Skip to content

Confidence, evidence and scores

DFX states how it knows each thing rather than hiding it in one number: an evidence tier, a match basis, and a stated breakdown for any score. A candidate or inference is labelled as one.

1 min read

  • Search hits carry a trust tier: sworn (filed under oath with a regulator), verified, observed, thin, unverified, or shell.
  • Ask DFX rows carry an evidence tier: Observed, Current signal, Emerging indicator (not a prediction), or Validated forecast.
  • Family office facts are marked Verified, Strongly supported, Inferred or Estimated. Strategy labels use candidate, probable and confirmed.
  • Scores (Research, Transition, Dealability, Why-now) show what they are made of next to the number. A score is a reason to look, not a verdict.
  • Signal pages show Confidence as a percentage, Historical lift against the base rate, and the falsifier: what would make the reading wrong.
  • Profiles mark each fact: Verified (stated in a filing or by the subject), Calculated (computed by DFX from verified records), Inferred (DFX's reading of indirect evidence), Claimed (one source, not corroborated), Historical (was true, not known to be true now) and Under review (disputed or awaiting a correction).

How accurate is DFX?

A fact is as accurate as the filing it came from, and DFX shows you that filing. Joins between records and derived figures are where error can enter, which is why each states its basis. If something is wrong, report it and it is checked by a person.

Did this answer your question?

Still stuck?

The help assistant knows the page you are on and your plan. A person reads every message.

Write to us
Browse every help article