free inference expands that space. Default inference fills gaps because nothing explicitly prevents it. Arbitration chooses one reading among several possible readings. Indeterminacy marks the cases where no legitimate answer can be produced without adding unsupported material.
Typical misreadings
The first mistake is to treat every inference as hallucination. Some inference is necessary for interpretation, but it must be bounded by canon, source hierarchy and response conditions. The issue is not inference itself. The issue is unauthorized inference.
The second mistake is to treat smoothness as resolution. A smooth answer can conceal uncertainty, unresolved authority conflict or missing evidence. Interpretive fidelity requires the system to preserve the limits of the source, not to replace them with a cleaner story.
Use in audit and routing
Use this family when an output seems plausible but cannot be defended from the available sources. The audit should identify which part of the response was retrieved, which part was inferred, which part was arbitrated and which part should have remained indeterminate.
For routing, this family supports inference prohibition, proof of fidelity, answer legitimacy, interpretive risk and source hierarchy pages. Its role is diagnostic: it locates where the response exceeded the evidence.
How to use this glossary family
This glossary family should be read as a conceptual map, not as a replacement for the individual canonical definitions. Its role is to show how the terms around Glossary: inference, arbitration, and interpretive error space relate to one another and why they should not be collapsed into a single generic idea.
A useful reading starts with the failure pattern. Ask what kind of mistake the family helps prevent: confusing visibility with authority, retrieval with legitimacy, citation with proof, persistence with current validity, action with authorization, or coherence with fidelity. The definitions then become routing surfaces. They help decide which page should be primary, which page should support it, and which concept should remain separate.
Boundary of the family
The family does not prove that a model, search engine or agent follows these distinctions. It provides the vocabulary required to test whether they do. In practice, it should be used with observations, audits, source hierarchies and proof discipline. Without those layers, a glossary can name a risk but cannot show whether the risk has occurred.