How verification informs the work

AK Venture Corp (AKVC) is a company that backs teams whose agent claims can be checked. AKVC Labs runs the platform that checks them.

The thesis

Production agents rest on probabilistic foundations. That is useful for proposing work. It is not a proof, and it is not a substitute for one.

The company works where that difference is the product. The companies in view make claims that a later reader can examine: in clinical decisions, autonomy, privacy-preserving systems, and authorization, an uncheckable claim is a defect.

Evidence, proof, and a record

Three layers make a claim checkable. AKVC Labs operates them.

Calibrated evidence scoring

Claims are taken from a company’s materials and set against an evidence file. A decision model assigns a calibrated score. The score is about that claim. It is not a rating of the company, and it is not a guarantee. Thin support should lead the model to abstain, rather than invent confidence.

Formal proof

Some claims can be stated precisely: algorithmic guarantees, policy invariants, safety properties. Those are checked in Lean 4 and Mathlib, or a similar kernel. A model may draft a proof. The kernel decides whether the proof is valid.

Audit trail

Inputs, model versions, scores, proof artefacts, and human sign-offs go into an append-only log chained by hashes. A later reader can see whether the record was altered.

How the record is used

A calibrated score can show which claims have support and which do not. A formal check can show which claims have been proved. Neither one is a decision by itself. People decide. A person remains accountable for any written assessment.

The platform itself is described on AKVC Labs.

What AKVC looks for

Teams building evidence layers, proof kernels, or products that sit on them. Work that can be read together: a specification, a kernel, and a go-to-market. PASS is a candidate. Verified is a kernel judgement.

This page describes that approach. The programmes are on the programmes page.