How it works
Five steps. The tool is chosen last.
OMEGA is a research system I built and direct. I set each question and make the decisions; AI agents in the system do much of the implementation and checking.
The method follows the question.
- Define the question. Precisely enough that it could come out either way.
- Work out what evidence could answer it. And agree it before anything runs.
- Design the experiment, calculation or proof. Whatever settles this question at the lowest cost.
- Run it, and actively try to break the conclusion. Alternatives, failure conditions, controls that must fail.
- Return the result with its boundary. What survived, what failed, what is still unknown.
Which tools
Every investigation follows the same five steps. What changes from one question to the next is the evidence that would answer it, and so the tool. Claim Check and agent qualification are two applications of the same method.
Depending on the question, that can mean simulation, scientific computing, exact computation, formal proof, fuzzing, controlled experiments, optimisation, reproducibility work or AI-assisted research. The tool is chosen last. AI can propose an answer; only the experiment shows what the evidence supports.
Step through a completed investigation
Choose a completed investigation and step through its actual record.
Claim Check
One way to engage OMEGA, for when the question is specifically: can this consequential technical claim be independently supported? One claim, a check shown to catch a planted failure, and a verdict of Holds, Fails or Can’t tell yet, for a fixed fee agreed in advance.
How Claim Check worksSee a sample report
Research and engineering work. Not regulated certification, an accredited audit, investment advice or a guarantee of performance.
Where the work runs
I turn a question into a controlled computational experiment and run it where it fits, from a local workstation to a EuroHPC supercomputer, keeping the evidence sealed and stating exactly what the result establishes.
OMEGA uses compute for its own research. The institutional allocations below belong to named, approved research projects and are restricted to them: they are not offered to clients or used for client work. They are listed as a record of what has actually run, not as capacity for hire.
ResourceAwardedAvailableUsedResult produced
Local workstationsApple Silicon and a GPU workstation, for development and experimentsNot applicableYesYesYes
JUPITER (EuroHPC)One approved research project in AI evaluationYesYesYes: a preregistered evaluation studyYes, not yet published
Isambard-AI (UKRI AIRR)20,000 GPU-hours, one named research project, to December 2026YesAccess establishedQualification jobs onlyNo
VLQ 24-qubit Star (EuroHPC)10 QPU-hours, one approved research purposeYesNot yet onboardedNoNo
Awarded, available, used and result produced are separate states. Each row claims only the states its evidence supports.