CQClinical Quantum Exchange

Evidence and receipts

Two verdicts, never one. Mechanism asks whether the circuit did what it was meant to, inside 4·√(0.5/shots). Performance asks whether it beat the classical floor. The second never inherits the first’s pass. Every run here lost or tied against ordinary maths, and that is published at full size.

Runs were executed offline against Quantinuum Nexus and committed at source revision d8993ce. Emulator is labelled emulator. Nothing on this page ran on a QPU.

Each receipt digest is signed with SLH-DSA, a NIST-standardised post-quantum signature scheme that Arc supports, and verified before anything is anchored or paid. The Arc transaction carrying the anchor is still ECDSA-signed — why that gap matters.

Chat is not the evidence layer

operating rule
  • — A message in a chat window is a convenience layer. A number only counts once it resolves to a committed receipt carrying its engine, shot count and seed.
  • — A job triggered from a phone still runs the full order: classical floor, dequantization gate, quantum leg, graded receipt, post-quantum seal, anchor, settlement. The channel changes; the gate does not.
  • — Bulk sweeps and gating stay on the desktop runners. The mobile channel triggers small jobs and reads receipts back — nothing more.

Noisy-tier bands, committed first

grading rule

A noiseless simulation says whether the construction is sound. A noisy emulator says whether it survives. The bands below are fixed before the run, so a result cannot be re-banded after the fact, and a run that lands outside its band is published rather than repeated.

  • sI-PASSInside the pre-committed tolerance at every probe.
  • sII-DEGRADEDThe signal's direction survives the noise; its magnitude does not.
  • sIII-FAILOutside the degraded bar at one or more probes.
sI-PASSNOISY-EMUL · H2-Emulator

Parity-coded cohort sampler held under emulator noise

Seven data qubits plus an [8,7,2] parity-check ancilla, 8,192 shots, post-selected on a zero syndrome.

4.8% of raw shots were error-detected and discarded. The kept cohort matched the target distribution to a Jensen-Shannon divergence of 0.0024, with a maximum per-atom deviation of 0.0034 against a 4σ band of 0.0453.

The error-detection construction works at the noisy tier. This is emulator evidence, not a QPU claim, and it says nothing about advantage.

job id: 514510f1-1083-4ead-ba58-895bb5f44032
sIII-FAILNOISY-EMUL · H2-Emulator

Coherence-sensitive estimator failed its own pre-committed bar

A 17-qubit Hadamard-test moment estimator with no error-detection lane available — a parity measurement would collapse the interference the estimator needs.

Noise suppressed the moment by about 33% at both committed probe angles. One probe landed at 0.260 deviation against a 0.250 degraded bar, so the receipt reads sIII-FAIL. The sign survived; the magnitude did not.

A measured honest negative. It is published at full size with its diagnosis, it pays nothing, and the bar was not moved to rescue it.

job id: 304b0eb8-9f94-4f40-b7d3-80151217746a

Gynaecology — suspected endometriosis

receipt PASS

Of the referrals waiting for a first gynaecology appointment, which ones should a clinician read first?

Mechanism
PASS
measured 0.0231 vs envelope 0.0625
Performance vs classical floor
LOSS
floor 0.742 AUROC · Logistic regression, 5-fold cross-validation
  • 0.701 Logistic regression, plain
  • 0.742 Logistic regression, balanced class weights — the bar
  • 0.738 Tuned RBF kernel, 5-fold grid
Claims
  • — The 14-qubit kernel circuit produced the intended overlap distribution within 4*sqrt(0.5/shots) on the H2 emulator.
  • — The resulting ranking did NOT beat the logistic-regression floor of 0.742 AUROC.
  • — A classical surrogate reproduces the kernel, so the advantage claim dissolves.
engine: Quantinuum Nexus H2-Emulator
backend qualifier: emulator
shots: 2048
seed: 20260910
device: H2-Emulator
job id: nx-emu-endo-2048-a17c
bell control: 0.0039
billed HQC: 11.9
commit: d8993ce
Execution
committed record

Figures committed from an earlier offline run. No live job id is claimed for this pathway.

Dequantization gate

The classical surrogate reproduces the kernel within error, so no advantage claim survives this gate. Reported either way, as the gate requires.

Pre-registration
sI-PASSNOISY-EMUL

Mechanism inside 4*sqrt(0.5/shots) at 2,048 shots; performance must beat the best POWERED member of the classical family, not the plain one.

ref: prereg/endo-triage@d8993ce
  • — A1 — floor family widened to include balanced class weights after the plain baseline was found unpowered. Bar unchanged.
  • — A2 — Bell control moved in-batch so the control shares the job's noise.

Amendments fix the tool, never the target.

Execution spec (pytket · H2-Emulator)

Submission contract only. No credentials, no job id, no result implied by this block.

builder: pytketconfig class: QuantinuumConfigdevice: H2-Emulatorqubits: 14shots: 2048seed: 20260910

verification: Bell |Phi+> control in the same batch; accept when anti-correlated/shots <= 4*sqrt(0.5/shots).

Post-quantum seal

Sign this receipt digest with SLH-DSA and verify it on the spot. The signature is a few seconds of work, so it runs on request here and automatically on every live job.

grade reason: Envelope complete and internally consistent.

Gynaecology — theatre list sequencing

receipt PASS

Given the theatre slots available next month, which ordering of the waiting list clears the most long-waiters?

Mechanism
PASS
measured 0.0288 vs envelope 0.0625
Performance vs classical floor
LOSS
floor 31 of 40 Long-waiters cleared · Greedy longest-wait-first then 2-opt
  • 27 Greedy longest-wait-first
  • 31 Greedy then 2-opt local search — the bar
  • 31 Simulated annealing, 8 ms budget
Claims
  • — The 24-qubit Ising circuit sampled the intended distribution within envelope on the H2 emulator.
  • — Best-of-N sampling cleared 29 long-waiters against a classical floor of 31 — a loss, and a shot-budget artefact rather than an optimiser.
engine: Quantinuum Nexus H2-Emulator
backend qualifier: emulator
shots: 2048
seed: 20260910
device: H2-Emulator
job id: nx-emu-gynae-2048-6b02
bell control: 0.0049
billed HQC: 22.3
commit: d8993ce
Execution
committed record

Figures committed from an earlier offline run. No live job id is claimed for this pathway.

Dequantization gate

Annealing reaches the same objective value in 8 ms. The quantum leg adds no measured advantage on this instance.

Pre-registration
sI-PASSNOISY-EMUL

Objective re-derived host-side from raw bitstrings; best-of-N must clear more long-waiters than greedy + 2-opt.

ref: prereg/gynae-backlog@d8993ce
  • — A1 — shot budget disclosed as a limit on best-of-N sampling before the run, so a loss cannot later be excused as budget.

Amendments fix the tool, never the target.

Execution spec (pytket · H2-Emulator)

Submission contract only. No credentials, no job id, no result implied by this block.

builder: pytketconfig class: QuantinuumConfigdevice: H2-Emulatorqubits: 24shots: 2048seed: 20260910

verification: Bell control in-batch; objective re-derived host-side from raw bitstrings.

Post-quantum seal

Sign this receipt digest with SLH-DSA and verify it on the spot. The signature is a few seconds of work, so it runs on request here and automatically on every live job.

grade reason: Envelope complete and internally consistent.

Dermatology — two-week-wait referrals

receipt PASS

Which two-week-wait skin referrals carry features that most often turn out to need urgent review?

Mechanism
PASS
measured 0.0198 vs envelope 0.0625
Performance vs classical floor
LOSS
floor 0.781 AUROC · Gradient-boosted trees, 5-fold cross-validation
  • 0.744 Logistic regression, balanced
  • 0.781 Gradient-boosted trees, tuned — the bar
  • 0.769 Resampled (SMOTE) + RBF kernel
Claims
  • — The 11-qubit feature map produced the intended kernel matrix within envelope.
  • — The quantum kernel scored 0.769 AUROC against a classical floor of 0.781 — a loss.
engine: Quantinuum Nexus H2-Emulator
backend qualifier: emulator
shots: 2048
seed: 20260910
device: H2-Emulator
job id: nx-emu-derm-2048-c3f1
bell control: 0.0029
billed HQC: 9.1
commit: d8993ce
Execution
committed record

Figures committed from an earlier offline run. No live job id is claimed for this pathway.

Dequantization gate

Reproduced within error at 128, 512 and 2048 shots. Flat metric with paired confidence intervals crossing zero: classically equivalent.

Pre-registration
sI-PASSNOISY-EMUL

Shot ladder 128/512/2048 with paired confidence intervals; a win requires an interval that excludes zero.

ref: prereg/derm-2ww@d8993ce
  • — A1 — paired bootstrap replaced the single-split comparison; a single draw had no statistical power.

Amendments fix the tool, never the target.

Execution spec (pytket · H2-Emulator)

Submission contract only. No credentials, no job id, no result implied by this block.

builder: pytketconfig class: QuantinuumConfigdevice: H2-Emulatorqubits: 11shots: 2048seed: 20260910

verification: Shot ladder 128/512/2048 with paired confidence intervals; Bell control in-batch.

Post-quantum seal

Sign this receipt digest with SLH-DSA and verify it on the spot. The signature is a few seconds of work, so it runs on request here and automatically on every live job.

grade reason: Envelope complete and internally consistent.

MSK physiotherapy — did-not-attend risk

receipt PASS

Which booked physiotherapy appointments are most likely to go unused, so the slot can be offered again?

Mechanism
PASS
measured 0.0215 vs envelope 0.0884
Performance vs classical floor
TIE
floor 0.5 AUROC · Logistic regression with balanced class weights
  • 0.612 Logistic regression, plain — unpowered, not a bar
  • 0.688 Logistic regression with class weights — the bar
  • 0.674 Resampled (SMOTE) + RBF kernel
Claims
  • — The 9-qubit angle-encoded map produced the intended distribution within envelope on a real Nexus job (69 programs, 1024 shots, H2 emulator).
  • — Quantum kernel scored 1 AUROC against a 0.5 floor on a 4-record held-out split — far too few records to resolve a win, so the pre-committed tie rule applies.
  • — This is emulator evidence. No hardware ran, and no advantage is claimed.
engine: Quantinuum Nexus H2-Emulator
backend qualifier: emulator
shots: 1024
seed: 20260910
device: H2-Emulator
job id: 3597614a-d89f-4634-8e8b-b1e649a44be0
bell control: 0.003906
billed HQC: 0
commit: d8993ce
Execution
executed on Nexus

Submitted to Quantinuum Nexus from account Arun Nadarasa on 2026-09-13. The job id above is live and resolvable. Emulator, not a QPU.

Dequantization gate

Surrogate matches within error. No advantage claim available on this pathway.

Pre-registration
sI-PASSNOISY-EMUL

Paired bootstrap against the balanced-weight floor; overlap of intervals reads TIE, never WIN.

ref: prereg/msk-physio@d8993ce
  • — A1 — tie rule stated explicitly before the run rather than inferred afterwards.

Amendments fix the tool, never the target.

Execution spec (pytket · H2-Emulator)

Executed. Nexus job id on the receipt; the cohort is synthetic and seeded.

builder: pytketconfig class: QuantinuumConfigdevice: H2-Emulatorqubits: 9shots: 1024seed: 20260910

verification: Bell control in-batch, bit order calibrated by X-probe, self-overlap diagonal as the mechanism check.

Post-quantum seal

Sign this receipt digest with SLH-DSA and verify it on the spot. The signature is a few seconds of work, so it runs on request here and automatically on every live job.

grade reason: Envelope complete and internally consistent.

Cardiology — echocardiogram demand

receipt PASS

How many echocardiogram slots will the service need each week over the next quarter?

Mechanism
PASS
measured 0.0255 vs envelope 0.0625
Performance vs classical floor
LOSS
floor 9.4% MAPE · Seasonal ARIMA
  • 14.8 Seasonal naive
  • 9.4 Seasonal ARIMA — the bar
  • 9.5 Kernel ridge on lag windows
Claims
  • — The 16-qubit lag-window kernel produced the intended distribution within envelope.
  • — Forecast MAPE was 10.1% against a classical floor of 9.4% — a loss.
engine: Quantinuum Nexus H2-Emulator
backend qualifier: emulator
shots: 2048
seed: 20260910
device: H2-Emulator
job id: nx-emu-echo-2048-2a70
bell control: 0.0044
billed HQC: 13.8
commit: d8993ce
Execution
committed record

Figures committed from an earlier offline run. No live job id is claimed for this pathway.

Dequantization gate

Surrogate reaches 9.5% MAPE. The quantum kernel is not doing anything the classical one cannot.

Pre-registration
sI-PASSNOISY-EMUL

MAPE on a held-out final quarter; the seasonal-ARIMA floor is fixed before the quantum leg compiles.

ref: prereg/cardio-echo@d8993ce
  • — A1 — held-out window fixed to the final quarter to stop retrospective window choice.

Amendments fix the tool, never the target.

Execution spec (pytket · H2-Emulator)

Submission contract only. No credentials, no job id, no result implied by this block.

builder: pytketconfig class: QuantinuumConfigdevice: H2-Emulatorqubits: 16shots: 2048seed: 20260910

verification: Bell control in-batch; model-free curve check alongside the fitted forecast.

Post-quantum seal

Sign this receipt digest with SLH-DSA and verify it on the spot. The signature is a few seconds of work, so it runs on request here and automatically on every live job.

grade reason: Envelope complete and internally consistent.

Endoscopy — multi-site slot allocation

receipt GAP

Across four sites, how should endoscopy slots be allocated so no site holds a disproportionate backlog?

Mechanism
BLOCKED
no measurement exists
Performance vs classical floor
NOT-RUN
floor 4 patients Max site imbalance · Integer programme, exact solve
  • 11 Proportional allocation
  • 4 Integer programme, exact solve — the bar
Claims
  • — No quantum measurement exists for this pathway. The 32-qubit register exceeds the account's 26-qubit emulator ceiling.
engine: Quantinuum Nexus
backend qualifier: not-run
shots: —
seed: —
device: —
job id: —
bell control: —
billed HQC: —
commit: d8993ce
Execution
committed record

Figures committed from an earlier offline run. No live job id is claimed for this pathway.

Dequantization gate

The gate was never reached because no quantum lane exists at this width.

Pre-registration
not-run

Register width audited against the account ceiling BEFORE submission; an over-width circuit is never run.

ref: prereg/endoscopy-slots@d8993ce
  • — A1 — width audit promoted to a standing pre-flight gate after this lane was found over the ceiling.

Amendments fix the tool, never the target.

Execution spec (hugr · Helios-1E-lite)

Submission contract only, and one that cannot currently be executed. Recorded so the blocked leg is legible rather than absent.

builder: hugrconfig class: HeliosConfig + HeliosEmulatorConfigdevice: Helios-1E-litequbits: 32shots: 2048seed: 20260910

verification: Would need a device wider than the account ceiling. Not submitted.

Post-quantum seal

Sign this receipt digest with SLH-DSA and verify it on the spot. The signature is a few seconds of work, so it runs on request here and automatically on every live job.

grade reason: 32 qubits required, 26-qubit emulator ceiling on this account. Only a compile-only structural audit lane is available: gate counts, depth and per-qubit idle windows.

Mental health — talking-therapy access

receipt PASS

Which referrals to talking therapies are waiting past the access standard, and what is driving it?

Mechanism
PASS
measured 0.0214 vs envelope 0.0625
Performance vs classical floor
LOSS
floor 0.803 AUROC · Random forest, 5-fold cross-validation
  • 0.751 Logistic regression, balanced
  • 0.803 Random forest, tuned — the bar
  • 0.788 Tuned RBF kernel
Claims
  • — The 13-qubit parity-window circuit produced the intended readout within envelope.
  • — Quantum kernel scored 0.784 AUROC against a floor of 0.803 — a loss.
engine: Quantinuum Nexus H2-Emulator
backend qualifier: emulator
shots: 2048
seed: 20260910
device: H2-Emulator
job id: nx-emu-mh-2048-77b5
bell control: 0.0034
billed HQC: 10.7
commit: d8993ce
Execution
committed record

Figures committed from an earlier offline run. No live job id is claimed for this pathway.

Dequantization gate

Reproduced within error. Recorded as a negative closure on this formulation.

Pre-registration
sI-PASSNOISY-EMUL

Parity-window cadence checked against the compiled circuit; performance judged against the tuned random forest.

ref: prereg/mh-capacity@d8993ce
  • — A1 — cadence check added pre-submit after a compile-order defect was caught locally.

Amendments fix the tool, never the target.

Execution spec (pytket · H2-Emulator)

Submission contract only. No credentials, no job id, no result implied by this block.

builder: pytketconfig class: QuantinuumConfigdevice: H2-Emulatorqubits: 13shots: 2048seed: 20260910

verification: Bell control in-batch; parity-window cadence checked against the compiled circuit.

Post-quantum seal

Sign this receipt digest with SLH-DSA and verify it on the spot. The signature is a few seconds of work, so it runs on request here and automatically on every live job.

grade reason: Envelope complete and internally consistent.

Multimodal triage — rare presenting subgroup

receipt GAP

In a mixed symptom, genomic and imaging record set, can the small subgroup that needs a different waiting-list route be picked out at all?

Mechanism
BLOCKED
no measurement exists
Performance vs classical floor
UNPOWERED-FLOOR
floor 0.76 Minority recall · Powered classical family: balanced linear, SMOTE + RBF, tuned RBF
  • 0 Linear SVM, plain — unpowered, not a bar
  • 0.76 Linear SVM, balanced class weights — the bar
  • 0.3 Resampled (SMOTE) + RBF kernel
  • 0.26 Tuned RBF kernel
Claims
  • — No quantum measurement exists for this pathway. The cohort failed its fitness sweep before any budget was released.
  • — An earlier train-side result on this cohort was measured against an unpowered plain linear baseline and is withdrawn as a performance claim.
engine: Quantinuum Nexus
backend qualifier: not-run
shots: —
seed: —
device: —
job id: —
bell control: —
billed HQC: —
commit: d8993ce
Execution
committed record

Figures committed from an earlier offline run. No live job id is claimed for this pathway.

Cohort fitness gate
unfit-cohort

Sample sizes from 320 to 10,240 were swept, then a 20,000-record oracle fit as an upper bound on what any classifier could recover. The ceiling reached 0.40 recall — below the bar, and below the balanced floor at 320 records. The signal is weak, not the data scarce. 'More data' is falsified; the cohort is not fit for a quantum receipt at any size, so no quantum budget was released.

bar 0.80 minority recall on held-out records · classical ceiling 0.4 at 20,000 records
Dequantization gate

The gate was never reached. The cohort-fitness sweep stopped the lane first: a 20,000-record classical oracle ceiling of 0.40 recall sits below the 0.80 bar and below the balanced floor at 320 records, so the signal is weak rather than the data scarce.

Pre-registration
not-run

0.80 minority recall on held-out records. Cohort fitness is swept classically FIRST; a ceiling below the bar stops the lane.

ref: prereg/rare-subgroup@d8993ce
  • — A1 — classical family widened to balanced, resampled and tuned members after the plain linear baseline collapsed at every size.
  • — A2 — a 20,000-record oracle fit added as an upper bound, so 'needs more data' could be tested rather than assumed.

Amendments fix the tool, never the target.

Execution spec (pytket · H2-Emulator)

Submission contract only, withheld by the fitness gate. Recorded so the refusal is legible rather than absent.

builder: pytketconfig class: QuantinuumConfigdevice: H2-Emulatorqubits: 11shots: 8192seed: 20260911

verification: Would need a cohort whose classical ceiling clears the bar. The fitness sweep failed, so the contract was never submitted.

Post-quantum seal

Sign this receipt digest with SLH-DSA and verify it on the spot. The signature is a few seconds of work, so it runs on request here and automatically on every live job.

grade reason: Cohort unfit. Classical ceiling 0.40 minority recall at 20,000 records against a 0.80 bar; no method — quantum, plain, balanced, resampled or tuned — reaches it at any size tried. No quantum budget released.