Core distinction
Plausible is not the same as supported.
An explanation is a story that could account for an observation. Evidence is what gives one explanation more support than another. WITNESS is designed around that distinction.
Observation: reach fell 38%.
That statement describes a measured change. It does not establish the cause. “Audience fatigue,” “algorithm change,” “weaker content,” “different traffic mix,” and “reporting anomaly” may all be possible explanations.
An AI answer can skip the hard part.
A language model is optimized to produce coherent continuations and useful responses. Without an evidence discipline, the first plausible explanation can sound like a conclusion even when the available observations do not distinguish it from alternatives.
WITNESS sequence
Evidence
What observations and source constraints are actually available?
Alternatives
What other explanations fit those same observations?
Discrimination
Which evidence strengthens one explanation while weakening another?
Uncertainty
What remains unresolved because the evidence cannot distinguish it?
Finding
Which explanation is best supported now?
Change condition
What future evidence would materially alter the Finding?