Part IX. Company and Investing · Chapter 76
Error-Correction-Stack Diligence
Fault tolerance needs decoders, real-time controls, calibration systems, and middleware — and a startup has formed around each. This chapter is about deciding whether any of them owns a bottleneck someone will pay to remove.
In this chapter 9 sections
Trace physical measurements through syndrome extraction, transport, decoding, feedback, and logical outcome; require latency, bandwidth, error-model, code-distance, and logical-error data at every owned interface before assigning technical or economic value.
Error correction is a timed information pipeline. Measurements leave the device, become syndrome bits, cross an acquisition and transport path, enter a decoder, and return as feedback or a tracked frame before the next dependent operation. A fast decoder in isolation can still miss the machine’s cycle contract.
Draw the syndrome path before valuing the product
The interface budget records code family and distance, physical-error model, syndrome bits per cycle, acquisition, transfer, decode and feedback latency, sustained throughput, tail latency, logical-error estimator, and before/after conditions. The total is , with every term in microseconds and with overlap stated separately.
Interface budget from measurement to correction
Shor, Steane, and Gottesman establish the code and stabilizer foundations; GKP establishes an oscillator encoding. The surface-code and fault-tolerance reviews provide synthesis, while the Gidney–Ekerå estimate shows how explicit error-correction assumptions enter an application resource account. None of these validates a company decoder at an unstated interface.
A decoder misses its cycle-time contract
The illustrative decoder takes 0.32 microseconds, which sounds compatible with a one-microsecond cycle. Acquisition, transport, and feedback add 0.88 microseconds, producing a 1.20-microsecond loop. Without proved overlap, the system misses the contract by 20 percent. The artifact preserves every term so the organization knows whether the next experiment belongs in decoder optimization, data movement, or control scheduling.
| Stage | Latency | Unit | Owner |
|---|---|---|---|
| Acquire | 0.28 | microseconds | readout/control |
| Transport | 0.35 | microseconds | interconnect |
| Decode | 0.32 | microseconds | decoder |
| Feedback | 0.25 | microseconds | control |
| Total | 1.20 | microseconds | system |
Artifact contract. A units-bearing latency/bandwidth/error table and executable cycle-budget test. Acquisition, transfer, decode, and feedback terms sum to the total; the total must fit the declared cycle; throughput and logical-error claims are tied to code distance and error model.
STAGES = ("acquire", "transport", "decode", "feedback")
def inspect_stack(dossier):
errors = []
if not dossier.get("source_ids") or type(dossier.get("code_distance")) is not int or not dossier.get("error_model"):
errors.append("source_code_or_error_model")
latencies = dossier.get("latencies", {})
for stage in STAGES:
row = latencies.get(stage, {})
if type(row.get("value")) not in (int, float) or row.get("unit") != "us" or not row.get("source_id"):
errors.append(stage + "_schema")
cycle, throughput, syndrome_rate = (dossier.get(name, {}) for name in
("cycle", "decode_throughput", "syndrome_rate"))
if type(cycle.get("value")) not in (int, float) or cycle.get("unit") != "us":
errors.append("cycle_schema")
if any(item.get("unit") != "Msyndromes/s" or type(item.get("value")) not in (int, float)
for item in (throughput, syndrome_rate)):
errors.append("throughput_schema")
total = sum(latencies.get(stage, {}).get("value", 0) for stage in STAGES)
fits = not errors and total <= cycle["value"] and throughput["value"] >= syndrome_rate["value"]
return {"decision": "invalid" if errors else ("interface-pass" if fits else "cycle-miss"),
"errors": sorted(set(errors)), "total_us": total,
"margin_us": None if errors else cycle["value"] - total}
dossier = {"scenario": "distance-7 syndrome loop", "source_ids": ["scenario:bench-2026-08-01"],
"code_distance": 7, "error_model": "matched circuit-level depolarizing model",
"latencies": {name: {"value": value, "unit": "us", "source_id": "scenario:trace-17"}
for name, value in {"acquire": 0.28, "transport": 0.35,
"decode": 0.32, "feedback": 0.25}.items()},
"cycle": {"value": 1.00, "unit": "us"},
"decode_throughput": {"value": 1.2, "unit": "Msyndromes/s"},
"syndrome_rate": {"value": 1.0, "unit": "Msyndromes/s"}}
base_result = inspect_stack(dossier)
bad_latency = {**dossier["latencies"], "decode": {"value": 0.32, "unit": "", "source_id": "scenario:trace-17"}}
bad_result = inspect_stack({**dossier, "latencies": bad_latency})
faster = {**dossier, "latencies": {**dossier["latencies"],
"decode": {"value": 0.10, "unit": "us", "source_id": "scenario:FPGA-batch"}}}
faster_result = inspect_stack(faster)
assert abs(base_result["total_us"] - 1.20) < 1e-12 and base_result["decision"] == "cycle-miss"
assert bad_result["decision"] == "invalid" and "decode_schema" in bad_result["errors"]
assert faster_result["decision"] == "interface-pass" and faster_result["margin_us"] > 0
print(f"PASS: 76 correction workbook total={base_result['total_us']}us invalid={bad_result['errors']} faster_margin={faster_result['margin_us']}us")Exact validation command: python3 tools/validate_briefs.py --briefs data/editorial_briefs_64_87.json --from 64 --through 87 --check-rewritten-sources --execute-artifacts
Overhead claims need an error model and distance
Value the owned interface, not the fastest kernel. Ask for tail latency under representative syndrome rates, accuracy under the declared correlated-error model, scaling with distance, integration data movement, and logical-error change in matched conditions. A decoder can be technically strong and commercially useful without yet proving system-level logical advantage; the memo should say both.
Audit a real-time correction-stack claim
Prompt. Budget a syndrome path for a declared code distance and cycle time.
Deliverable. Table of acquisition, transport, decode, feedback, bandwidth, code distance, physical-error model, and resulting logical metric.
Pass condition. Latency terms reconcile in microseconds, throughput covers the syndrome rate, and no logical improvement is credited without matched before/after conditions.
Model answer: sound decoder, unproved system advantage
Format. Decoder budget that meets algorithmic accuracy but misses real-time feedback.
The model answer credits algorithmic decoder performance and rejects the real-time claim under the stated sequential budget. The fixture returns 1.20 microseconds and fails the one-microsecond cycle. It would pass only after a documented pipeline overlap or component improvement changes the sum. Logical-error benefit remains a separate gate requiring matched system measurements.
Verification. The fixture reproduces the latency overrun and changes to pass only when a documented batching or hardware change brings the total below the cycle.
Companion work
Artifacts for this chapter
These entries resolve to checked-in local source. Commands are reproduced exactly from the chapter manifest, and source-embedded fixtures are exported as direct downloads.
diligence dossier
Syndrome-to-feedback interface budget
Reproduce or test
python3 tools/validate_briefs.py --briefs data/editorial_briefs_64_87.json --from 64 --through 87 --check-rewritten-sources --execute-artifacts
Provenance
Sources and review
- Peter W. Shor. Scheme for reducing decoherence in quantum computer memory. Physical Review A. 1995primary paper
- Andrew M. Steane. Error correcting quantum code. Physical Review Letters. 1996primary paper
- Daniel Gottesman. Stabilizer codes and quantum error correction. California Institute of Technology / arXiv. 1997doctoral thesis
- Daniel Gottesman, Alexei Kitaev, and John Preskill. Encoding a qubit in an oscillator. Physical Review A. 2001primary paper
- Craig Gidney and Martin Ekerå. How to factor 2048 bit RSA integers in 8 hours using 20 million noisy qubits. Quantum. 2021primary peer-reviewed resource estimate
- U.S. Government Accountability Office. Quantum Computing and Communications: Status and Prospects. GAO. 2021government technology assessment
- Austin G. Fowler et al.. Surface codes: Towards practical large-scale quantum computation. Physical Review A. 2012peer-reviewed review
- Earl T. Campbell, Barbara M. Terhal, and Christophe Vuillot. Roads towards fault-tolerant universal quantum computation. Nature. 2017peer-reviewed review
The load-bearing claims in the chapter are mapped inline to this registered source set. A citation supports only the bounded claim beside it.