Record dates use UTC and show when an assessment or clarification was saved. Review dates show the evidence check. Earlier records were recovered from QubitWire’s source history; no earlier score movement is inferred.
No dated name or ownership transition is recorded here. An alias is a search aid, not evidence of a legal ownership relationship.
Product scope
Superconducting cat qubits with hardware-biased error protection.
A company profile may discuss several systems. Capabilities remain attached to the product and conditions named in each source; they do not transfer automatically to related organizations. A verified security relationship has not been attached to this record.
These source updates add context. QRank scores retain their existing assessment dates.
Alice & Bob · Synthetic-data proof of concept
Alice & Bob reports LDPC-cat decoding on NVIDIA GPUs through NVQLink and Quantum Machines control electronics. The company explicitly identifies synthetic input data and says this proof of concept does not constitute real-time decoding.
Source checked Limits, research question & sources
Parallel decoding feasibility is not evidence of sustained closed-loop error correction on a physical cat-qubit processor.
Question to follow: Physical syndrome provenance, sustained throughput and latency distribution against the actual QEC cycle budget.
Underlying evidence family: https://alice-bob.com/newsroom/alice-bob-decodes/. Related announcements do not automatically add independent validation.
Dated company evidence, its limits and the next test to follow. These notes do not change a QRank score.
Alice & BobSuperconducting bosonic cat qubits
Sources reviewed 20 Oct 2026. Event dates remain attached to the original evidence.
Alice & Bob described a September 2026 result on stabilizing cat states with a voltage-biased superconducting circuit, developed with ENS Lyon researchers. The announcement concerns a physical ingredient of its bosonic approach; it is not evidence that an entire fault-tolerant machine has been completed.
The appeal of a protected physical encoding is that some error processes may become easier to manage before a larger code is built. The comparison must nevertheless follow every remaining error channel. Suppressing one kind of fault can be valuable while another becomes the limiting factor. A useful company card should show the operating point, the resources required to maintain it and what changes when the device is coupled to another encoded unit. It should also distinguish a stable stored state from a reliable computation that performs many operations. For readers, the most revealing next experiment is not necessarily a larger headline number. It may be a carefully measured trade-off demonstrating that better protection still leaves enough control, measurement speed and reproducibility to build a useful processor around the device.
For a reader evaluating Alice & Bob, the practical question is whether protection of one error channel improves the complete logical operation needed by an application. A useful next evidence check is specific: ask for both error channels, leakage, repeated-operation performance and the overhead of the associated correction scheme. Strong noise bias is useful only if the remaining error mechanisms and gates stay manageable. These are proposed tests for judging a later update, not claims that the company has already passed them.
Security status: No public security verified in reviewed material; directory/watch profile only
Facts, comparisons and next evidence
At a glance
label
Quantum role
value
Bosonic qubit design
label
Evidence date
value
2026-09-30
label
Assessment action
value
Evidence enrichment
Dated milestones
date
2026-09-30
Date scope
source publication date; unknown if undated
event
Cat-stabilization result published
status
Published primary evidence; retain the narrower maturity statement
what it does not establish
Strong noise bias is useful only if the remaining error mechanisms and gates stay manageable.
Quantum role
layer
Bosonic qubit design
Question
Whether protection of one error channel improves the complete logical operation needed by an application.
Questions for comparison
Protection versus remaining error channels
Encoded operation and measurement overhead
Reproducibility across fabricated devices
Related approaches
name
Nord Quantique
relationship
Proposed comparison of relevant dimensions; not a claim of equivalent products or a ranking
dimensions
Protection versus remaining error channels
Encoded operation and measurement overhead
Reproducibility across fabricated devices
name
Quantum Circuits
relationship
Proposed comparison of relevant dimensions; not a claim of equivalent products or a ranking
dimensions
Protection versus remaining error channels
Encoded operation and measurement overhead
Reproducibility across fabricated devices
Limits and missing evidence
Strong noise bias is useful only if the remaining error mechanisms and gates stay manageable.
Linked full paper and operating conditions
Multi-device statistics
Logical operations with full overhead accounting
What would change this assessment
trigger
Ask for both error channels, leakage, repeated-operation performance and the overhead of the associated correction scheme.
Evidence needed
Dated primary technical report, product specification or customer/issuer disclosure directly answering the test.
Test
Identify the configuration, time boundary and measured outcome; retain unsuccessful attempts and stated limitations.
Limit
Strong noise bias is useful only if the remaining error mechanisms and gates stay manageable.
A starting question
question
What is the point of a cat qubit?
answer
It stores information in specially arranged states of an oscillator. The design can reshape the error problem. A useful assessment asks which errors are reduced, which remain and what extra operations are required to correct them.
Company account of component research. Cat stabilization does not establish a complete logical processor.
Read, question, follow the source
A closer look at the evidence
Company evidence guideAlice & Bob: ask what the cat-qubit testbed lets a partner measureUse the Helium announcement to ask precise questions about encoded experiments and partner access.Context reviewed +
Alice & Bob unveiled its Helium Quantum System in June 2026 as a platform for on-premise deployment and research collaboration. The announcement describes a system engineered to encode a first logical qubit using as few as eighteen cat qubits, and an intended upgrade path. A later September announcement reports a collaboration with CEA on the quantum software stack and hybrid computing. These sources support a profile about a testbed and integration programme; they do not establish that every advertised logical capability has already been demonstrated for external users.
The distinctive question is how a partner would test the architecture’s central promise. A biased-noise approach aims to suppress some error processes strongly enough that the remaining correction problem becomes more manageable. That promise needs a measurement vocabulary that separates storage, gates, measurement, repeated correction and complete logical operations. A long-lived memory statistic cannot stand in for every one of these. The profile should let a reader ask which experiment was performed and which errors remained relevant during that experiment, without dismissing the value of improved protection against a particular error.
A useful partner brief would define a small experiment with a clearly observable outcome. For example, compare the same encoded task over several correction-cycle counts while retaining the decoder, treatment of rejected runs and calibration window. Then describe what a meaningful failure would look like: a regime where increasing protection does not improve the selected outcome, a control operation that changes the assumed noise bias, or a measurement cost that dominates the task. These are proposed evaluation questions. This record does not claim that Alice & Bob has published such a complete data set for Helium.
Integration evidence deserves its own paragraph because a testbed has to operate within a wider workflow. Researchers need to submit programmes, understand execution limits, retrieve results and connect those results to the machine configuration. A collaboration on the software stack can make these interfaces more concrete. It should still be described at the level reported by the parties. An agreement to work together differs from a supported interface, a public release and a customer-accepted deployment. Versioned documentation would be a particularly useful next source for this profile.
Technology
Cat-qubit architecture and error-correction testbed
Listing status
No direct listed security verified in consulted company sources
Ticker
Not asserted
Public exposure
Organization evidence profile; no investable instrument claimed
Current evidence
System announcement and research collaboration
Logical performance
Specific external-user experiment not independently verified
Access
Research-partner invitation; unrestricted public access unverified
Questions to take with you
Does suppressing one error solve every error problem?
No. The complete encoded operation still has to account for the error processes that remain, including those associated with control, measurement and correction. The value of a biased-noise architecture depends on how those parts work together.
Why focus on a testbed?
A testbed makes a proposed architecture measurable through specific experiments and interfaces. Its usefulness can be assessed before a universal machine exists, provided the profile states what is available and what remains a target.
Evidence context only. This guide does not change QRank scores or supply a current price.
Research collaboration on software-stack integration in hybrid centres
No production support contract or public interface inferred.
SOURCE REVIEW
Cat qubits need an explicit error model
Alice & Bob develops cat-qubit hardware and describes Helium as a testbed for error correction. The approach seeks to make one error channel easier to manage while building the remaining protection into a larger architecture.
What this establishes—and its limits
Built-in protection must not be read as removal of every error. A testbed result, threshold claim and complete fault-tolerant computer require different evidence.
Evidence to look for next
Check both bit-flip and phase-flip behaviour, leakage, logical operations and the resources needed for additional protection.
Sources reviewed . This review does not create or change a QRank assessment.
A design intended to encode a logical qubit is not a completed demonstration of universal fault-tolerant computation. Suppression of one error channel does not remove all logical errors.
The next question to watch
Demonstrate the system’s logical operations and full error budget experimentally.
A research question, not a promised milestone.
QubitWire coverage
Security · Published · Source report · Sources checked
Three demonstrators bring core error-correction operations onto hardware. The company explicitly says its logical error rate is not yet below threshold.
Hardware · Published · Source report · Sources checked
A new Alice & Bob and CEA collaboration is about making different machines work together, not declaring a winner.
Open research questions
Editorial questions for this profile. No reviewed closure decision has been recorded; unanswered questions do not lower the official score.
Open research question
Demonstrate the system’s logical operations and full error budget experimentally.
What evidence would help
A source addressing this question must identify the named system, experiment or deployment, its conditions, measured outcome and remaining limitations. Does the result report errors, circuit depth and a credible classical baseline?
What would resolve it?
An editorial review must explain which part of this question the evidence resolves. A roadmap or an announcement alone does not close it.
No resolved, partial or disputed outcome is asserted without a reviewed transition and its evidence.
Which named system or service can I use under the documented access route?
What evidence would help
Check the current terms for research-partner engagement. Identify the system version, eligibility, restrictions and whether the result used hardware or simulation.
What would resolve it?
Current primary documentation must identify the offering, eligible user and access conditions. An unspecified company-level access claim is insufficient.
No resolved, partial or disputed outcome is asserted without a reviewed transition and its evidence.
Can you access the named system today, and under which plan?
What evidence would help
Provide the source location, relevant product or generation, workload, resource constraints and reported uncertainty. Keep measurements, simulations and announced targets distinct.
What would resolve it?
A source reviewer must map the evidence to this exact scope. Missing disclosure remains unknown; it is not technical failure.
No resolved, partial or disputed outcome is asserted without a reviewed transition and its evidence.
Questions are drawn from the existing brief and sector reading guide. Research brief reviewed 23 Sept 2026. Following stores a local bookmark; it does not send email alerts or imply a scheduled review.
Since your review
Scope: Alice & Bob. Reading a page never marks its records reviewed.
Keep a baseline and explicitly mark the evidence you have reviewed. This optional journal stays in this browser.
Existing material records remain available after setup. Routine rechecks and initial entries are excluded from material counts.
Alice & Bob described a September 2026 result on stabilizing cat states with a voltage-biased superconducting circuit, developed with ENS Lyon researchers. The announcement concerns a physical ingredient of its bosonic approach; it is not evidence that an entire fault-tolerant machine has been completed.
Alice & Bob reports LDPC-cat decoding on NVIDIA GPUs through NVQLink and Quantum Machines control electronics. The company explicitly identifies synthetic input data and says this proof of concept does not constitute real-time decoding.
Keep up with the companies you follow.
Follow companies to find their published coverage here. Email alerts are not enabled.
