Ingest
Collect new public material from each configured station inside explicit source, date and volume limits.
Attributed evidence
How it works
Every stage produces an inspectable output. The system is designed to reduce a noisy public conversation without severing the evidence chain.
The pipeline
Collection is incremental: later cycles pull only new material, preserve prior states, and skip empty snapshots rather than manufacturing movement.
Collect new public material from each configured station inside explicit source, date and volume limits.
Attributed evidence
Identify pain points, objections, requests, questions and praise with verbatim supporting excerpts.
Structured themes
Collapse semantically similar language while retaining distinct-source and distinct-voice counts.
Stable concepts
Match concepts across consecutive snapshots and classify what appeared, disappeared or materially shifted.
Versioned change
Track concept mutation, decay and related fingerprints across cycles and separate research niches.
Theme lineage
Link falsifiable expectations to evidence, assess their conditions and update a calibrated trajectory.
Bet verdicts
Surface cross-source bursts and sentiment shifts only when configured evidence thresholds are met.
Actionable signal
Expectation
Define a specific condition and time horizon before future evidence arrives.
Evidence
Each new cycle adds attributable observations and recalculates the trajectory.
Verdict
The bet becomes on track, at risk, met or missed—with evidence attached.
Built-in restraint
Trend extrapolation is not causal inference. Public discourse is not a representative population sample. Watchdogs makes those boundaries explicit rather than hiding them behind a score.
Interpretation rules
Put a watchdog on it