Canonical terminology

WITNESS glossary.

The words matter because WITNESS deliberately separates evidence from interpretation. These definitions establish the canonical vocabulary used across the product and its public knowledge graph.

Creator intelligence

The process of reconciling creator signals, comparing context, testing explanations, preserving uncertainty, and determining what the evidence supports.

Observation

A bounded description of what was detected, reported, measured, or supplied before causal interpretation is applied.

Evidence

Source-grounded information used to support, weaken, distinguish, or constrain a claim or explanation.

Inference

An explanatory conclusion drawn from evidence. Inference is not automatically established truth.

Finding

An evidence-backed conclusion that preserves supporting observations, alternatives, uncertainty, limitations, and conditions that could change it.

Watch

A persistent intelligence question that can be reevaluated when material new evidence arrives.

Record

A governed knowledge artifact that preserves evidence, provenance, methods, uncertainty, and commitment boundaries around creator or work knowledge.

Signal Integrity

The WITNESS capability for determining how much an evidence input deserves trust before downstream reasoning relies on it.

Competing explanation

An alternative account of the same observations that must be considered before selecting a leading explanation.

Uncertainty

The explicit representation of what is unknown, incomplete, weakly supported, disputed, or dependent on unresolved evidence.

Provenance

Information about where evidence came from, how it was obtained, and the chain of source or custody relevant to evaluating it.

Reconciliation

The process of aligning source identity, timestamps, metric semantics, units, provider constraints, provenance, and contradictions before comparison.

Baseline

A historical or contextual reference state used to determine whether a current signal is normal, novel, or materially changed.

Material change

A change significant enough to alter a Finding, priority, hypothesis, or creator decision rather than merely creating noise.

Change condition

A specified new observation or evidence state that would materially alter the current conclusion.

Protected principal

The design principle that the creator is the party WITNESS exists to serve and protect; the creator's evidence is not treated as the product.

Canonical evaluation

A governed evaluation state kept distinguishable from raw source reports and model-generated analysis.

Evidence-backed AI

AI reasoning in which sources, observations, inference, uncertainty, alternatives, and conclusions remain inspectably connected.