How the Leakage Report is computed

An award is reported as leakage if and only if its signed (action) date falls inside an exclusion's active window for the same confirmed entity:

activation_date <= action_date
  AND (termination_date IS NULL OR termination_date > action_date)

The termination date is treated as the exclusive end of the window — an award signed on the termination date is not counted.

• confirmed_uei — award recipient_uei equals the exclusion's UEI.
• confirmed_cage — award recipient_cage equals the exclusion's CAGE code.
• Held — name-only or unmatched exclusions are never asserted against an entity.
• Indeterminate — an exclusion with no published activation date can never be reported as active.

confidence is 1.000 only on an exact-key match; held and ambiguous records carry no confidence and never produce a finding. This is the same discipline the healthcare engine uses for NPI matching.

Only the bitemporally-current version of each row participates in the default view (valid_to IS NULL). Substituting a valid-time predicate —valid_from ≤ D AND (valid_to IS NULL OR valid_to > D) — reproduces the exclusion and award picture exactly as it stood on any past date D. That is what makes the report defensible for set-aside-eligibility-at-award-date and FCA-defense use cases.

The exclusion spine is SAM.gov. Award history is USASpending.gov, which is DATA Act public domain and is preferred over SAM for any pre-2022 field to avoid the D&B / DUNS redistribution bar. Pre-2022-04-04 records are flagged and their D&B-sourced fields and legacy DUNS are withheld from public display and the API.

The study exposes exact regulatory facts only. There is no derived “risk”, “fraud”, or “likely-excluded” field anywhere in the data model or the page. A finding is the two dated facts the regulator published, and every record links to the official source to confirm current status. A CI invariants gate fails the build if any inference or score column is introduced.

Published SQL (with expected-result comments) re-derives every figure directly in Postgres. The same computation is a pure, unit-tested function: src/lib/procurement/leakage.ts. Snapshot generated 2026-06-20; methodology leakage/v1; Fonteum.

← Back to The Leakage Report