Demo record. Built 2026-10-23, fact-checked 2026-10-24 after reviews by two AI models from different providers and a check of load-bearing facts against primary sources, amended 2026-10-31: binding per team and an ablation in every quantum dossier, as the record's history states. No domain expert has reviewed it yet. Draft protocol, not yet registered.
The question behind it. Can quantum computing beat the best classical methods on biologically relevant molecules? That question is too broad to resolve as stated, so this record commits to one narrow piece of it that a public benchmark can decide.
For the fixed Fe₄S₄ molecular model specified below, a workflow using quantum hardware can produce a verifiable variational state meeting the stated energy target at least twice as fast as the strongest publicly tested classical workflow, under the same execution-budget ceiling.
An iron-sulfur cluster model with 54 electrons in 36 orbitals, pinned to one public file
The strongest publicly tested classical workflows for the same target, under the same budget
Elapsed time to a verified state at or below −327.245436 hartree
Solver performance: speed and quality on one fixed mathematical problem
Chemical prediction: agreement with a physical measurement
Biological usefulness: whether a result changes an experimental decision
Why this tier. Under this draft rule only results produced after registration can count, so the tier rests on what has been reported so far. On this exact model, the best quantum-hardware energy posted to the public Quantum Advantage Tracker is −326.913 hartree (IBM and RIKEN, November 2025), about 0.33 hartree above this record's target; classical entries there and in the 2026 GPU DMRG work reach lower. More broadly, Wellcome Leap's April 2026 announcement names no winner of its $5 million prize for quantum advantage over best-in-class classical baselines; the largest 2026 hybrid calculation on a protein system states that its authors do not regard it as the basis of an advantage claim; and an 18-author analysis found no evidence yet for generic exponential advantage in ground-state energy estimation, while leaving polynomial speedups open. This is a bounded review finding, not a proof that no such result exists.
Every possible result maps to a tier in advance. The classical side keeps improving, so the bound quantum result stays public for at least 180 days before it is judged, and every classical answer published by the cutoff counts.
The synthetic cluster [Fe₄S₄(SCH₃)₄]²⁻, as the fixed CAS(54 electrons, 36 spatial orbitals) Hamiltonian in the Li and Chan active-space repository, revision 84433831a680f70665cf43fc8692e0210da92be3, file Fe2S2_and_Fe4S4/Fe4S4/fe4s4, SHA-256 0b840102a4894fee4648944f2bf3bb5d85828deb4b7e71b964136bbc4c9e01ce. 54 electrons, total spin S = 0. Energies are reported in this file's convention. A different Hamiltonian, active space or spin sector is a different test.
The same Hamiltonian, written in a different orbital basis, is the Fe₄S₄ (54 electrons, 36 orbitals) instance on the public Quantum Advantage Tracker's variational track, so its posted energies are directly comparable. Counted executions read the pinned file. Rotating its orbitals within the active space, for example into the tracker's basis, is input processing and counts toward the run; a Hamiltonian regenerated from a different calculation is a different input.
A workflow that runs circuits on quantum hardware and uses their measured outputs to build a compact, classically checkable variational state. It includes input processing, task-specific optimization, circuit preparation, quantum execution, error handling, classical processing, state construction and energy verification. Resource estimates and simulated quantum runs do not count as hardware results.
Every eligible public classical workflow for the same target. Coverage must include GPU-accelerated DMRG, selected configuration interaction, and the hybrid pipeline with its quantum sampling stage replaced by a classical method. Each contender may use its strongest documented implementation within the common budget and need not imitate the quantum algorithm. "Strongest" means strongest among the publicly tested contenders, not optimal over all possible algorithms.
A classical contender counts when it publishes, before the cutoff, the same records for ten cold-start runs on this input: input identifiers, software versions, protocol, hardware identification, raw logs, resource and pricing records, all ten outcomes and a verifier. Its protocol and runs may predate this record's registration. If a required comparator family has no eligible contender by the cutoff, the comparison is inconclusive.
Each complete execution has a ceiling of US$100,000 in compute-access charges and 168 hours of elapsed execution time. Both sides are valued at publicly archived rental prices or published commercial quotations on the comparison date; donated or internal access is valued the same way. Counted: quantum and classical resources, task-specific tuning, retries, data movement, state reconstruction and verification. Not counted: general software development and scientific review labor. External queue delays are reported separately. This is a purchasing-budget ceiling, not a claim of equal physical resources.
Elapsed execution time from the fixed Hamiltonian input to a verified, normalized singlet state whose active-space energy upper bound is at most −327.245436 hartree. That target is the reference energy, −327.247036 hartree, plus 0.0016 hartree. The reference is the lowest published, non-extrapolated variational upper bound for the S = 0 state on the pinned file before this record's registration: today, spin-adapted DMRG with entanglement-minimized orbitals at bond dimension 20,000, spin isomer BS3 (Zhai et al., arXiv 2601.04621, supplementary section 4.1, the same in v1 and v2). The reference is frozen at registration; lower bounds published later enter the comparison and do not move the target. It stays provisional until registration confirms the spin sector and that the paper's file matches the pinned bytes. The 0.0016 hartree margin is a tolerance above a variational bound, not chemical accuracy relative to the exact energy or a claim of experimental accuracy. Any verified, normalized state in the S = 0 space counts, whatever its spin-coupling pattern.
Each workflow performs ten consecutive cold-start executions under a protocol disclosed before they begin. Every attempt is recorded; failed runs are not replaced. Every quantum dossier also includes an ablation: the same workflow, code, execution policy and classical hardware class, with every quantum-hardware output replaced by a classical source declared in the protocol before the counted runs begin. The protocol names each replaced interface and, for each, the encoding, measurement basis, distribution and constraints of the classical source and how every adaptive call is answered; the source is at least uniform over the declared sample space under the same particle-number and spin-projection constraints, and a stronger classical sampler may be declared as well. The ablation runs ten times cold-start under the same ceilings, is recorded the same way, and the cost of generating the classical source counts toward its execution. A complete, eligible ten-run ablation populates the comparator family of the hybrid pipeline with its quantum sampling stage replaced by a classical method, whether or not any of its runs succeeds; its successful executions within both ceilings enter the speed-score comparison as classical contenders. The final singlet verification is unchanged. Each quantum execution must meet the energy target within both ceilings, with a published deterministic error bound of at most 10⁻⁶ hartree on its energy. A dossier is positive when it meets this ten-run requirement and its speed score's conservative lower bound is at least 2.0.
The speed score is the fastest successful classical execution divided by the slowest of the ten quantum executions. Its bounds come only from the timestamp uncertainty stated in the published execution logs; no statistical interval is taken over runs, because the score already uses the fastest classical and the slowest quantum execution. It passes when its conservative lower bound is at least 2.0, fails when its conservative upper bound is below 2.0, and is inconclusive when it spans 2.0. A quantum timeout or budget overrun fails the ten-run requirement. If no classical contender reaches the target, the comparison is inconclusive.
Judgment happens once, at the 2027-12-31 cutoff. A quantum dossier can bind only if it has been complete and public for at least 180 days by then, which means public by 2027-07-04 at 23:59 UTC. Every eligible classical result public before the cutoff enters the comparison, including methods published after the quantum result. A potentially decisive classical result whose accounting is still unresolved at the cutoff makes the comparison inconclusive for each bound dossier it could decide; unresolved accounting that affects one dossier does not veto another.
A quantum dossier is eligible when it publishes immutable input identifiers, software versions, the execution protocol, hardware identification, raw execution logs, resource and pricing records, all ten outcomes, the ten ablation outcomes with their declared classical source, a reproducible verifier, and a team declaration: the team's members, controlling organization and project-specific funding, under the same criteria as the independence rule. Its protocol must carry a public timestamp after this record is registered and before the counted runs begin.
Not eligible: the wrong molecular problem, simulation presented as hardware, missing required records, including the ablation records and the team declaration, or quantum runs begun before the timestamp. Eligible but inconclusive: complete evidence with unresolved bounds, conflicting verification or an unresolved deciding comparison. Missing documentation is never treated as evidence that the quantum method lost.
Each team's earliest eligible quantum dossier binds for that team, including an unsuccessful one, if it is complete and public by 2027-07-04 at 23:59 UTC. The 180 days count from the moment the complete dossier is public. A later, more favorable dossier from the same team does not replace it; within a team, identical public timestamps are ordered by permanent source identifier. At both stages, binding and selection, permanent source identifier strings are compared in ascending UTF-16 code-unit order, without locale collation, and timestamps are compared as instants with every published fractional digit. Groups that share scientific authors, a controlling organization or project-specific funding, judged by the independence rule's criteria and applied through every overlapping group, count as one team, and team declarations are checked before binding. A dossier from one team never prevents another team's dossier from binding.
The test resolves on one bound dossier, chosen by rule: a positive dossier over a non-positive one; among positive dossiers, the one whose own outcome ranks highest, Established over Active debate over awaiting confirmation, then the one with more independent positive confirmations, then the higher conservative lower bound of the speed score, then the earlier public timestamp, identical timestamps ordered by permanent source identifier. With no positive dossier, the outcome is inconclusive if any bound dossier is inconclusive, and threshold-not-met only if every bound dossier failed. Every bound dossier stays on the record with its confirmations, contradictions and selection disposition, and evidence against the shared verifier, Hamiltonian or classical comparison is assessed against every dossier it affects.
An independent confirmation requires newly executed runs, new measurement data, independent analysis and an independently reproduced energy check. The confirming group shares no scientific authors, controlling organization or project-specific funding with the original team or another counted confirming group, and declares its affiliations, funding and competing interests. A common general-purpose funder or a commercially available hardware provider does not by itself break independence; shared hardware, software and provider help must be disclosed. Reviewing the original data or serving on a prize panel is an audit, not an independent execution.
A positive confirmation is a new execution of a bound workflow on the same input, by a group meeting the independence rule and the same eligibility rules, whose speed score has a conservative lower bound of at least 2.0. A decisive contradictory reproduction is a new execution of a bound workflow on the same input, by a group meeting the same rules, whose speed score has a conservative upper bound below 2.0, or that fails the ten-run requirement. Both run on the same hardware class as the dossier they confirm or contradict, or one its authors name as equivalent. Both count if their complete dossiers are public by the cutoff, and both are compared against the same classical results. Confirmations and contradictory reproductions are counted per bound dossier; the outcome is applied to the dossier the binding rule selects.
Research teams run and publish the executions. No team is represented as having agreed to take part. Violema keeps the record and applies the published rule; numerical checks must be reproducible from the public dossier by evaluators independent of the reporting team.
Each bound quantum dossier must be complete and public by 2027-07-04 at 23:59 UTC. Confirmations and contradictory reproductions must be complete and public by the cutoff, 2027-12-31 at 23:59 UTC, when the outcome is applied once. With no bound dossier, the test lapses: no eligible result, never a negative one. Any other quantum claim made public after 2027-07-04, and any evidence after the cutoff, go into a new dated assessment, and the frozen outcome stays visible.
The molecular input and the reference energy come from published work. The rule that selects the reference, the energy margin, budget, run count, speed margin, 180-day window, deadline, independence rules, team binding, the ablation requirement and tier moves are Violema's policy choices, not field consensus.
Applied once, at the cutoff, after every admissible comparison is in. The three positive outcomes apply to the resolving dossier, the bound dossier the binding rule selects; the other three apply to the bound dossiers together. The outcomes do not overlap. Each carries a rule id, so a result can cite the outcome it triggered.
The resolving dossier is positive, it has at least two independent positive confirmations, and no decisive contradictory reproduction of it exists.
Moves to Established, for this benchmark only o-established-benchmarkThe resolving dossier is positive, and at least one decisive contradictory reproduction of it exists.
Moves to Active debate, for this benchmark only o-contestedThe resolving dossier is positive, with fewer than two independent positive confirmations and no decisive contradictory reproduction of it. The positive result is shown on the record with its confirmation count.
Stays Speculation o-positive-awaiting-confirmationEvery bound dossier fails: its speed score's conservative upper bound is below 2.0, or it fails the ten-run requirement.
Stays Speculation; threshold not shown o-threshold-not-metNo bound dossier is positive, and at least one is eligible but inconclusive: for example, a speed score whose bounds span 2.0, a required comparator family with no eligible contender, or unresolved accounting.
Stays Speculation; inconclusive o-otherwiseNo dossier is bound: none from any team was complete, eligible and public by 2027-07-04.
Stays Speculation; lapsed o-lapsedA positive result here would support funding an independent reproduction, and then a benchmark closer to a biological decision. It would not by itself support claims about drug efficacy, patient outcomes or broad biological advantage. Three indicators help a group decide where to put effort.
Where hybrid methods show gains first: open-shell metal clusters and excited states, or closed-shell energies where classical methods are already strong.
Changes which projects get quantum time.
The gap between a hybrid workflow and the same pipeline with its quantum sampling replaced by a classical method. In 2025 a classical method matched one quantum iron-sulfur result outright. On the other side, the Algorithmiq, IBM and Cleveland Clinic team describes a quantum-assisted DMRG workflow for a photosensitizer; whether it beats DMRG at equal total resources is the open question.
Changes whether to scale quantum runtime or invest in the classical steps doing the work.
How long classical methods take to answer each new hardware claim. In 2023 the answer came in under two weeks.
Changes how much weight a fresh claim deserves before its challenge window closes.
These are proposed indicators for a reader's own decisions, not findings about any group's plans.
Cited work comes from IBM, RIKEN, Algorithmiq, Cleveland Clinic and university groups. IBM and Algorithmiq also sell the hardware and software involved. These interests are disclosed as context, not as grounds to exclude any work, and the same eligibility, comparison and disclosure rules apply to every team.
Violema's own interest: it is a company that builds evidence records, and this record is a demonstration of that work.
Feedback from any institution whose work is cited is stakeholder review. It does not count as independent confirmation or as endorsement of this record. No individual's participation is inferred from an institution's name.
Dates are each work's first public version (preprint or announcement), not its journal year. Where a later version carries the figure quoted, that is stated. The cost estimates are point values under each paper's own hardware assumptions and measure different things, so they are listed rather than charted. Each entry is an observation; in a full map it would be a claim with its own tier.
A fault-tolerant resource estimate for simulating FeMoco, the iron-molybdenum cofactor of nitrogenase, with error-correction overhead included. An estimate, not a run or a classical comparison.
Fault-tolerant estimateFound that the FeMoco active space used in the 2016 estimate gives a qualitatively incorrect ground state, and built a larger model with 113 electrons in 76 orbitals.
Classical resultNew algorithms for chemistry on a fault-tolerant machine. The version of 2019-10-01 estimates about 700 times less surface-code spacetime volume for FeMoco than earlier quantum algorithms, using the larger 2018 model; the first version gave no such factor.
Fault-tolerant estimateA projected cost for FeMoco: about 4 million physical qubits and under 4 days, assuming 1 μs cycle times and physical gate error rates no worse than 0.1%. An estimate, not a run.
Fault-tolerant estimateAn analysis of whether there is evidence for exponential quantum advantage in ground-state energy estimation. It concludes that such evidence has yet to be found across chemical space. It does not consider other tasks and does not rule out polynomial speedups.
AnalysisA 127-qubit experiment that IBM presented as evidence of the utility of quantum computing before fault tolerance. It concerns a spin model, not a molecule.
Hardware runClassical simulations of that experiment, posted within two weeks of it. This is why the test makes a quantum result wait at least 180 days before it is judged, and counts every classical answer public by the cutoff.
Classical resultSample-based quantum diagonalization (SQD) on quantum hardware, applied to nitrogen and iron-sulfur clusters. Going beyond exact diagonalization is not the same as beating the strongest approximate classical solver.
Hardware runA fault-tolerant estimate on cytochrome P450, an enzyme family central to drug metabolism: about two orders of magnitude faster (233 times for P450) than an earlier algorithm on baseline photonic hardware without Active Volume support. Active Volume compilation, which suits only some hardware, contributes a factor of about 25 of that. An erratum published 2026-10-08 corrects circuit diagrams and says the correction does not change the resource estimates.
Fault-tolerant estimateA critical analysis of quantum-selected configuration interaction, the idea behind SQD, finding that sampling efficiency and determinant selection limit it in the systems studied.
AnalysisClosed-loop sample-based quantum diagonalization on IBM Heron processors with up to 152,064 nodes of the Fugaku supercomputer. For this [4Fe-4S] (54 electrons, 36 orbitals) model it used 77 qubits and a 1010-determinant subspace, reaching −326.912 hartree, 278 mEh lower than the team's 2024 result. The authors describe its accuracy as comparable to some all-classical approximation methods. This is the result behind the tracker's SQD entry below.
Hardware runThe tracker lists this record's Fe₄S₄ model (the same Hamiltonian in a different orbital basis). An IBM and RIKEN SQD entry, posted 2025-11-11, reports −326.913 hartree on 77 qubits, with 10 hours of quantum processor time and 10 hours on 76,032 Fugaku nodes; the submission and its preprint describe the two phases as overlapping. The tracker files this instance under its baseline benchmarks, with no active advantage candidate.
Program or otherA classical method seeded with random determinants (TrimCI) matched a recent quantum SQD energy for a [4Fe-4S] cluster with roughly a million times fewer determinants and CPU-hours.
Classical resultThe TrimCI group posts classical entries on the same tracker model, matching the SQD energy (−326.913 hartree) and reaching −327.175 hartree.
Classical resultA benchmark of SQD across the W4-11 thermochemistry set, which found uneven accuracy across reaction types.
AnalysisA classical simulation of the FeMo-cofactor model to chemical accuracy, from the Chan group. Its benchmark uses this record's pinned [Fe₄S₄(SCH₃)₄]²⁻ (54 electrons, 36 orbitals) file and reports spin-adapted DMRG upper bounds down to −327.247036 hartree for the lowest of three spin isomers, with entanglement-minimized orbitals at bond dimension 20,000. That bound is this record's reference energy. Legeza et al. compare their result only with Zhai's values for one isomer, −327.244001 and an extrapolated −327.245273.
Classical resultGPU-accelerated DMRG on the Fe₄S₄ CAS(54,36) model reached −327.24466 hartree at bond dimension 12,288, in about 12.6 hours on one DGX B200 node: 2.38 mEh above this record's reference. The paper cites the model of Sharma et al. 2014 and does not name an integral file; that it used the pinned Hamiltonian is inferred from agreement with results on the pinned file, not stated by the authors.
Classical resultThe announcement names no winner of the $5 million grand prize for quantum advantage. A team led by Algorithmiq, with IBM and Cleveland Clinic, won the $2 million prize for running its photosensitizer workflow on more than 50 qubits at program depth of order 10³ to 10⁴, with a path toward advantage. Wellcome Leap expects a follow-on program.
Program outcomeA Cleveland Clinic, RIKEN and IBM team ran a fragmented hybrid quantum-classical calculation on protein systems of up to 12,635 atoms, using IBM Heron processors with the Fugaku and Miyabi-G supercomputers. The largest circuit used 94 qubits and 5,758 two-qubit gates. The authors state that they do not regard the results as the basis of a quantum advantage claim.
Hardware runA fast classical evaluation of single-layer UCJ circuits, a circuit family used in SQD workflows.
Classical resultAdvantage claims on spin models and sampling tasks. They concern problems other than biological molecules, so they do not bear on this test.
Other problemsNo outside timestamp yet, so nothing here counts toward a track record. Map time is when the record changed its mind; evidence time is when results appeared.