Validation Units

Validation Units (VUs) are the commercial unit of completed Validation Intelligence work.

1 VU = one unique, completed canonical relationship evaluation between one material information unit and one independent canonical evidence item.

VUs measure the evidence relationships actually completed. They are not tokens, requests, sessions, model calls, or arbitrary credits.

Current /v1/validation calculation

Completed canonical pairings determine usage. For example:

VU = information unit count × independent evidence count
Information unitsIndependent evidence itemsCompleted relationshipsVUs
111 × 11
22Potentially 2 × 2 canonical evaluations4

The endpoint decomposes the supplied information into material units. Evidence independence is established in the canonical result; callers should not estimate VUs from request count or raw evidence-array length alone.

What adds no new VUs

In-progress work, polling, replay, failed execution, duplicate evidence, duplicate task delivery, and rejected idempotency conflicts add zero VUs. Completion and usage accounting are atomic.

  • Duplicate evidence can remain in the response for traceability, but duplicate_of evidence does not create another independent relationship evaluation.
  • Replaying an identical completed request with the same Idempotency-Key adds 0 VUs.
  • Failed or incomplete relationship graphs add 0 VUs.
  • Projected or inherited results add 0 new VUs when they do not perform a new canonical relationship evaluation.

For example, one information unit supplied with two identical evidence items still has one independent evidence item. The response can preserve both supplied items, while the completed graph consumes 1 × 1 = 1 VU.

What does not change the quantity

VU quantity does not depend on:

  • the overall determination (supported, unsupported, conflicting, or uncertain);
  • the relationship label;
  • source prestige;
  • model, provider, client, latency, or presentation.

These inputs and outcomes do not redefine the unit of work. Only explicit acquire requests retrieve public HTML. Supplied text is evaluated unchanged and citation metadata triggers no acquisition. Inspection failure fails closed. See Source Inspection.

The frozen meter version is vu_v1. Pricing, plan sizes, rates, and commercial packaging are not defined by this documentation. See Usage and Limits to inspect organization capacity and consumption.