Methodology
From signal to Finding.
WITNESS is built around an investigation sequence that keeps raw observations, reconciled evidence, inference, alternatives, uncertainty, and final Findings distinguishable.
1 — Observe
Describe before explaining.
WITNESS begins with bounded, authorized observations. A change in reach, completion, subscribers, profile activity, or another signal is recorded as an observation before the system assigns a cause.
2 — Reconcile
Make the evidence comparable.
Source identity, timestamps, reporting windows, metric definitions, units, provenance, provider constraints, and contradictions are reconciled before cross-source comparison. If the evidence cannot be reconciled cleanly, that limitation survives into the investigation.
3 — Compare
Use context instead of isolated numbers.
Current observations are compared with baselines, prior periods, related metrics, historical Findings, known entities, releases, campaigns, and other relevant workspace context.
4 — Infer
Form the explanation that best fits.
The system can generate an explanatory hypothesis, but plausibility alone is insufficient. Inference is explicitly downstream of the evidence already collected.
5 — Challenge
Try to break the explanation.
WITNESS evaluates competing explanations, contrary evidence, missing observations, structural anomalies, and alternative causal stories. A Finding should survive challenge better than the alternatives, not merely appear first.
6 — Explain
State what is established, inferred, disputed, and unknown.
The result exposes the supporting evidence, material limitations, uncertainty, competing explanations, and the conditions that could change the conclusion.
7 — Watch
Let new evidence reopen the question.
Important questions can persist as Watches. New material evidence can trigger reevaluation rather than forcing the creator to reconstruct the original context manually.
Evidence quality
Signal Integrity runs underneath the methodology.
A clean-looking input can still be weak, anomalous, manipulated, incomplete, or contextually mismatched. Signal Integrity evaluates evidence quality so downstream reasoning is constrained by what the input deserves.