The epistemic standard
Intelligence that keeps its reasoning attached.
WITNESS is designed so an answer cannot quietly become evidence for itself. Observations, source evidence, inference, competing explanations, uncertainty, and established findings remain distinguishable.
Investigation protocol
Observe
Collect authorized signals and describe what is present before explaining why it happened.
Reconcile
Resolve source identity, timing, units, provenance, contradictions, and data-quality differences before comparison.
Compare
Compare current evidence against baselines, prior periods, related signals, releases, campaigns, entities, and known patterns.
Infer
Form the explanation that best fits the available evidence without treating plausibility as proof.
Challenge
Actively test the leading inference against competing explanations, missing evidence, anomalies, and contradictory observations.
Explain
State what is established, inferred, disputed, or unknown; show the evidence path and the conditions that could change the conclusion.
Watch
Persist material questions and notify the workspace when new evidence meaningfully changes the state of understanding.
Core invariants
Insufficient evidence is a legitimate result. WITNESS should not manufacture certainty to complete a narrative.
A model may propose, summarize, compare, or infer. Its output does not become established truth merely because it was generated confidently.
Inter-agent transport can move claims and evidence. It does not determine what is true.
Why this matters
AI can explain anything. WITNESS is built to make the explanation answer to the evidence.
A plausible story can be useful, but it is not automatically a justified conclusion. WITNESS is designed around the difference between a coherent explanation and an evidence-supported finding.
Canonical entity record · Public knowledge map · Signal Integrity · AI integration surfaces