Opaque-system analysis
Did the algorithm change?
Creators can observe how distribution changes. They usually cannot observe the proprietary ranking system that caused it. WITNESS is designed to reason inside that boundary.
Start with what is observable.
Discovery exposure, source mix, reach, impressions, returning behavior, completion, engagement, timing concentration, and cross-platform divergence can all provide evidence about how treatment changed. None of those measurements grants direct access to a platform's hidden ranking logic.
Competing explanations
Platform distribution change
A change in how broadly or to whom the platform exposes the work.
Audience-demand change
The same distribution opportunity produces weaker audience response.
Content-mix change
Different works, formats, topics, or release cadence alter the observed pattern.
Source-mix change
Traffic composition changes even if the underlying audience remains healthy.
Seasonality or timing
The comparison period differs structurally from the current one.
Reporting or evidence anomaly
The apparent change is partly or wholly a measurement problem.
WITNESS can infer treatment without inventing mechanism.
If discovery collapses while returning-audience quality improves, the evidence may support a distribution-shift Finding. It does not justify claiming that a particular hidden ranking rule changed unless independently observable evidence establishes that claim.
Unsupported leap: “The platform changed algorithm rule X.”