Private preview · research prototype · no checkout · no customer data

Method / the control loop

Evidence, with
a handbrake.

The method is intentionally slower than recommendation theatre. Each record moves through a local evidence lifecycle; uncertainty creates an exception, not a confident sentence.

01DiscoverFind a candidate source
02ExtractSave the observable passage
03ClassifyFact / interpretation / unknown
04ReviewHuman checks the boundary
05RecheckTimestamp and link state

What a record needs

Provenance before polish.

  • Direct URL and access date in UTC
  • Observable fact, kept separate from interpretation
  • Unknowns and prohibited inferences
  • Freshness rule and reviewer state
  • Approval gate before any handoff
BLOCKED

What we do not infer

A pricing page does not prove total cost. A programme page does not prove acceptance. A product listing does not prove stock, quality, delivery, safety or outcome.

If the claim cannot be supported by the saved source, it stays out.

Reader pattern

If X matters, inspect Y next.

Every guide ends with a small, observable next step. The goal is not to make a decision for you; it is to make the decision legible.