Evidence first, claims second
Most quantum-computing claims fail for reasons that have nothing to do with quantum mechanics: undefined baselines, missing error bars, single-backend results presented as general, and artifacts nobody else can point at. The T5.4 programme exists to make that failure mode visible before it reaches a slide deck.
Current work
T5.4 Claims Audit
In progressA worked audit of a high-profile public fidelity claim, scored against all nine truth rules. The published result was arithmetically sound but established generalisability on only one platform, and at a 12.1% error rate against the roughly 0.1% that fault-tolerant designs assume. The verdict names the specific actions that close each remaining gate.
Read the sample auditEQL compiler
Shipped to W1-2A TypeScript implementation of the EigenOS Quantum Language: lexer, recursive-descent parser, 20 AST node types across 8 precedence levels, a 60-plus unit catalogue with base dimensions, a 25-gate catalogue, and a 10-platform hardware registry. Canonical hashing and Ed25519 provenance. The current build type-checks clean and passes 73 of 73 tests.
Read the language referenceStandards programme
Roadmap publishedA submission path for EQL into the quantum-computing standards landscape: IEEE P3109 and P3120 at high priority, the OMG Quantum Working Group and ETSI at medium. The honest position is a 50 to 70 percent chance of meaningful recognition within 24 months, not a promise of ratification.
Read the standards roadmapCERN and White Rabbit
Co-development thesisAn 18-month co-development charter covering timing-aware experiment provenance, timing-fault analysis, and offline link-stability analytics. The funding reality is stated plainly: CERN does not typically fund companies, so the model is in-kind software contribution under an LGPL-licensed open-source ecosystem, with the academic bridge running through VU Amsterdam.
Read the collaboration thesisThe numbers we hold ourselves to
These are the figures that can be independently checked. Where a result is provisional or a tier has not been reached, the number says so.
1 worked, in progress
Every rule scored, including the ones that did not pass.
73 / 73
Lexer, parser, AST and unit catalogue covered.
10
Superconducting, trapped-ion, neutral-atom and photonic.
60+
The basis for the dimensional type checker in W3-4.
148
Quantum computing entities catalogued for the Q1 2027 outreach cycle.
87.9%
Verified against the source artifact, with the error rate stated alongside it.
On honesty as a positioning strategy
Publishing a verdict that fails rules is not a weakness of the audit method. It is the demonstration that the method is real. An auditor that only ever returns good news has no signal, and the first serious user to test that will find it immediately.
It is also the only reason a vendor would ever submit a claim in the first place.
Put a claim in front of the rules
If you have a result you want to be able to defend, submit it. If you want to know whether your result would survive an audit, submit that.