I write a lot about post-quantum cryptography on this site, and I stand behind every line of it — FIPS 203, 204 and 205 are real standards, harvest-now-decrypt-later is a real risk, and the migration work is genuinely happening. But this site also carries language about quantum computing itself — qubit counts, a "PRODUCTION" status badge, a QAOA optimiser said to rank millions of records in milliseconds — and that language deserves the same scrutiny I'd apply to anyone else's claims. So here is the honest version, written the way I'd want a skeptical engineer to write it about my own work.
The short version: the cryptography migration is real and dated. The quantum-hardware and quantum-optimisation claims are, at best, a description of where the field is heading and where we're experimenting — not a description of a system running in production today. Anyone reading both pieces of content side by side deserves to know which is which, so here's the distinction, made explicit.
NISQ is not a marketing term — it's the actual era we're in
"NISQ" stands for Noisy Intermediate-Scale Quantum, a phrase John Preskill coined in 2018 to describe exactly the hardware generation we still have in 2026: quantum processors with somewhere between a few dozen and a few thousand physical qubits, none of them protected by full error correction, all of them accumulating noise fast enough that circuit depth is the binding constraint, not qubit count. Every publicly documented superconducting, trapped-ion or neutral-atom processor in the world today is a NISQ device. There is no exception. Nobody — not IBM, not Google, not IonQ, not a well-funded startup, not us — is running a fault-tolerant, error-corrected quantum computer in production. That machine does not exist yet, for anyone.
That single fact should reframe how any "quantum" claim gets read. When a page says a QPU status is "production," the honest question is: production of what, running which class of problem, with what error rate, verified against which classical baseline? Those questions have answers for real deployments. They rarely have satisfying answers for early quantum-optimisation pilots, because the field itself hasn't produced them yet.
Physical qubits and logical qubits are not the same number
This is the distinction that gets flattened almost every time quantum computing shows up in a pitch deck or a landing page, including, until now, mine. A physical qubit is a single noisy hardware element — a superconducting loop, a trapped ion, an atom in an optical tweezer. A logical qubit is an error-corrected qubit built by encoding information redundantly across many physical qubits, using a quantum error-correcting code, so that the logical qubit's effective error rate is far lower than any individual physical qubit's error rate.
The ratio matters enormously. Depending on the code and the physical error rate you start from, encoding one reliable logical qubit can require dozens to over a thousand physical qubits. Google's 2023 surface-code result crossed the break-even threshold — meaning a logical qubit finally became more reliable than its constituent physical qubits — using a code distance that consumed dozens of physical qubits for a single logical qubit, at very low logical error rates that are still nowhere near "production" for arbitrary workloads. Harvard and QuEra's 2023 neutral-atom result demonstrated dozens of logical qubits operating together for the first time, a genuine milestone — and it was reported globally as exactly that: a landmark experiment, not a shipped product.
So when a number like "127" is presented next to the word "qubits" without saying physical or logical, a reader has no way to know whether it's describing a chip's raw qubit count (plausible — several vendors have shipped physical-qubit processors in that range and higher) or a claim of 127 error-corrected logical qubits running a real workload (not achieved by anyone, anywhere, as of today, at any vendor). Our site's copy doesn't make that distinction, and it should. I'd rather say plainly: where we reference qubit counts, we mean physical qubits on hardware we access through a cloud quantum-computing provider, not logical qubits we've built or error-corrected ourselves. We haven't built a processor. Nobody outside a handful of hardware labs has.
What QAOA actually is, and what it isn't proven to do
QAOA — the Quantum Approximate Optimization Algorithm — is a real, well-published algorithm, first described by Farhi, Goldstone and Gutmann in 2014. It's designed for exactly the kind of problem our copy gestures at: combinatorial optimisation, like matching or ranking a large set of items against multiple weighted criteria. That part is legitimate — it is a real technique, and running a QAOA circuit against a small instance of a matching problem on real quantum hardware, or a simulator, is something a team can genuinely do today.
What QAOA has not done — not for us, not for anyone publicly, as of this writing — is demonstrate a proven, reproducible advantage over classical solvers on a real-world optimisation problem at meaningful scale. The honest state of the research is: QAOA works on small, often academic, problem instances; its performance is highly sensitive to circuit depth and the specific parameters chosen; and every serious benchmark comparison I'm aware of shows classical heuristics — simulated annealing, tabu search, well-tuned integer programming solvers — still winning on real-world-sized combinatorial problems, including anything resembling ranking millions of records by multiple criteria. If our copy implies a QAOA pipeline is the thing actually ranking 12.4 million credits in production, that's not accurate. What we've done is run early QAOA experiments on small representative subsets, in parallel with the classical solver that does the actual production ranking work today. The classical solver is real and it works. The QAOA experiments are real and they're research. Collapsing that distinction in the copy was a mistake, and it's one I'm correcting.
What's real, what's research, and what was marketing shorthand
Laid out plainly, without the ambiguity:
- Real, live, dated: the post-quantum cryptography migration work — evaluating and piloting ML-KEM (FIPS 203), ML-DSA (FIPS 204) and SLH-DSA (FIPS 205) against our own key-management, TLS and signing estate, exactly as described in the companion piece on this site. This is engineering work against published NIST standards, not speculative.
- Real research direction, not yet production: QAOA-based optimisation experiments run against small representative instances, alongside — not instead of — the classical solver that handles real production ranking and matching today. Genuinely interesting, genuinely early.
- Marketing shorthand that overstated the position: a "PRODUCTION" status badge next to a qubit count, and a claim that a QAOA pipeline — rather than the classical system — is doing the millisecond-scale ranking of millions of records. Neither claim survives the physical-vs-logical-qubit distinction or the actual state of QAOA benchmarking, and both should be corrected rather than left standing next to a rigorous, well-cited cryptography series that would make any careful reader ask why one part of the site is precise and the other isn't.
Why I'm writing this instead of just quietly fixing the copy
Because the fix that matters isn't a wording change — it's the standard. I'd rather have one honest, technically literate essay explaining exactly where quantum computing is useful to us today (largely: preparing for the cryptographic threat it will eventually pose) and where it isn't yet (running our actual production optimisation), than a page that reads well to a non-technical visitor and falls apart under five minutes of scrutiny from anyone who's read a NISQ paper. The PQC migration content on this site earned its credibility by being specific, dated and checkable. The quantum-computing-for-optimisation claims should earn theirs the same way: by being equally specific about what's running today, what's a pilot, and what's a research direction we're honestly excited about but haven't shipped.
That's the only kind of "quantum" claim worth making publicly in 2026 — not because the ambition is wrong, but because a founder who gets the details right on the hard cryptography story and vague on the hardware story has, whether he means to or not, told a reader which parts of the site to trust less. I'd rather there be no such gap at all.