Post-Selection Laundering

This article is part of the Quantum Snake Oil Dictionary — a series examining terms used in quantum technology marketing. The series is divided into Red Flag Terms (terms with no established technical meaning that almost always signal hype or fraud) and Misused Terms (legitimate concepts routinely stripped of context in marketing). This entry is a Red Flag Term.

“Post-Selection Laundering”

A note before we begin. This article examines the practice of using extensive data filtering in quantum computing experiments and then describing the filtered results using language associated with fault-tolerant quantum computing. I am not referring to any specific company, product, or individual. Post-selection is a well-known and sometimes legitimate technique in quantum computing research. The issue arises when its use is obscured or when the filtered results are presented as evidence of fault tolerance.

What It Claims

A vendor or research group reports a quantum computation with “zero detected errors” or “perfect logical fidelity.” The numbers sound impressive, suggesting the system is operating fault-tolerantly. The claim, implicit or explicit, is that the quantum hardware performed the computation without making mistakes.

Where It Breaks Down

The question is not how many errors were detected in the final reported results. The question is how many results were discarded before the counting began.

Post-selection means running a quantum computation many times and keeping only the results that pass some set of quality checks. The survivors, by construction, look clean. This is analogous to a teacher reporting a 100% pass rate after quietly removing the failing exams from the pile.

Post-selection is a legitimate and well-understood technique in quantum computing research. Researchers use it to filter for runs where no detected errors occurred, allowing them to study the behavior of the circuit in the error-free subspace. The critical requirement is transparency: the post-selection rate (the fraction of runs that survived) must be reported, because it determines whether the technique can scale to useful computation.

The scaling problem is the core issue. Post-selection survival rates drop exponentially with circuit depth. If each gate has a small probability of error, the probability that an entire circuit runs error-free is roughly (1−p)^N, where p is the per-gate error rate and N is the number of gates. For shallow circuits on a few qubits, the survival rate can be acceptable. For deep circuits on many qubits, the survival rate approaches zero and no amount of filtering produces useful output at practical rates.

This is exactly why fault-tolerant quantum error correction exists. FTQC detects and corrects errors during the computation, allowing the circuit to keep running. Post-selection detects errors after the computation and discards the affected runs. The first enables arbitrarily long computation. The second collapses under its own filtering overhead as circuits scale.

The Laundering Pattern

Post-selection becomes a red flag when three conditions are met simultaneously:

The filtering is multi-layered and partially obscured. The system applies per-shot parity checks, per-term quality gates, and per-run admissibility criteria at different stages. Each layer discards some fraction of the data. The total rejection rate across all layers may be high, but no single summary number is provided.

The claim uses FTQC language. The vendor describes the filtered results using terms like “fault-tolerant,” “zero logical errors,” “logical fidelity F=1.0000,” or “governed fault tolerance.” These terms carry specific meanings in the QEC community, and applying them to post-selected results blurs the line between mitigation and correction.

The rejection rate is withheld. When asked directly what percentage of total data (shots, terms, or runs) was discarded across all filtering stages, the vendor either does not answer, redefines the question, or provides a number that covers only one stage of filtering while omitting the others. If the number supported the FTQC claim, there would be no reason to withhold it.

A specific tell: claiming “Data rejection / post-selection: None” while simultaneously describing a per-shot parity filter or a per-term veto mechanism elsewhere in the same document. The “None” claim typically refers to whether entire runs were discarded, while per-shot and per-term filtering within each run goes unmentioned in the summary.

What Legitimate Practice Looks Like

Post-selection is standard practice in quantum computing research and there is nothing wrong with using it, as long as three conditions are met:

The post-selection rate is reported. Every paper using post-selection should state what fraction of total shots survived to the final result. Without this number, the result cannot be evaluated.

The language matches the method. Results obtained through post-selection should be described as “post-selected” results, not as evidence of fault tolerance. The distinction between discarding bad results and correcting errors during computation is not semantic; it determines whether the approach can scale.

The scaling behavior is addressed. A credible paper will discuss how the post-selection rate changes as circuits get deeper or as the problem size increases. If the rate drops exponentially, the approach has a known scaling ceiling, and the paper should acknowledge it.

Questions to Ask a Vendor

“What is the total rejection rate across all filtering stages?” Not just per-run rejection, but per-shot parity filtering, per-term admissibility gates, and any other quality checks. The sum across all stages is the number that determines whether the technique scales.

“How does your survival rate change as circuit depth increases?” If the survival rate drops exponentially with depth, the approach cannot support the deep circuits required for useful quantum computation.

“Are you using the term ‘fault-tolerant’ in the standard QEC sense?” The standard definition requires that logical error rates decrease as code distance increases (the below-threshold criterion). If the vendor’s definition differs, ask them to state it explicitly and explain why the standard definition does not apply.

The Bottom Line

Post-selection is a filter, not a fix. It can produce clean-looking results on shallow circuits, and it is a perfectly legitimate technique when reported transparently. Calling it fault tolerance, burying the rejection rate, or using FTQC language to describe filtered results misleads anyone trying to evaluate the system’s actual capabilities. The question is never “how many errors did you find in the data you kept?” It is always “how much data did you have to throw away?”