OMEGA

Discuss a technical challenge

For researchers, engineering teams and organisations. A few sentences is enough: what you are trying to find out, and what decision depends on it.

Discuss a technical challenge

Or write directly to warren@omegaprotocol.org

Warren Smith, founder and independent researcher. OMEGA is currently led by me alone. AI agents in the system do much of the implementation and checking.

Research and engineering work. Not regulated certification, an accredited audit, investment advice or a guarantee of performance.

What you can bring, and what comes back

You don’t need to know how to investigate it. Working that out is the job: the question comes first, and the method is chosen to fit it.

You might bring

  • a result you don’t trust
  • a model or algorithm that behaves strangely
  • two approaches you can’t choose between
  • an expensive experiment you want to de-risk first
  • an AI system you’re not sure you can rely on
  • a technical claim that needs an independent check
  • an important question nobody knows how to test

The answer can be

Yes
It holds, and you are told exactly under what conditions. A spending limit was proved to hold in every state a protocol model can reach.
No
It does not, with the evidence and a way to reproduce it. A battery model gained 2.44% of its lithium.
A better approach
The question was right, but the method, model or experiment should change, and you get the evidence for the alternative.
The evidence is not enough
What is missing, and what would settle it. A power-grid solver’s results looked reasonable, but with no second, independent solver to compare against, “plausible” was as far as the evidence went.

Whichever it is, you receive

  • the answer, and what it means for your decision
  • the evidence, and exactly what was tested
  • the code, data and, where possible, a check you can re-run

Every finding says exactly what was tested and what the result does, and does not, establish.

Three ways to work with OMEGA

Open investigation

You have an important technical question and don’t yet know what kind of work would answer it.

Bring it as it is. Working out what would answer it is part of the job.

Discuss a technical challenge

Claim Check

You have one specific claim a decision depends on, such as “the paper’s result reproduces from the released code”.

I check that one claim independently and return Holds, Fails or Can’t tell yet, for a fixed fee agreed in advance.

How Claim Check worksSee a sample report

Agent qualification

You need to know whether a particular AI setup can reliably do a particular job before you rely on it.

I test that exact setup on realistic and deliberately difficult tasks. The result never says “this agent is safe”: it says what worked, what failed and where a person must stay in the loop.

Related public work, not client work: an AI agent’s tool server passed 66 of 66 tests and its own replay check said VERIFIED, while it wrote 100,000 bytes where it was allowed 4,096. The finding

Agent qualification

How an investigation runs

  1. Pin down the question. Precisely enough that the answer could come out either way.
  2. Decide what evidence would actually answer it. Agreed with you before anything runs.
  3. Find the cheapest test that would settle it. An experiment, a calculation, a simulation or a proof.
  4. Run it, then try to prove the conclusion wrong. Other explanations, cases where it should break, and checks shown to catch a planted mistake.
  5. Report what survived. What held, what failed and what is still unknown.

The tool is chosen last. AI can propose an answer; only the experiment shows what the evidence supports.

OMEGA starts from your question and uses whatever can answer it, rather than selling one technology: a simulation, a controlled experiment, a mathematical proof, or a program tested with thousands of generated inputs.

Step through a completed investigationHow it works

Fields

Some questions ask whether something works as claimed; others ask which way to go before an expensive step. Choose a field to see example questions, and the completed public work in it where it exists:

Explore questionsEach example is a proposed investigation, never a past client.

Where the work runs

Experiments run on local workstations and, for one approved research project so far, on a EuroHPC supercomputer. Institutional allocations like that belong to named research projects and are restricted to them: they are not offered to clients or used for client work.

Research infrastructure, and what has actually run on it

Who you reach

Warren Smith

Founder and independent researcher, OMEGA

I work on scientific simulations and software, on testing AI systems, and on mathematical proofs that software does what it should. The connecting idea is simple: turn difficult technical uncertainty into something that can actually be tested.

OMEGA is currently led by me alone. There is no account manager and no handoff: the person who reads your question is the person who designs and runs the investigation. Where a question needs domain expertise I don’t have, I say so in the first reply.

Research on agents that each pass their own checks while together breaching a shared limit has been accepted as a briefing and poster at AMLUCS 2026 and as a research poster at the IDAI Workshop on Multi-Agent Safety and Security 2026.

Presentations & postersFixes and reports in other people’s softwarePublic code on GitHubResearch archive