Work with Omega
Bring a question.
If there is something important about a system, model, experiment or technical result that you need to know, Omega can investigate it independently.
You don’t need to know how to test it.
Tell Omega:
- what you’re working on
- what you need to know
Omega will say whether the question looks answerable, what access or evidence would be needed, and what a small first piece of work could look like.
See what this could look like
AI and its toolsExplore ↓
- Your question
- Did this permission or control actually govern what the agent executed?
- What Omega might do
- Compare what was authorised with the operation that actually ran. Inspect the effect from outside the agent and test whether a refusal really prevents it.
- What you get
- A finding on the permission boundary, with the operation and observed effect behind it.
- Related real work
- Agent permission boundaries →
Scientific modelsExplore ↓
- Your question
- Does this model actually conserve the quantity it claims to conserve?
- What Omega might do
- Recalculate the quantity from the model’s output, check it across the run, and compare refinements or a control model.
- What you get
- A conservation check, the observed drift or agreement, and the conditions under which it was tested.
- Related real work
- PyBaMM lithium conservation →
Software and calculationsExplore ↓
- Your question
- When this solver reports that it converged, did it actually?
- What Omega might do
- Check the reported result against independent sanity checks and test which inputs the solver should accept or reject.
- What you get
- The tested inputs, observed outputs and a finding on what the solver’s success signal supports.
- Related real work
- pandapower transformer vector groups →
Maths and proofsExplore ↓
- Your question
- Does this guarantee actually follow from the stated assumptions?
- What Omega might do
- Make the assumptions explicit, look for a counterexample, and check a proof or finite model where appropriate.
- What you get
- A proof, a counterexample or an unresolved question, with machine-checkable evidence where the method allows it.
- Related real work
- Collective-bound proofs and counterexamples →
Scores and measurementsExplore ↓
- Your question
- Does this score or checker actually measure the property it is being used to represent?
- What Omega might do
- Compare the score with a separate reference check and introduce known failures to see whether the checker catches them.
- What you get
- Evidence of what the checker distinguishes, what it misses and where the comparison stops.
- Related real work
- An AI judge’s pass versus a solver’s policy check →
What you get
- the finding
- the evidence behind it
- what was and was not established
- a reproducible check where appropriate
Bring a question.
Want the details?
Project terms and how the work beginsExplore ↓
Independent checks for important technical questions
Bring a question. Build the evidence. See what the evidence shows.
Start with a specific question or an open research problem. Before work starts, agree what will be checked, what access is needed and what you will receive. If the evidence does not settle the question, the report will say so.
From your question to a result
What you bring
- A system, model, experiment or technical claim
- The question that matters, in your own words
What Omega does
- Checks whether the question can be answered with evidence available or obtainable
- Agrees exactly what is being asked and which version is being checked
- Builds or chooses a way to test it
- Shows that the check can catch a relevant error
- Runs the check and keeps a record of what happened
What you get
SUPPORTED CONTRADICTED INCONCLUSIVE
- A clear answer where the evidence supports one, or an explanation of why it does not
- A record of the test and the evidence behind the finding
- What the result does not show, what remains unknown and how to check it again where possible
The check depends on the question. Omega might count again from raw data, run the same input two ways, measure an effect from outside the system, or check a mathematical proof. Some questions cannot be answered with the evidence or access available; the report says so.
What gets built serves the question
Omega may run an experiment, count again from raw data, compare two versions, observe what actually happened, or check a proof by machine. That can include mathematical or formal questions, where a proof, a computation or an independent check could settle part of the problem. The question decides the method.
You do not need to choose a tool in advance. The report explains the check used and what it can and cannot show.
Bring a question that matters
Start in ordinary technical language. If the question contains a statement that can be settled either way, Omega will pin it down, along with what would count as settling it, before the decisive run.
- Did this control govern what actually executed, and the effect that followed?
- Does this model conserve the quantity it claims to conserve?
- Does this score measure the thing it is being used to represent?
- Can this question become a check that runs on every release?
The first decision is whether the question can be settled from the evidence available or obtainable. There is no claim that arbitrary questions can be. A clear no, or a verdict of INCONCLUSIVE where the question was pinned to one statement, is a legitimate successful result.
Bring us a question
You describe the system and the question in one or two sentences, with whatever evidence or access already exists. The reply says whether it can be answered, what access would be needed, and whether one small fixed-scope piece of work is the right shape.
Full deliverablesExplore ↓
Deliverables
- 1
The question, the system and the work agreed in writing
- 2
The version and inputs that were checked, where they can be fixed
- 3
What Omega could observe or test, and what it could not
- 4
The check or experiment, including a test that shows it can catch a relevant failure
- 5
A record of what ran and the evidence needed to understand or repeat it
- 6
A finding that says what the evidence shows, what it does not show and what remains unknown
- 7
A new check after a fix, when a fix is offered and a new run is agreed
Sensitive information and where work can runExplore ↓
Sensitive systems can stay where they are
The work does not always need the system, its source code, its production data or its credentials to be handed over. Raw sensitive material can stay inside the owner's own environment.
Where the question allows it, Omega can provide a fixed set of checks for your own team to inspect and run inside your environment.
Before execution, agree
- what runs
- what Omega receives
- what remains local
- what evidence is required
- what that evidence can and cannot support
Omega can report only on evidence it can inspect. A stronger check may need different access. Omega has no accreditation or security clearance and has not claimed prior work inside a cleared environment.
Different forms of workExplore ↓
The work can take different forms
- 01Investigate a specific question: find out what the evidence shows, or why it cannot yet answer.
- 02Check a system or result independently: test it against agreed questions and show where the check could fail.
- 03Keep checking as the system changes: rerun a check that has already been tested, and record what changed.
- 04Work together on research: design and run experiments for an open question with the people who know the field.
These are options, not steps you have to buy in order. Repeated checks make sense only after the check itself has been tested. No price or result is promised here.
Research collaborationExplore ↓
Work together on research
Bring an open question, two competing explanations, or a result that needs a stronger test. Omega can help design and run the experiment with the people who know the field.
Agree the question, each person’s role, what can be shared and what would count as a useful result before work starts. You remain responsible for what the finding means in your field. See the current research and open questions.
What Omega’s work does and does not certifyExplore ↓
What an assessment is not
- Not certification. Omega is not a certification body and certifies nothing to anyone.
- Not a statement that a system is safe. A property that passed is a property that passed.
- Not a penetration test, a general AI red team, or a promise to find a vulnerability.
- Not extra engineering capacity on work you already know how to do.
- Not universal verification. What was not tested is listed, not implied.
The system owner keeps the judgement about what to do with the evidence. Omega does not manufacture a pass, and does not offer certification, accreditation, security approval or regulatory compliance.