One fact layer · three data products · every claim sourced

Account Facts

Three data products that a general-purpose AI cannot reliably produce — because they are derived from a reviewed, source-linked fact layer and keep "insufficient info" honest instead of guessing.

How the fact layer works

  • Three-valued evaluation: satisfied / not satisfied / insufficient info. Missing input, illegal dates or non-numeric values never become a definite true/false.
  • Publish gate: only approved claims whose evidence version is unchanged are published. Drafts, disputed and unknown values are never encoded as "free / eligible / fully online".
  • Source binding: every claim cites its source, and a source must explicitly declare which product it covers — same-domain is not enough.
  • Existence of a path ≠ guaranteed approval. Absence of a record ≠ path does not exist.

Fact-layer revision 9a8402ca089d · data baseline 2026-10. Rules change frequently — confirm with the official source before applying.

FAQ

What is the fact layer?

A reviewed, source-linked dataset of conditional claims (product × identity × channel × time). Only approved claims whose evidence has not changed are published; drafts, disputed or unknown values never become "free / eligible / fully online".

Why show "insufficient info" instead of "not eligible"?

Because absence of a recorded path is not proof that a path does not exist. We only mark "not eligible" when a claim explicitly states so (e.g. a tourist pass is excluded).

Are these three products the same data as the bank pages?

Yes. All of them derive from the same fact layer and the same shared renderer, so amounts, conditions and evidence stay consistent across pages.