The Heartbeat Is a Stabilizer: Quantum Computing's Older Cousins

In 1494, a Franciscan friar in Venice named Luca Pacioli published Summa de Arithmetica with a long section on a method Italian merchants had been using for two centuries. The method was double-entry bookkeeping. Every transaction got two entries — a debit on one account, a credit on another — summing to zero. If the books didn’t balance, somebody had made a mistake. The books told you a mistake existed before they told you where.

In December 2024, Google’s Quantum AI team announced in Nature that their Willow chip had pushed a distance-7 surface code below threshold for the first time. One hundred and five physical qubits encoded a single logical qubit, with logical error rate at 0.143% per cycle and exponential error suppression as code distance grew. The thing that surprised the field was not the chip. It was that error-corrected logical qubits had finally outlived their own components. The redundancy paid off.

The mathematical machinery Google used is the same machinery Pacioli wrote up. Anil Arya and John Fellingham, in a working paper at Ohio State, have argued formally what the Venetian merchants had figured out empirically: double-entry bookkeeping is a low-rate error-correcting code. The chart of accounts is a parity-check matrix. The trial balance is a syndrome check. When the syndrome doesn’t reconcile, you know an error happened; you don’t yet know where. The decoder, in the accounting case, is the auditor.

I work on a fleet of long-running agents — processes that run continuously, journal their state, and watch each other for signs of trouble. We did not set out to recapitulate quantum error correction. We did anyway, because the problem shape forced our hand. Once you notice the pattern, it is everywhere. Several of quantum computing’s hardest engineering problems were solved a long time ago in cousin domains the field has not been reading. What follows is four of those cousins, the engineering tricks that port across, and the places the analogies genuinely fail.

The heartbeat is a stabilizer

Ask the dumb question first: how do you tell whether a long-running process is alive without interrupting it?

The naive answer — “just read its state” — fails the moment you try to implement it. Reading working memory either blocks the process or returns a stale view. Tailing the stdout buffer can corrupt or crash the process if the buffer is being written when you read it. So you don’t read the process. The process writes a heartbeat record every cycle; a separate watcher reads the heartbeat; if the heartbeat goes stale, the watcher acts. The heartbeat is a low-bandwidth projection of liveness — correlated with the real state, but never the real state itself.

Quantum error correction does the same thing for the same reason. A logical qubit lives across many physical qubits. You cannot read the data qubits’ amplitude — measurement collapses the superposition. So you introduce stabilizer ancillas: extra qubits whose measurement reports the parity of subsets of data qubits, never the data qubits themselves. The ancilla collapses; the protected state survives. A classical decoder takes the time series of stabilizer outcomes and infers what error happened, then schedules a correction. The Willow result hinges on running dozens of stabilizers in parallel with a decoder fast enough to keep up. IBM’s Quantum Loon, demonstrated in November 2025, ran a Relay-BP decoder on qLDPC codes in under 480 nanoseconds per round; Riverlane’s Local Clustering Decoder, in Nature Communications a month later, hit the equivalent in under one microsecond on FPGA hardware.

Both fields independently arrived at the same engineering rule: when direct observation costs more than you can pay, design redundant projections that are correlated with the real state but cheap to extract. Different physics. Same trick.

Even the failure modes match. A heartbeat record with no cadence model — present or absent only — confuses “process healthy but slow” with “process died and rebooted twice in the gap.” Surface codes have the same problem under the name measurement noise on syndromes: a bad ancilla read looks identical to a real error. Both fields fix it the same way: repeat the measurement, majority-vote across rounds, and require sustained syndrome chains before triggering a correction.

Audit as syndrome extraction

If the heartbeat is a stabilizer, the auditor is a decoder. Move one notch up the abstraction stack and the picture sharpens.

Hindenburg Research dissolved in January 2025, when its founder Nate Anderson published a note saying the firm’s work was done. Over its eight-year run it published reports on dozens of public companies — Nikola settled fraud charges with the SEC; Lordstown went bankrupt; several Adani entities are under investigation; Icahn Enterprises was charged by the SEC in 2024. The method, which Anderson described in the dissolution note, never required access to the targets’ real books. Hindenburg probed public projections: SEC filings, foreign business registries, court records, UCC liens, vessel AIS data, import manifests. Each projection is a low-bandwidth view of the same underlying entity. When the projections did not reconcile, Hindenburg had a syndrome. The short position was the bet that the inconsistency resolved to fraud.

This is structurally a stabilizer ancilla measurement. Each filing is one parity equation over the real state. No single filing tells you the truth; together they over-constrain it. The forensic accountant runs minimum-weight perfect matching by hand: what is the simplest fraud pattern consistent with all the inconsistencies I am seeing?

Even the failure modes match. Hindenburg never broke Tether. The reason, as far as anyone can tell from public reporting, is that Tether kept all its public projections mutually consistent for years. Consistent projections look identical to a clean state. In QEC this is called a correlated error — a fault that simultaneously corrupts both the data and the stabilizers in a way that produces a self-consistent (and wrong) syndrome chain. The code passes; the answer is silently wrong. Surface codes are vulnerable to correlated burst errors for the same reason Hindenburg was vulnerable to Tether.

The transfer that has not been made: quantum verification — proving a quantum server actually ran the computation it claimed — is a hard open problem. Urmila Mahadev’s 2018 protocol and the line of work it started are mathematically beautiful and operationally heavy. The audit profession has been verifying computations performed by hostile counterparties for centuries. Several of its primitives have not crossed over: statistical sampling at calibrated rates, multi-firm independent attestation (the Sarbanes-Oxley rule), append-only journaled provenance with external time anchors. Randomized benchmarking exists in the quantum world but characterizes hardware, not specific computations. The audit cousin would hand you a methodology checklist on day one.

The Goldilocks zone

Quantum hardware engineers will tell you noise is the enemy. Quantum biology researchers will tell you noise, in the right band, is a resource.

Both can be right because they live in different parts of the same curve.

Martin Plenio and Susana Huelga showed in 2008, in New Journal of Physics, that local dephasing noise enhances excitation transport in quantum networks. The mechanism is straightforward in retrospect: pure coherent dynamics traps energy in destructive-interference dark states; too much noise destroys the wave structure entirely; in between sits a regime where the environment knocks the system out of the trap without destroying the transport. The biological systems Nature selects for — the Fenna-Matthews-Olson complex in green sulfur bacteria, the light-harvesting antennas in plants — appear to live in this Goldilocks zone. The 2007 long-lived-coherence headline that started the quantum-biology wave has been substantially walked back; a 2025 Chemical Society Reviews paper by Jha and colleagues put room-temperature electronic coherence at ∼60 femtoseconds, too short to drive picosecond energy transfer. The result that survived peer review was not the headline. It was the noise-assisted transport underneath it.

Joseph Connell published the same shape of curve in 1978 in Science, under the name intermediate disturbance hypothesis. Biodiversity peaks not in undisturbed reserves and not in storm-wracked ones, but in regions experiencing moderate disturbance. (Jeremy Fox argued in Trends in Ecology & Evolution in 2013 that the universal version of IDH should be abandoned — about 60% of empirical studies don’t find the hump-shaped curve. The trade-off mechanism survives even where the universal claim doesn’t.)

The transferable insight is not “QEC engineers should stop fighting noise” — below threshold, fighting noise still pays. The insight is that systems forced to operate in noisy regimes (variational quantum algorithms in the NISQ era) should not assume noise is purely adversarial. An April 2026 preprint (arXiv:2604.23005) extends Plenio-Huelga to show the optimal noise profile is spatially non-uniform: different gates want different amounts. Calibrated dephasing, not uniform isolation, may be the right framing for the hybrid algorithms we’re stuck with until fault tolerance arrives.

Q-day is NotPetya at a different time horizon

In 2019, Craig Gidney and Martin Ekerå estimated it would take roughly 20 million physical qubits to break RSA-2048. By May 2025, Gidney had revised that to under one million. By February 2026, a preprint from the Sydney startup Iceberg Quantum reportedly pushed the estimate below 100,000. Every reduction came from algorithmic improvement, not hardware. The threat surface is moving faster than the hardware roadmap, faster than Moore’s Law in fab improvements would predict, and the trajectory is set by people working on circuit synthesis rather than people working on cryogenics.

Structurally, that is a NotPetya. NotPetya was the canonical correlated catastrophic event on the cyber side: one zero-day in the M.E.Doc tax-software supply chain, roughly $10 billion in global losses, and a years-long fight in which Merck successfully argued its insurers could not invoke a war-exclusion clause to deny the claim — settled in January 2024. The cyber insurance industry rebuilt itself around correlated-loss catastrophes after that. The toolkit it built — outside-in observability, parametric triggers, systemic-event sublimits, attestation-as-underwriting-input — was designed exactly for events like Q-day. It also handles the specific ugliness of harvest-now-decrypt-later: adversaries storing encrypted traffic today to decrypt after a future Q-day event, which means the policy that should have covered the breach may have lapsed by the time discovery happens. Liability insurance solved that latency problem decades ago, under “occurrence vs claims-made” — mesothelioma claims arrive thirty years after asbestos exposure. Same construction ports.

Read each piece against the quantum stack. Outside-in observability: Cloudflare reported in early 2026 that more than two-thirds of human TLS traffic to its edge was using hybrid post-quantum key agreement (X25519MLKEM768) — an underwriter can scan that without the applicant’s cooperation. Parametric trigger: “If a quantum computer capable of breaking RSA-2048 is publicly demonstrated by date X, payout = $Y.” Avoids the hardest part of any quantum-loss claim — proving a specific breach was caused by quantum decryption. Correlated-loss carve-out: every RSA-protected system breaks at once on Q-day; the quantum exclusion is a copy of the war exclusion with the perils relabeled. Attestation: has the applicant migrated their HSMs to ML-DSA? Have they rotated Bitcoin custody off P2PKH addresses? Each is a question with a verifiable on-the-wire answer.

The product that doesn’t yet exist is the parametric quantum exclusion paired with chain-anchored attestation. Insurers in 2026 are adding quantum readiness to renewal questionnaires (Insurance Times tracked this through 2025) but no parametric quantum product is shipping. Per Moody’s, more than 90% of global businesses lack a quantum cybersecurity roadmap, while two-thirds of human web traffic is already running post-quantum primitives — the infrastructure is migrating past the policies covering it. If you want to build the product, the cold-start customer list is small and concrete: specialty Lloyd’s syndicates and the cyber reinsurance desks at Munich Re and Swiss Re. Ten meetings, not a thousand cold pitches; their actuaries already wrote the war-exclusion language Q-day will reuse.

Where the analogies break

None of these cousins are equivalences. Each is a structural mapping that holds along one axis and fails along others. Naming where each breaks is the price of being allowed to use them.

Heartbeats and stabilizers diverge at the substrate. Long-running process failures are mostly independent — one dies, the others continue — while quantum errors are correlated, with leakage events, cosmic ray strikes, and control-electronics crosstalk producing simultaneous faults. The transfer is at the observation-engineering layer, not at the physics.

Audits and stabilizer measurements differ in their degrees of freedom. Auditors can request supplementary documents, interview management, recompute from primary records. Stabilizer measurements get one shot — the qubit is consumed. The trick of probing public projections rather than demanding private state was invented twice, once for each set of constraints.

The Goldilocks zone is much narrower for QEC than for ecology or photosynthesis. Biological systems have evolved over a billion years to recruit specific noise spectra; engineered qubits live in noise spectra determined by fabrication and cryogenics. The lesson is noise can be a resource, not noise always is. Misapplied, this becomes a license to stop fighting noise below threshold, which is wrong.

Q-day is binary; cyber risk is continuous. The cyber toolkit ports if you read it as primitives for catastrophic correlated events, not as steady-state actuarial models.

What you can use today

If you take one thing from this, take the engineering rule, not the analogies.

When direct observation is too expensive — because it collapses the state, interrupts the process, or biases the subject — design redundant projections that are correlated with the real state but cheap to extract. The watchdog reads the heartbeat, not the agent. The auditor reads the trial balance, not the books. The decoder reads the syndrome, not the qubit. The underwriter reads the TLS handshake, not the self-assessment.

Three concrete actions, in increasing effort.

Audit-grade integrity logs for any long-running computation. Whatever observable proxies your system already exposes — heartbeats, queue lengths, parity checksums, build hashes — anchor them to an external timestamp source. Hash a daily integrity log, post the hash to a public blockchain, keep the underlying log private. Now “we ran the system correctly between dates A and B” is cryptographically defensible after the fact. Right primitive for any claim that may be challenged years later, including harvest-now-decrypt-later.

Pre-register your benchmarks before you run the hardware. Behavioral science learned painfully that exploratory analysis on the data that generated the hypothesis is fraud-adjacent; the audit profession learned the same thing as auditor independence. Quantum-advantage claims are vulnerable to the same failure mode — design the benchmark, run the chip, declare victory. Pre-registration is free and catches the mistake.

Read the cousin literature before re-deriving from first principles. Some of these results are formal (Arya & Fellingham). Some are wrapped in domain idiom that obscures the structure (Connell’s intermediate disturbance hypothesis; Plenio & Huelga’s noise-assisted transport). All of them are cheaper to read than to re-discover.

Quantum computing has plenty of genuinely new problems — coherence at the substrate level, fault-tolerant gate synthesis, magic-state distillation. Those deserve the field’s full novelty discount. The rest — observability, the audit trail, correlated-tail risk pricing — should be solved by reading what’s already been written. The cousins have been waiting.

Sources: Google Quantum AI, “Quantum error correction below the surface code threshold,” Nature, December 2024; Arya & Fellingham, “Double-entry bookkeeping as an error-correcting code,” Ohio State working paper; Bravyi et al., IBM Quantum Loon demonstration, November 2025; Riverlane Local Clustering Decoder, Nature Communications, 2025; Anderson, Hindenburg Research dissolution note, January 2025; Plenio & Huelga, New Journal of Physics 10 (2008); Jha et al., Chemical Society Reviews, 2025; Connell, Science 199 (1978); Fox, Trends in Ecology & Evolution, 2013; Gidney & Ekerå, Quantum 5 (2021); Gidney, “How to factor 2048-bit RSA integers with less than a million noisy qubits,” arXiv preprint, 2025; Mahadev, FOCS 2018, “Classical verification of quantum computations”; Cloudflare Radar post-quantum adoption report, Q1 2026; Moody’s, “Quantum cyber risk readiness,” 2026.