How to read a quantum claim

Synthetic educational example

A chosen white-frequency-noise law illustrates averaging; it does not calculate a clock’s accuracy.

scale = A / √t; A = 10⁻¹² at one second

Assumptions

• One synthetic white-frequency-noise law over the whole interval.
• No systematic offset, drift or change in correlation.
• t is in seconds; A supplies the stated one-second fractional scale.

Increasing time 100-fold divides this model’s fluctuation scale by ten. An omitted fixed bias remains omitted.

Change one thing: Double A, then compare 100 seconds with 10,000 seconds. Name an effect that prevents indefinite extrapolation.

No real clock specification, nuclear-clock fit or hardware-control advice.

• NIST: Handbook of Frequency Stability AnalysisHistorical handbook: averaging-time dependence, noise models, drift and measurement conditions; no nuclear-clock performance inferred. Sources checked 2026-10-11.

Synthetic educational example

Two independent equal-probability links illustrate coordination, not an actual network.

No storage: 1/p² rounds. Perfect memory: 2/p − 1/(2p − p²) rounds.

Assumptions

• Independent links and rounds with equal p > 0.
• Perfect storage retains the first success indefinitely at no cost.
• Every joint success is usable by definition; quality is not modelled.

Waiting for both successes in one round is different from retaining the first. Real memory decay, latency and control costs must be restored for a device assessment.

Change one thing: At p = 0.1, list three omitted effects that could reduce the apparent memory benefit.

Rounds are not seconds; these are bounding toy assumptions, not IonQ measurements.

• Architectural Principles for a Quantum InternetInformational architectural background; neither a mandatory standard nor a current performance survey. Sources checked 2026-10-11.
• Quantum Networking at the Speed of Quantum ComputationOctober 7 preprint: one heterogeneous ion–diamond memory link; no general application advantage. Sources checked 2026-10-11.

Synthetic educational example

Select the fastest imaginary setting meeting an invented minimum fidelity of 0.94.

Eligible if F ≥ 0.94; choose the greatest rate among eligible rows.

Assumptions

• All rows describe the same invented event and metric.
• The task’s minimum is stipulated externally.
• No real application threshold or architecture ranking.

B wins only this rule. At a minimum of 0.98, C is the only eligible setting; at 1.0, none is eligible.

Change one thing: Compare minima 0.75, 0.94, 0.98 and 1.0. Explain why no eligible setting is not a score of zero for a technology.

Do not multiply rate by fidelity to invent useful qubits per second.

• Quantum Networking at the Speed of Quantum ComputationOctober 7 preprint: one heterogeneous ion–diamond memory link; no general application advantage. Sources checked 2026-10-11.
• Architectural Principles for a Quantum InternetInformational architectural background; neither a mandatory standard nor a current performance survey. Sources checked 2026-10-11.

Synthetic educational example

An idealized three-stage route demonstrates multiplicative loss accounting.

Combined survival = t₁ × t₂ × t₃

Assumptions

• Transmission factors apply to the relevant surviving events.
• Independent multiplicative loss factors.
• No retries, correlated loss, false counts or recovery.

A component’s 90% transmission does not mean a three-stage route transmits 90%.

Change one thing: Improve only the middle 0.90 stage to 0.99: the product is 0.8019. Compare with improving all stages to 0.95.

Not a photonic-computer success probability or a manufacturer specification.

• NIST: superconductive nanowire single-photon detectorsMechanism and detector tradeoffs; NIST examples are not manufacturer specifications. Sources checked 2026-10-11.

Synthetic educational example

Four fictional instrument settings illustrate Pareto screening.

Prefer signal ↑, noise ↓, stability ↑. Dominated means no better in any goal and worse in at least one.

Assumptions

• Dimensionless invented indices with the stated directions.
• No uncertainty, safety limit or hidden cost.
• Nondominated does not mean safe or optimal.

C dominates A; B trades higher signal for worse noise and stability. B and C are nondominated.

Change one thing: Impose fictional noise ≤ 3 and stability ≥ 4. Which settings remain, and which is then dominated?

No physical tuning recommendation, PNNL dataset or claimed AQUA-WOLF result.

• PNNL: an AI agent for amplifier tuningOctober 5 project announcement: gain-first rewards, later noise/stability objectives and future hardware validation. Sources checked 2026-10-11.

Synthetic educational example

An Amdahl-style calculation counts the unchanged work and additional overhead.

Total speedup = 1 / (1 − f + f/s + o)

Assumptions

• Same complete task and output quality before and after.
• Only f is accelerated; 0 ≤ f ≤ 1 and s > 0.
• o is nonnegative extra time divided by the whole baseline time.

A spectacular local improvement can have a small complete-job effect. A result below one means the synthetic complete job became slower.

Change one thing: Use f = 0.1, s = 10 and o = 0.15: total speedup is about 0.9434. Identify the missing costs in a real comparison.

No company-specific timing, energy, cost or predicted quantum advantage.

• Quobly and SiPearl: exploratory CPU–QPU integrationNon-exclusive MoU and proposed feasibility work, not a delivered joint machine or benchmark. Sources checked 2026-10-11.
• Aaronson: Quantum Machine Learning Algorithms—Read the Fine Print2015 input, access and output assumptions; not a claim that all quantum algorithms fail. Sources checked 2026-10-11.

Synthetic educational example

Recover a repeated classical bit only when every visible entry agrees.

Return the agreed visible bit; otherwise no unique answer.

Assumptions

• Classical repeated-bit word, not quantum encoding.
• Perfectly known erased locations.
• No majority-vote substitution or optimal decoder.

A location flag helps identify the puzzle. It does not create information or resolve a conflict by itself.

Change one thing: Compare 1, blank, 1 with 1, 0, 1. Which extra assumption would majority voting introduce?

This analogy does not permit copying an unknown qubit or calculate a quantum threshold.

• Quantum Error Correction for Quantum MemoriesResearch review: syndromes, Pauli frames, memory and computing deadlines. Sources checked 2026-10-11.

Synthetic educational example

Compare ideal channels against each original pure input, using phase-flip probability 0.18.

Fidelity to original input; phase-flip channel E(ρ) = (1 − p)ρ + pZρZ.

Assumptions

• Four specified pure inputs and ideal mathematical channels.
• Fidelity is to the original input, not the reset target.
• No encoded logical state, imperfect gates or readout noise.

Reset can prepare |0⟩ perfectly while failing unknown-state preservation. The target determines whether an operation succeeds.

Change one thing: Average each column over these four equally weighted inputs. Explain why that average is not a universal device-quality score.

Not a quantum-memory benchmark and not a new Noise Lab reset simulation.

• IBM Quantum Learning: channel representationsMathematical channel definitions, separate from real hardware characterization. Sources checked 2026-10-11.
• Rutgers: adaptive feedback and reset experimentOctober 9 institutional account of repeated measurement and reset; not protection of arbitrary encoded information. Sources checked 2026-10-11.

Synthetic educational example

Independent Bernoulli trials with known p = 0.18 separate count and fraction variation.

Mean count = Np; count SD = √(Np(1 − p)); fraction SD = √(p(1 − p)/N).

Assumptions

• Independent identically distributed trials with fixed known p.
• No bias, drift or correlations.
• Count and fraction have different units.

Count variation grows with N while fraction variation shrinks. These model standard deviations are not a guarantee for every sample.

Change one thing: Run the native lab at p = 0.18, 200 shots and seed 42. Compare 33 observed ones with the expected 36.

No claim that real-device shots meet these assumptions.

• NIST: confidence intervals for proportionsWilson interval and independent-binomial sampling model; no complete quantum-device uncertainty budget. Sources checked 2026-10-11.

Synthetic educational example

These fixtures were checked against the existing native sampler, without changing it.

Update before emission; unsigned 32-bit state; shifts 13 left, 17 right, 5 left; divide by 2³²; compare u < p.

Assumptions

• phase-flip-hzh-v1 and xorshift32-v1.
• A nonzero unsigned 32-bit seed; each shot consumes one draw.
• Software pseudorandomness, not secure or physical quantum randomness.

A seed alone is not the recipe. Algorithm, versions, inputs, comparison rule and shot count must agree.

Change one thing: Change only the seed to 43, then change only the shot count. Which change tests replay and which explores sample variation?

Parity validates this software convention, not a real quantum device.

• IBM Quantum Learning: states and measurementFoundational course; no cloud job or device experiment performed. Sources checked 2026-10-11.

Synthetic educational example

An unreworked all-pass toy illustrates why a component result is not a complete production result.

All-pass fraction = yᵏ

Assumptions

• Identical independent pass probabilities.
• Every required component must pass.
• No screening, replacement, repair, redundancy or correlations.

Repeatability and assembly need their own evidence. Testing and replacing parts can substantially change real outcomes.

Change one thing: Compare 0.99¹⁰⁰ with 0.999¹⁰⁰, then name three ways real manufacturing changes the model.

No actual foundry yield, cost, revenue or investment recommendation.

• NIST: superconductive nanowire single-photon detectorsMechanism and detector tradeoffs; NIST examples are not manufacturer specifications. Sources checked 2026-10-11.

Synthetic educational example

A synthetic binary reporting model separates the underlying event from the instrument’s report.

q = (1 − e₁₀)p + e₀₁(1 − p)

Assumptions

• Two conditional reporting probabilities under a limited binary model.
• Independent repeats only when discussing sampling concentration.
• No measured processor or complete quantum-noise model.

More independent observations concentrate around 0.714, not 0.8. A small sampling interval can miss a systematic discrepancy.

Change one thing: At p = 0, q equals e₀₁. At p = 1, q equals 1 − e₁₀. Explain what each test checks.

No corrected-truth display or silently clipped inverse estimate.

• NIST: confidence intervals for proportionsWilson interval and independent-binomial sampling model; no complete quantum-device uncertainty budget. Sources checked 2026-10-11.
• A Quantum Engineer’s Guide to Superconducting QubitsReview first submitted April 13, 2019; version 5 revised July 7, 2021. Historical superconducting control, noise and readout; no present-day product specification. Sources checked 2026-10-11.

Synthetic educational example

A classical single-worker schedule illustrates waiting and latency.

startᵢ = max(arrivalᵢ, finishᵢ₋₁); finishᵢ = startᵢ + serviceᵢ; latencyᵢ = finishᵢ − arrivalᵢ.

Assumptions

• One worker and the declared request order.
• Nondecreasing arrivals and nonnegative finite service times.
• Arbitrary units; no measured decoder or quantum cycle.

The burst makes later requests wait although the worker handles every request in two units. A mean service rate does not describe every latency.

Change one thing: Choose a synthetic latency deadline of four units, inclusive. Request 3 misses it. A toy deadline miss is not a quantum logical error.

Real systems can use parallelism, batching and classical frame tracking; their operation-dependent deadlines need source-specific evidence.

• Quantum Error Correction for Quantum MemoriesResearch review: syndromes, Pauli frames, memory and computing deadlines. Sources checked 2026-10-11.

Synthetic educational example

Two exact one-qubit examples separate amplitudes, preparation and measurement.

Normalize (1, ±2) by √5; use H = (1/√2)[[1,1],[1,−1]].

Assumptions

• Real two-component pure states in the stated computational basis.
• Exact ideal Hadamard; probabilities are squared magnitudes.
• No arbitrary large-state input or reconstruction.

The same initial measurement probabilities do not make the states identical. Changing the measurement procedure exposes a different property.

Change one thing: Explain why one computational-basis shot cannot reveal either entire amplitude vector.

Not an algorithmic advantage or free storage/readout of exponentially many classical answers.

• IBM Quantum Learning: states and measurementFoundational course; no cloud job or device experiment performed. Sources checked 2026-10-11.
• Aaronson: Quantum Machine Learning Algorithms—Read the Fine Print2015 input, access and output assumptions; not a claim that all quantum algorithms fail. Sources checked 2026-10-11.

Synthetic worked example

How does the stated expression change with word size?

This exercise evaluates the ratio 4b/(b+1) for selected positive integer word sizes. It illustrates a leading-expression comparison associated with the QROM discussion. It does not reproduce the paper’s proof, a full circuit construction or a hardware experiment.

The expression has a definite unit of comparison only within the specified resource model. Calling its result a speedup without saying what was counted would erase the most important assumption.

r = 4b / (b + 1), with positive integer word size b

1. Choose a word size b. Substitute it into both numerator and denominator; keep the +1 term rather than rounding the denominator to b.
2. Compare several sizes. The ratio approaches four from below as b grows, but it does not equal four at every finite size.
3. Return to the source before applying the expression to a real algorithm. Workspace, table shape, lower-order terms and the rest of the computation can change the practical accounting.

For b=32, the ratio is 128/33, approximately 3.878788. This number describes the chosen expression. It is not a measured elapsed-time reduction, an output-fidelity improvement or a financial return.

The useful lesson is to preserve the resource model while explaining the result. A reader can appreciate a strong reduction without being told the whole computer became four times faster.

Try the next question: Evaluate b=1 and b=1000. Explain why the limiting value is not a universal real-workload result.

This is a leading resource-count ratio, not a device benchmark or a stock-performance measure.

• A New Quantum Data-Loading Trick Shrinks One Expensive Bill →

• Halving the cost of QROM again via dense encodingPrimary preprint on a leading Toffoli-cost construction under specified workspace and table-loading assumptions. It does not measure elapsed hardware runtime.

Synthetic worked example

How can known uncertainty change a simple estimate?

Two synthetic independent unbiased observations estimate the same quantity. Their standard deviations are known and fixed. Under those assumptions, inverse-variance weighting gives more influence to the less uncertain observation.

This is a scalar educational example, not a reproduction of the temporal-mode method in the homodyne paper. Real correlated measurement data require an appropriate covariance treatment and independent calibration.

wᵢ = 1 / sᵢ²; estimate = Σ(wᵢxᵢ) / Σwᵢ; SD = √(1 / Σwᵢ)

1. Use observations 10 and 12 with standard deviations 1 and 2. Their inverse-variance weights are 1 and 1/4.
2. Normalize the weights and calculate the estimate. Compute the combined standard deviation from the reciprocal of the total weight.
3. Compare the equal-weight mean. Explain why the comparison depends on unbiasedness, independence and correctly known uncertainty.

The smaller uncertainty does not establish that the detector produced less physical noise. It reflects how the available information was combined under the assumed model.

An unmodelled bias or mistaken uncertainty estimate can undermine the result. Selecting weights after looking for a favourable answer can also create a misleading comparison. Calibration and evaluation roles should remain separate.

Try the next question: Give the first observation a fixed unknown bias. Explain why the lower reported standard deviation does not guarantee a better answer.

The observations and standard deviations are synthetic, not measured voltages from the cited experiment.

• A Better Quantum Measurement Can Start With How You Read the Detector →

• Noise-aware ultrafast homodyne measurementMay 2026 preprint on noise-aware estimation. Its statistical method is distinct from physically removing noise from a device.
• Ultrafast homodyne research journal recordJournal record of the homodyne-estimation work. The scalar exercise above illustrates statistical assumptions; it does not reproduce the experiment.

Synthetic worked example

What changes when two equally noisy readings are correlated?

Two readings have the same standard deviation s and correlation rho. The standard deviation of their average is s sqrt((1+rho)/2). This is a two-variable covariance calculation, not a universal model of every detector.

The example holds the individual uncertainty fixed and changes only correlation. It shows why the word independent matters in an averaging claim.

SD of the average = s × √((1 + ρ) / 2), for two equal-SD readings

1. Set s=1. Evaluate rho=0, 0.5 and 0.9.
2. At zero correlation, averaging reduces the standard deviation by the familiar square-root factor. Positive shared noise leaves more uncertainty in the average.
3. Check the endpoints. At rho=1 the two readings share their variation completely; the average has the same standard deviation as either one. Perfect negative correlation is a special ideal case, not a free experimental resource.

The calculation does not change the physical source of the noise. It changes what follows from the assumed joint statistics.

A real experiment needs evidence for its covariance model and should test whether that model remains stable. A correlation estimated from the same data used to celebrate an improvement may require a separate validation step.

Try the next question: Explain how a common drifting offset can make the independence assumption inappropriate.

The formula covers two specified readings. A longer time series needs its own correlation model.

• Noise-aware ultrafast homodyne measurementMay 2026 preprint on noise-aware estimation. Its statistical method is distinct from physically removing noise from a device.
• Ultrafast homodyne research journal recordJournal record of the homodyne-estimation work. The scalar exercise above illustrates statistical assumptions; it does not reproduce the experiment.

Synthetic worked example

Why do more samples not solve every measurement problem?

Assume independent random errors with single-sample standard deviation s and a fixed bias b. The mean-squared error of the average is b²+s²/n. The model deliberately separates two kinds of error.

The example uses b=0.5 and s=2. These are arbitrary educational numbers, not estimates of a quantum instrument.

MSE = b² + s² / n, with fixed bias b and independent random errors

1. Calculate the squared bias, which remains 0.25 regardless of sample count.
2. Calculate the variance contribution s²/n for n=4, 100 and 10,000.
3. Add the two terms. Explain why the total approaches the squared bias rather than zero.

A narrow uncertainty bar can coexist with a systematic offset. More measurements address the random component under the model; they do not automatically establish correct calibration.

The appropriate next action depends on the source of the bias. The exercise does not prescribe a laboratory procedure, but it makes clear why more data and better accuracy are not identical claims.

Try the next question: Set b=0 and compare the limiting behaviour. Then double b and identify which term changes.

The fixed bias is an illustrative input, not an estimate of bias in a source experiment.

• A Better Quantum Measurement Can Start With How You Read the Detector →

• Noise-aware ultrafast homodyne measurementMay 2026 preprint on noise-aware estimation. Its statistical method is distinct from physically removing noise from a device.
• NIST-developed quantum sensors improve nuclear monitoringNIST account of transition-edge sensing for a specified X-ray measurement. A sensing instrument answers a different question from a quantum computer.

Synthetic worked example

When does the simple collective-probe expression lose its advantage?

The exercise compares normalized factors 1/sqrt(N) and 1/(vN). It is an idealized pedagogical contrast using probe count N and an illustrative visibility factor v. It does not derive a complete metrological bound or reproduce the funded project.

Time, state preparation, losses, repetitions and estimator design are not modelled. Those omissions prevent the table from being a prediction of actual experimental performance.

Independent factor = 1 / √N; collective factor = 1 / (vN), for v > 0

1. Fix N=4. The independent-probe factor is 0.5 in this normalized example.
2. Change v from 1 to 0.75 to 0.5. Watch the collective expression worsen even though the probe count stays unchanged.
3. At v=0.5, the two expressions match. At v=0, this particular expression provides no finite useful factor. Interpret the result as a limitation of the model, not a physical infinity in a detector.

The example explains why an entanglement headline is not enough. A fair practical comparison must state what resources and imperfections were counted.

The funded sensing project and this idealized calculation have different evidence states. These table values are calculated examples, not measurements produced by the project or the pictured equipment.

Try the next question: Include a synthetic preparation-time penalty and explain what additional normalization would be necessary.

The table establishes no dark-matter detection, forecast sensitivity or universal quantum advantage.

• Four Entangled Qubits Could Become a New Kind of Dark-Matter Listening Device →

• Early Career award: dark matter with entangled qubitsInstitutional account of a planned four-probe sensing comparison. A funded programme is not a dark-matter detection or a completed sensitivity result.

Synthetic worked example

How many independent attempts are expected before one success?

For independent identical trials with fixed success probability p, the expected number of attempts to the first success is 1/p. This is an expectation, not a guarantee for any individual sequence.

The calculation excludes adaptive protocols, changing probabilities, correlations and time-dependent preparation costs. It is a useful prompt to ask about discarded attempts, not a complete experiment model.

Expected attempts = 1 / p, for independent identical trials and p > 0

1. Calculate the expectation at p=0.8, 0.2 and 0.01.
2. Explain why five expected attempts at p=0.2 does not mean every fifth trial succeeds.
3. Keep accepted-output quality separate from acceptance probability. A beautiful retained subset does not reveal the cost of obtaining it.

Reporting the attempts helps readers assess the resource cost of postselection. It does not automatically invalidate postselection as a scientific technique.

For a practical workflow, the next accounting step includes the duration and other resources per attempt. If the probability changes during the run, the fixed geometric model should not be applied without modification.

Try the next question: Compare a fast low-success attempt with a slower high-success attempt using explicitly synthetic times.

Published accepted outcomes alone do not establish the success probability or discarded-attempt cost.

• A Quantum Chip Ran a Particle-Collision Model. No Collider Was Built. →

• Particle-collision simulation with W-state preparationHistorical 2025 research on a defined particle-collision model. A circuit representation is not a physical collider or a general application advantage.

Synthetic worked example

What does a late-night Pacific deadline mean elsewhere?

This is a date-conversion exercise using an example input: 23 November 2026 at 23:59 in America/Los_Angeles. It does not announce or verify an event deadline.

An offset depends on the date and location. A missing time zone cannot be guessed; repeated or nonexistent local times require explicit resolution.

Example input: 2026-11-23 23:59, America/Los_Angeles

1. Keep the original example date, time and IANA zone together.
2. Convert the same instant to UTC and America/New_York using date-specific offsets.
3. For a real deadline, confirm the organizer’s current date, time and zone before making plans.

The UTC date is the following day. A deadline with an official time should not be reduced to a bare calendar date.

The local reference check rejects ambiguous and nonexistent local times. That validation is separate from confirming a real event’s deadline.

Try the next question: Why can a deadline with no verified time zone not receive a reliable countdown?

A reliable countdown needs an actual deadline and a verified time zone.

• Python: IANA time-zone supportDocumentation of date-specific zone conversion and ambiguous local times. The table is a locally computed example, not an event-status check.

Synthetic worked example

How does a fictional consideration ceiling differ from an actual payment?

Use a fictional base amount of 100 and a contingent maximum of 20. A fraction f of the conditionally payable amount produces 100+20f. The units are arbitrary and do not describe the IonQ agreement.

The model explains a distinction in arithmetic. It does not estimate the probability of a condition, the fair value of an earnout, the number of shares issued or the tax treatment of a transaction.

Fictional amount = 100 + 20f, for 0 ≤ f ≤ 1

1. Evaluate f=0, 0.5 and 1.
2. Keep the maximum separate from an expected value. An expected value would require a justified distribution and a contractual valuation model.
3. Return to the actual agreement before describing a real transaction. The real contract may use a different mechanism entirely.

The maximum is 120 in the fictional example, but none of the scenarios establishes the actual outcome of a real deal. A headline ceiling should not become a statement that the full amount has already been paid.

An actual transaction has its own contractual conditions and source record. The fictional example illustrates the difference between a ceiling and an outcome; it does not establish what a real agreement will pay.

Try the next question: Explain what additional evidence would be needed to estimate a probability-weighted value.

Fictional arithmetic does not forecast a real transaction outcome or investment return.

• IonQ to acquire Elisity to accelerate zero trust for the quantum eraIssuer announcement of a proposed transaction. A signed agreement, a consideration ceiling and a completed acquisition are different stages.