Atinamos Labs

Atinamos Labs

AI agents•verification•autonomous commerce•applied research

WHAT LABS IS FOR

We test the parts that need evidence, not just a convincing diagram.

Atinamos Labs is where we build and test the parts of agentic systems that are easier to discuss than prove: discovery, payment, execution, evidence and buyer control.

Labs is also our public research surface: the place where selected methods, working papers, experiment notes and limitations can be inspected rather than hidden behind a product claim.

PUBLIC RESEARCH

Ideas should survive being written down properly.

Published work carries a date, version and status. A public draft is exactly that: open to inspection, not presented as settled fact or peer-reviewed research.

Experiment report · Published record

Autonomous Buyer Proof #3: discovery, evidence, policy and payment

An Atinamos-operated buyer independently discovered external machine services, checked exact-route viability, queried published evidence, applied a fixed procurement policy, selected a seller without human seller selection, paid 0.001 USDC and exactly validated the purchased result.

Read publication
Experiment report · Published record

Deterministic Assurance: IBANforge IBAN validation

Two paid external x402 controls settled on Base and the returned valid field matched independently computed ISO 13616 MOD-97 expectations for one valid and one deliberately invalid IBAN fixture.

Read publication
Publication practice

Methods, status labels and corrections

How we separate observation from interpretation, publish limitations and handle revisions.

Read our methods

Browse all public research

CURRENT WORK

Experiments with a job to do.

Public registry updated 2026-09-09

verification active

Assurance

Evidence-first assurance for autonomous services. We record what happened and leave the trust decision with the buyer.

View project
verification active

Verification

A public evidence surface for checking independently recorded service behaviour without turning it into a universal trust score.

View project
autonomous commerce live

Render Check

A machine-buyable render test used to explore autonomous service discovery, payment, execution and proof.

View project
AI agents active

WebMCP experiments

Practical work on websites that can expose useful, bounded actions directly to capable agents while remaining usable by people.

View project
A WORKING RULE
“You can pay to be tested. You cannot pay to be trusted.”

We prefer evidence that can be inspected over scores that ask to be believed. Where a buyer needs a policy decision, the buyer should control the policy.

BUILD

Start with the smallest real thing that can answer the question.

OBSERVE

Record what actually happened — including the failures.

CLAIMS

Say no more publicly than the evidence can support.

USEFUL?

If an experiment earns its place, turn it into something useful. If it does not, stop.