A small company with a narrow claim
EcoKure Pty Ltd is an Australian early-stage technology company building deterministic verification and evidence-control infrastructure for regulated and high-assurance workflows. It is founder-led and pre-revenue, and this page is written on the assumption that you would find both out anyway.
The details a procurement team asks for
| Item | Detail |
|---|---|
| Legal entity | EcoKure Pty Ltd, an Australian proprietary company |
| ACN | 699 693 779 |
| ABN | 98 699 693 779 |
| Registered | 30 June 2026, Queensland, Australia |
| Governing law | Queensland, Australia |
| Stage | Pre-revenue. Founder-led. No outside capital taken to date. |
| Headcount | One. Stated because a vendor-risk assessment will ask, and because it changes what we should be trusted with. |
| Certifications | None held. Controls are built to be auditable, which is a different and weaker claim. |
| Intellectual property | Four Australian provisional patent applications covering DTL, DAX, DCLA and DELA. Filed, not granted. |
The finding that made this a company
The first working piece was a deterministic mathematics checker. The question behind it was narrow: could an exact symbolic method answer a question faster and more reliably than asking a language model to reason its way to an answer a calculator already knows.
It could. But testing it against the best available models exposed something larger than the thing being looked for. Frontier models are far more fragile under adversarial and edge-case conditions than their averages suggest, and the failures are not the kind a confidence score catches.
That is the origin of the whole architecture. Not an argument that models are bad, and not an attempt to build a better one. An argument that where the consequence of a wrong answer is regulatory, financial or physical, the answer should be checked by something that cannot be persuaded, and the check should leave a record somebody else can examine.
The same instinct runs through the rest of it. We withdrew three published benchmark results after finding they did not hold, including the most impressive one. That record is on the evidence page rather than deleted, because a company selling verification cannot be selective about verifying itself.
Four commitments, each with a cost we have actually paid
Downgrade the claim to the provable version
Where we are unsure whether a claim is safe, it goes to the weaker wording. This has cost us every headline number we once had.
Publish the failures with the results
Retractions stay online beneath the original posts. Pilot reports include what did not work and the cases the system declined to decide.
Make it checkable without us
Public repositories, a published audit protocol, and evidence packs a third party can verify without our source code or our cooperation.
Fail closed, everywhere
An unavailable check is recorded as unavailable, never as a pass. It is the least convenient design choice available and it is the whole point.
The questions you were going to ask anyway
| Question | Answer |
|---|---|
| How many customers do you have? | None. Pre-revenue. No pilot has yet run through to a production licence. |
| How big is the team? | One person, with external technical review. If your risk process has a minimum vendor size, we will not clear it yet. |
| What happens if you stop? | A fair question at this stage. Source escrow and the fact that the gates are public repositories are both part of the answer, and it should be discussed before a pilot, not after. |
| Are the patents a moat? | No, and we do not claim it. No freedom-to-operate search has been run, and our own prior art note concludes the individual mechanisms are not novel. The durable asset is a non-public database of registered attacks and lanes. |
| Is this AI? | No. The checkers are deterministic rulebooks. EcoKure sits after a model, not inside one, and it works the same regardless of which model produced the output. |
| Why should we take a meeting? | Only if a change in rules or evidence has already cost you work. If it has not, we are not useful to you yet and would rather say so. |
Start with one workflow
Thirty minutes, one workflow, and an honest answer about whether this is worth either party's time.