Antefacts

For procurement

You are choosing between AI suppliers on numbers each of them produced about themselves.

Every vendor benchmark was run on data the vendor chose. Every consultancy recommendation came from a firm that also sells implementation. When the losing vendors ask how the decision was made, the answer is a slide deck. Antefacts runs the selection so that the answer is a record anyone can rerun.

The service

A fixed-price evaluation of the candidates you name

You author the criteria, the weights and the thresholds; they are yours because the decision is yours. We seal them, hash-committed and independently timestamped, before any vendor is seen. Each candidate system is run against held-back items none of them has had, on identical published terms, in a recorded configuration. The scoring harness is published so that a vendor who disputes the outcome can rerun its own submission and check every hash. The result is a bounded, pre-registered measurement of each system and a machine-checked certificate that the procedure was followed exactly as you sealed it.

Two to four candidate systems, one domain, one fixed price agreed before the criteria are sealed. Results can be confidential to you. The fact that the evaluation happened, its scope, its date, any deviation and any challenge are published regardless, because that is what makes the record worth having.

Why it holds

What the losing vendor is left with

A vendor who disputes a conventional bake-off can argue about criteria that were never written down, data it was not shown, and a scoring method that lives in someone's spreadsheet. A vendor who disputes an Antefacts evaluation is left with a pre-registration whose hash was public before it submitted anything, a harness it can run itself, and a defined challenge route with a deadline and a published outcome. Most do not challenge. Those that do produce a record that strengthens the decision rather than a dispute that undermines it.

Who it is for

Buyers whose decisions get audited

Regulated firms selecting an AI system that will itself be assessed under the EU AI Act. Public bodies whose procurement is subject to review. Boards that want the supplier decision on the record before the contract is signed. Standards bodies and industry groups choosing an auditor or a preferred supplier for their members. Anyone whose signature ends up on a choice between systems they did not build.

Neutrality

We have no stake in which vendor wins

We take no fees from and no equity in any vendor evaluated. Any participation fee is published and identical for every candidate. Our commercial relationship with each vendor, including none, is a published field on the pre-registration, countersigned by you. Where a vendor is a competitor of ours or of a company in our group, that vendor is not scored and the evaluation says so. The rules are on the neutrality page and our disclosures on the licensees and disclosures page.

What you receive
01

The sealed pre-registration

Your criteria, weights, thresholds and sample sizes, hashed and timestamped before any vendor was seen.

02

The measurements

Each candidate's result as a stated value within a stated interval under a stated configuration. Bounded, never overclaimed.

03

The harness

The published scoring code, so you, your auditor or a vendor can rerun the evaluation without us.

04

The certificate

A machine-checked statement that the published scores are the output of the sealed method applied to the recorded inputs.

05

The record

An append-only, hash-chained account of every step, deviation and challenge, with roots published where we do not control them.

06

A challenge route

Every vendor may rerun its own submission. A challenge is a defined record with a deadline and a published outcome.

Start

Bring the decision, not the criteria

Tell us what you are choosing between and what the decision has to survive. We will help you turn that into criteria you can seal, then run the evaluation. Contact us.