DOE's Genesis Q Sample Agreement Specifies Verification Circuits Standardized Across Awardees: What DE-FOA-0003657, Published 17 September 2026, Asks for 100 Logical Qubits
Every year at Goldsmiths' Hall in London, coins drawn at random from the Royal Mint are counted and weighed in front of the King's Remembrancer, and the assay work follows afterwards. The Trial of the Pyx has run in some form since 1248, and the Goldsmiths' Company has been responsible for the proceedings since 1580. The detail that matters is where the standard lives. The coins are judged against trial plates held outside the Mint by the national measurement authority, alongside standard weights, to specifications independently established for each coinage. The body that strikes the money does not set the benchmark it is judged against.
On 17 September 2026 the US Department of Energy opened the Quantum Genesis Q Competition, with up to $215 million in planned funding and payments attached to demonstrated quantum computing performance. Read against the competition documents rather than the announcement, DOE has done considerably more definitional work than the press release suggests, including something close to a trial plate of its own.
Practical takeaway. DOE has published working definitions of a logical qubit and a hard operation, its sample agreement specifies verification circuits standardized across awardees, and its published terms require that its own staff be given access to operate an awardee's machine. Applicants may propose modifications to the model terms, subject to DOE approval, and the implementation — and for some architectures the metric itself — is negotiated. Anyone reading a quantum performance claim between now and fall 2028 should ask which methodology produced the number, not only how large it is.
What DE-FOA-0003657 Defines: a Logical Qubit, a Hard Operation, and Scientific Relevance
The request for applications sets out three key performance metrics. A logical qubit is "an effective qubit spread across multiple subsystems that can be operated under a quantum error correction regime to suppress errors or environmental noise." A hard operation is one "from a family of operations which are difficult in the participant's error correction scheme (e.g., non-Clifford operations in stabilizer QEC) but necessary to realize computational universality," and the RFA is explicit that these are "expected to require many elementary quantum operations, becoming the scarce resource in a fault-tolerant computation." Scientific relevance, the third metric, is demonstrated against a problem and solution algorithm specified at the start of the program.
That second definition is where the announcement and the underlying document part company. DOE's public article describes machines "capable of performing hundreds of millions fault-tolerant operations." The RFA's own table puts the first-generation target at 100 logical qubits and 105 hard operations, with a scientific workflow; the sample agreement's milestone text speaks of approximately 104 to 105 hard operations "with a space-time volume of 108," and notes that compilation and routing overhead is not included in those baseline figures and is the awardee's responsibility. Elementary operations, hard operations and space-time volume are three different quantities, and the agreement's completion criteria use hard operations and space-time volume together with the correctness of the outputs. DOE also states that the table's numbers are illustrative and that a specific realization at any generation may differ.
How DOE Pays: $250,000 for an Approved Plan, $1.25 Million for a Prototype, and One Synchronized Evaluation
The payment structure is more specific than an incentive pool attached to a threshold. Awardees receive $250,000 after DOE approves a validation and verification plan developed with validation subject-matter experts, which DOE may approve or reject on feasibility at its sole discretion, and $1,250,000 for a validated intermediate prototype milestone. The $100 million general incentive pool is then split evenly among all awardees that demonstrate a first-generation system at the goal, with two $50 million bonus pools split evenly among those reaching 150 and 200 or more total logical qubits in addition to meeting the other first-generation goals. The RFA works the arithmetic through a four-performer example, and the resulting shares range from $25 million to $100 million.
The timing is a single event, not a race to file first. The sample Other Transaction Agreement makes the incentive milestones "subject to a synchronized, common evaluation date across all active Awardees," sets that date at Quarter 8, allows DOE to extend it unilaterally, and makes each payment contingent on DOE's determination of how many awardees qualified for that pool. The RFA anticipates the final evaluation in September 2028, with the option to extend the program if no awardee reaches the goal. DOE also warns applicants proposing longer timelines that other performers may claim the milestone payments first.
The funding language is equally plain: "All funding under this RFA is subject to the availability of appropriated funds," and funding identified for a future phase, milestone or incentive pool "does not constitute a commitment, promise, or entitlement to such funding." DOE records $2.5 million in Fiscal Year 2026 funding for the competition, with outyear funding contingent on congressional appropriations. Both numbers are true at once, and a reader who carries away only the $215 million learns the wrong thing about what is currently in hand.
Troyer, Nayak and Martinis Posted "Scalable Logical Qubits" on the Same Date
Also dated 17 September 2026, "Scalable logical qubits" (arXiv 2609.20549) by Matthias Troyer, Chetan Nayak and John Martinis states its motive in one sentence: "To make the progress on logical qubits measurable and comparable we introduce the definition of scalable logical qubits." Their definition asks for more than DOE's. A scalable logical qubit is preserved through long computations by repeated quantum error correction, is capable of fault-tolerant universal operations with low-latency real-time decoding and feedback, and is replicable to the hundreds or thousands that applications require. They then characterize it along four coupled dimensions — reliability, scale, capability and performance — and discuss the trade-offs among them.
The two documents are not in conflict. DOE's hard-operation metric is reaching for the same capability dimension the paper names, and the RFA's insistence that reliability and stability are what the hard-operation count really tests is the same instinct. The difference is jurisdictional. The paper proposes a definition the whole field can be measured against, on any machine and by any buyer. The RFA and its agreement fix definitions and a common test for one program, among the entrants who sign it.
The Common Test: Circuits Specified by the V&V Team and Standardized Across Awardees
The sample agreement's incentive milestones are where DOE's benchmark actually sits. Each of the three pool milestones requires an awardee to "demonstrate circuits on at least" 100, 150 or 200 logical qubits, and in each case "the circuits will be specified by the V&V team" and "will be standardized across awardees." The milestone text then describes what those circuits would compile to on today's reference architecture — roughly 104 to 105 hard operations at a space-time volume of 108 — while allowing that the figures "could vary based on genuine innovation and advances in architecture." A common set of test circuits, written by the buyer's verification team and run by every competitor, is a trial plate in all but name. The qualification matters: this is the model agreement published with the RFA, and applicants may propose modifications to those terms subject to DOE approval, so the executed agreements may differ.
Two negotiated layers sit around it. On the intermediate milestone, "DOE and the applicant will, during negotiation, specify, to the extent possible, the methodology used to define and evaluate these metrics." More consequentially, DOE will work with selected applicants "whose quantum computing architecture is not naturally measurable in total logical qubits to identify equivalent technical metrics during negotiation," and judgement of the acceptability of those alternative metrics is "at the sole discretion of DOE", covering both the first-generation milestone and eligibility for the two bonus pools.
That combination is a sensible design for a competition open to superconducting, trapped-ion, neutral-atom and photonic machines at once: one hardware-specific yardstick applied rigidly would pick a winner in the specification. It does mean the comparability sits at two levels rather than one — identical circuits above, negotiated implementation and, for some entrants, substituted metrics below.
The Assay DOE Did Build: National-Laboratory Staff Operating the Machine
The verification provision is the strongest part of the package. "Deep collaboration with and access to DOE and its contractors is required," the RFA says. To verify and validate performance under the agreements, DOE intends to use staff from its national laboratories and reserves the right to engage federally funded research centers, university affiliated research centers "and other similarly neutral third parties." Those personnel "are expected to be given both physical and virtual access to the quantum computers and allowed to operate the device to verify and validate performance," and that verification determines whether milestones were met.
That is an assay in the Pyx sense: the buyer's own people, and neutral third parties, operating the machine rather than reading a vendor's report about it. A separate lab call, described by DOE as up to $45 million in planned funding with $14 million in Fiscal Year 2026 and outyear funding contingent on appropriations, is intended to build a hardware-agnostic validation and verification testbed inside the national laboratories. Taken together with the standardized circuits, DOE has assembled both halves of the arrangement Goldsmiths' Hall depends on — an independent assayer and a common plate — inside a single funding instrument rather than in a standing institution.
How Quentir Reads It
Measurement questions become governance questions the moment public money turns on them, and logical-qubit counts have been travelling as settled figures for a while. In the Defense Monitor we traced how Circle's August disclosure read 813 logical qubits off a public leaderboard and imported that number into a corporate risk statement — a figure borrowed from one methodology and reused in a context that never examined it. Our reading of Quantinuum's C4-Helix demonstration on Helios, published here on 4 September 2026, showed the same thing from the hardware side: a logical-qubit count is only interpretable alongside the code, the encoding overhead and the operations actually available.
The QUOPS framework published on 10 September 2026 by Timothy Proctor and colleagues is a third piece of measurement work from this month, benchmarking the size of the largest computationally relevant circuit a machine can execute successfully together with the speed at which it executes it. Applied across processors from Quantinuum, Google and IBM, it puts roughly five orders of magnitude between measured computational capability and utility-scale applications. Our reading is that a benchmark framework, a proposed field-wide definition and a federal instrument arriving within eight days of one another reflect the same pressure: numbers that used to settle arguments now settle payments.
The practical consequence for anyone buying or underwriting quantum capability is narrow. A performance claim is a claim about a methodology, and the claims evaluated under this competition will have been produced under a common set of verification circuits, negotiated implementations and, in some cases, substituted metrics. How much of that methodology reaches the public record is genuinely open: the documents published so far set out the evaluation machinery without committing DOE to disclose each awardee's negotiated terms. Asking which definition a supplier's headline count was produced under is a reasonable question to put in writing, and the four dimensions the arXiv paper sets out are a usable structure for asking it.
Quentir reads these instruments as they are issued. The dated coverage on the blog is free; the Signature Brief is the paid edition alongside it.
The next fixed point is 19 October 2026, when applications close. The one after that is the Quarter 8 evaluation, which DOE may extend unilaterally, and where the interesting question will be how much of the reasoning behind each accepted or rejected milestone is ever shown.
Sources: US Department of Energy, Office of Science, "DOE Launches Competition to Accelerate Development of World’s First Fault-Tolerant Quantum Computer" (17 September 2026); grants.gov opportunity listing DE-FOA-0003657, "The DOE Quantum Genesis Q Competition" (posted 17 September 2026; applications due 19 October 2026), with the request for applications (Version 1.0) and the sample Other Transaction Agreement published as attachments to it; Matthias Troyer, Chetan Nayak and John Martinis, "Scalable logical qubits", arXiv:2609.20549 (17 September 2026); Timothy Proctor et al., "Benchmarking the computational power of quantum computers", arXiv:2609.12146 (10 September 2026); The Goldsmiths’ Company, "The Trial of the Pyx". Public sources checked 18 September 2026.
Published intelligence, built to inform your own decisions. Published: September 18, 2026.