Is there a fair classical baseline at all?
And is it the strongest reasonable one — not a strawman chosen to make the quantum result look good?
A quantum-advantage claim is a governance object before it is a physics result.
The Quantum Advantage Verification Layer is the KCH governance technology applied to one high-value problem — deciding whether a claim that a quantum method beats the classical alternative is admissible against a fair classical baseline — and it is the current focus of our research.
The precondition
The baseline is the part most easily left out.
A claim that a quantum method beats the classical alternative is only as credible as the classical baseline it is measured against — and that baseline is the part most easily left out, quietly weakened, or made moot by a later, better classical algorithm. A speedup, a sampling-hardness result, or a correctness bound is meaningful only relative to a precisely defined problem, a fair and strongest-reasonable classical comparison, and a path from evidence to conclusion that someone else can re-derive.
QAVL makes those preconditions load-bearing. It reasons over a claim as a precise object rather than a headline — the problem being solved, the cost of the quantum approach, the cost of a fair classical baseline, the cost model that makes those comparable, and the advantage test itself.
What QAVL checks
A disciplined set of questions, before any advantage is credited.
Is the claim what it says it is?
A genuine hardware or information-theoretic advantage — and not a metaphor, or a quantum-inspired classical method dressed up as quantum advantage?
Can the result be reproduced?
From artifacts, rather than taken on assertion?
Is every trust-bearing fact backed by a proper authority?
Rather than self-certified by whoever benefits from it?
Did a human with standing co-sign?
Where the stakes require it — and is the audit chain intact?
The governing commitment is non-compensation: the fair classical baseline is not one input to be traded off against a large speedup. It is a foundational precondition, and its absence ends the claim — no measured advantage, however impressive, can buy it back.
The advantage number is considered only after the foundational checks have passed.
Why this matters
The difference between plausible and accountable.
Evidence governance is the difference between a result that is plausible and one that is accountable. It is what could let a national laboratory cite an AI-assisted or quantum result in a publication, a funder or reviewer admit an analysis into a proceeding, or a research collaboration preserve an anomalous finding without prematurely claiming a discovery.
Three conditions make it timely: AI and quantum results are entering high-consequence workflows faster than governance can be retrofitted; funding and policy attention are turning explicitly toward evaluation, provenance, and trustworthiness; and the reproducibility and auditability pressures in computational science create demand for exactly this discipline.
QAVL is also our fastest, cleanest place to demonstrate the broader KCH technology: verifying a quantum-advantage claim requires no human subjects, no clinical data, no specialized hardware, and no access to a physical quantum device — only the claim, its evidence, and the discipline to judge them fairly.
The connective work around that discipline — linking national laboratories, universities, funders, and research collaborations who do not yet share a common language into a coordinated effort — is the company’s grand-collaborator role, described in full on the Grand Collaborator page.
Evidence and honest scope
What the early results support — and what they do not.
An early implementation of QAVL exists as a working classical kernel. In bounded, pre-registered testing, it correctly separated admissible claims from inadmissible ones, blocked a battery of adversarial attempts to smuggle a claim past the governance checks, and produced signed audit records that could be reconstructed offline. Related work on governed memory showed, in a bounded adversarial study, that an ordered, governed configuration outperformed simpler single-mechanism approaches on predefined safety and auditability measures.
We state the limits plainly, because they are the point.
These results are bounded to their test suites. They support the design; they do not demonstrate production performance, security against every adversary, or clinical or physical safety.
The early kernel uses implementation devices appropriate to a prototype, not a production security architecture. Production would require hardened cryptography, independent security review, and the full operational surround.
The broader thesis — that one evidence-governance discipline serves AI outputs and scientific measurement alike — is a research claim, not a demonstrated result, and remains open until independent domain studies test it against ungoverned and simpler-governed alternatives.
Our standing rule is to amend, hold, or retract any claim that a fair, matched comparison does not support. Leaving the work smaller and truer is a contribution, not a loss.
Collaborate with us
The KCH Research Program is seeking collaborators.
We welcome independent validation of the evidence model and its failure modes, joint benchmarks and comparative studies, simulation-stage pilots in governed scientific and enterprise workflows, and co-authored work establishing evidence governance as a measurable discipline.
If you work in AI governance, quantum information, formal methods, or research policy and see where this belongs, we would welcome the conversation — start one here.
This page describes a research program. It makes no discovery claim and asserts no independently established performance results; every capability statement is bounded by its evidence.
The Immanence Project LLC · Denver, Colorado · Prepared in the structural mode of the Hierarchy of Concepts.