Investigation patterns
Questions worth investigating.
WITNESS is most useful when a metric alone cannot answer the creator's real question—or when the obvious explanation may be wrong.
“My reach dropped. Is my audience losing interest?”
Reconcile discovery, returning audience behavior, completion, engagement, timing, and related signals before treating lower reach as audience decline.
“Two platforms disagree. Which signal should I trust?”
Compare provider definitions, reporting windows, account identity, provenance, data quality, and Signal Integrity before combining or prioritizing the evidence.
“Views are up. Did the release actually get stronger?”
Separate distribution volume from audience depth by comparing completion, returning behavior, follow-on actions, and baseline treatment.
“Was this spike organic, paid, anomalous, or structural?”
Challenge the apparent cause using timing patterns, source changes, campaign evidence, historical baselines, and structural anomaly signals.
“Did this campaign change the audience or just the traffic source?”
Compare audience-quality signals before and after the event rather than equating acquisition volume with audience transformation.
“Is the algorithm treating my work differently?”
Build a bounded inference from observable distribution and audience patterns while keeping unobservable platform internals explicitly uncertain.
“What changed after the release?”
Reconcile time-aligned signals across connected sources, compare with the pre-release baseline, and surface only material changes.
“Will WITNESS tell me if this conclusion stops being true?”
Convert the Finding into a Watch with explicit change conditions so new material evidence can reopen the investigation.
The common pattern
Observation ≠ explanation.
Every use case begins with a measured change or question. WITNESS then reconciles the evidence, compares context, forms an inference, challenges it against alternatives, explains the conclusion with uncertainty, and can Watch for material changes.
Boundaries matter