Syndrome-Conditioned Quantum Control

• Syndrome-conditioned quantum control is the dynamic use of error syndromes to steer decoding, mitigation, and routing decisions, fundamentally reshaping error correction strategies.
• Methodologies like SAGMS and QAOA adapt message gains and control logic based on live syndrome data, yielding significant improvements in logical error rates and reduced overhead.
• The approach spans multiple layers—from physical gate adjustments to network-level routing—demonstrating its potential to enhance robustness in scalable quantum architectures.

Searching arXiv for the provided topic and linked works to ground the article in current literature. Syndrome-conditioned quantum control is the use of measured error syndromes to adapt correction, decoding, mitigation, or routing decisions in a closed loop. In the stabilizer setting, the conditioning signal is the observed pattern of stabilizer outcomes; in broader formulations it can be a residual syndrome, a syndrome history, a parity of detector outcomes, a no-decay event, or an aggregate syndrome distribution. The concept appears at several layers: inside message-passing decoders, in adaptive syndrome-extraction schedules, in logical-layer mitigation and estimation, in physical no-jump gate design, and in network control planes that infer time-varying error structure from passive syndrome telemetry (Cordova et al., 11 May 2026, Aharonov et al., 29 Dec 2025, Sun, 24 Sep 2025, Fan et al., 7 Jun 2026).

1. Conceptual scope and control layers

A useful organizing principle is that syndrome-conditioned control is not restricted to applying a recovery Pauli after decoding. The conditioning variable can shape the dynamics of the decoder itself, the choice of which checks to measure next, the logical measurement basis, the mitigation policy, or even the route and code used by a quantum network.

Layer	Conditioning signal	Controlled object
Physical gate/readout	No-decay event, jump record	Gate waveform, readout acceptance
Syndrome extraction	Partial syndrome, detector parity	Which checks to measure, when to stop
Decoder dynamics	$r^{(\ell)}$, $f_\ell$, soft readout	Message gains, stopping, update schedule
Logical layer	$s$, $s_{1:j}$, detector region parity	Pauli frame, QP correction, measurement basis
Network layer	Syndrome histograms $h^{\mathrm{obs}}$	Route, swap tree, code selection

In one formulation, syndrome-conditioned control “broadly refers to adapting the correction strategy based on the observed stabilizer measurement outcomes (syndromes), i.e., measurement-based feedback,” while in physical quantum control it can mean “adapting gate sequences or recovery operations conditioned on syndrome bits” (Cordova et al., 11 May 2026). At the logical layer, SALEM treats the syndrome record as a feedforward control signal: the controller applies the ML recovery $\sigma_s$, then further inverts the residual syndrome-conditioned logical channel $\Lambda_{L|s}$ using quasi-probabilistic correction or global post-processing (Aharonov et al., 29 Dec 2025). In neutral-atom and atomic-qubit settings, successful operation is post-selected on a syndrome indicating no decay, so the control problem becomes conditional wave-function evolution under no-jump dynamics (Sun, 24 Sep 2025). At the network layer, SCOPE defines syndrome-conditioned quantum control as a “closed-loop control-plane mechanism” that bases route and code decisions on live syndrome distributions rather than on topology or scalar fidelities (Fan et al., 7 Jun 2026).

This breadth is important because it separates syndrome-conditioned control from the narrower notion of conventional decoding. A plausible implication is that syndrome use is best viewed as a control resource that can be consumed at multiple abstraction levels rather than as a terminal classical by-product of error correction.

2. Formal structure of conditioning

For a QLDPC code with parity-check matrix $H$ over $GF(4)$, a received error pattern $e \in \{I,X,Y,Z\}^n$ produces a binary syndrome through the trace inner product,

$f_\ell$0

Decoding seeks $f_\ell$1 such that $f_\ell$2 over $f_\ell$3. In iterative decoders, the relevant control state is often the residual syndrome

$f_\ell$4

where $f_\ell$5 is the fraction of unsatisfied stabilizers (Cordova et al., 11 May 2026).

At the logical layer, conditioning is naturally expressed with Bayes updates. SALEM uses

$f_\ell$6

where $f_\ell$7 denotes a logical error event. This posterior governs reweighting, selective resampling, acceptance or rejection, and syndrome-dependent quasi-probability inversion (Aharonov et al., 29 Dec 2025). A related estimation-theoretic formulation models the decoded output as a classical–quantum state

$f_\ell$8

and distinguishes two regimes: classical syndrome-aware protocols, in which the logical measurement basis is fixed and $f_\ell$9 is used only in classical post-processing, and quantum syndrome-conditioned control, in which the logical measurement basis and control operations depend on $s$0 (Tsubouchi et al., 5 Mar 2026).

The distinction is operationally sharp. For classical syndrome-aware protocols under Pauli noise and maximum-likelihood decoding,

$s$1

so syndrome use can improve the effective logical error rate by at most a factor of two on average, implying at most a quadratic reduction in sampling overhead (Tsubouchi et al., 5 Mar 2026). By contrast, for even-distance codes with ambiguous syndromes and syndrome-conditioned quantum control,

$s$2

which yields an exponential reduction in effective logical error with the number of logical qubits (Tsubouchi et al., 5 Mar 2026). This directly addresses a common misconception: classical post-processing of syndrome records is not equivalent to quantum syndrome-conditioned control.

A distinct but related physical formalism appears in no-jump control. Under the Lindblad equation, the no-decay branch evolves with

$s$3

and the renormalized “SECOND” dynamics adds a nonlinear term that preserves normalization while retaining the non-Hermitian influence of decay channels (Sun, 24 Sep 2025). The key point is that conditioning on no decay does not remove environmental back-action; it reshapes the controlled dynamics.

3. Decoder-internal syndrome-conditioned control

One of the clearest algorithmic realizations is the syndrome adaptive gain Min-Sum decoder. In SAGMS, the check-node update is

$s$4

with

$s$5

The global factor $s$6 damps messages when many checks are unsatisfied and ramps toward $s$7 near convergence, while unsatisfied checks receive a moderate local boost $s$8 (Cordova et al., 11 May 2026).

The control interpretation is explicit: the syndrome steers the decoder’s effective gain online. Under the Uniform Message Approximation, $s$9, whereas BP4 uses $s_{1:j}$0, which acts as an implicit gain compression. Because the BP4-matching SMS factor $s_{1:j}$1 decreases strictly with $s_{1:j}$2, any fixed $s_{1:j}$3 incurs a growing penalty as check degree varies; SAGMS avoids this by making $s_{1:j}$4 depend only on $s_{1:j}$5 and $s_{1:j}$6, not explicitly on $s_{1:j}$7 or $s_{1:j}$8 (Cordova et al., 11 May 2026).

On generalized bicycle QLDPC codes, SAGMS uses $s_{1:j}$9, $h^{\mathrm{obs}}$0, $h^{\mathrm{obs}}$1 with $h^{\mathrm{obs}}$2, and the main matched-channel result at $h^{\mathrm{obs}}$3 for $h^{\mathrm{obs}}$4 and $h^{\mathrm{obs}}$5 is

• BP4: $h^{\mathrm{obs}}$6,
• SMS: $h^{\mathrm{obs}}$7,
• SAGMS: $h^{\mathrm{obs}}$8, with a statistically significant improvement over SMS and about $h^{\mathrm{obs}}$9 lower FER than BP4 at the same $\sigma_s$0 (Cordova et al., 11 May 2026). Its weighted complexity remains close to MS: $\sigma_s$1

Related decoder-level control appears in quantum data-syndrome BP. DS-BP augments the Tanner graph with syndrome-bit variables and jointly infers data and syndrome errors from

$\sigma_s$2

Its scalar check update

$\sigma_s$3

implements measurement-noise-aware message passing, and a serial schedule reduces oscillations on short cycles (Kuo et al., 2021). A complementary approach uses analog, rather than hard, syndrome data. In soft-syndrome decoding, the Gaussian readout gives $\sigma_s$4; a virtual variable node per check carries $\sigma_s$5 and updates the effective syndrome sign and reliability during min-sum iterations. For QC lifted-product QLDPC families, the hard-syndrome threshold at fixed $\sigma_s$6 occurs near $\sigma_s$7, whereas the soft-syndrome decoder shifts it to $\sigma_s$8, while also removing the “second threshold” caused by wrong-syndrome locking (Raveendran et al., 2022).

QAOA decoding provides a different syndrome-conditioned mechanism: the measured syndrome parameterizes a reward Hamiltonian, and the resulting QAOA sample distribution over corrections is used to choose a control action. At level $\sigma_s$9, check-based QAOA for the $\Lambda_{L|s}$0 Hamming code and generator-based QAOA for the $\Lambda_{L|s}$1 code match maximum-likelihood decoding, while also exposing multiple degenerate corrections with comparable probabilities on the $\Lambda_{L|s}$2 Shor code (Lai et al., 2022).

4. Syndrome acquisition, adaptive extraction, and recovery synthesis

Syndrome-conditioned control also acts on the measurement circuit itself. Adaptive syndrome extraction concatenates a $\Lambda_{L|s}$3 inner code with a hypergraph-product outer code and measures the two inner generators first. If $\Lambda_{L|s}$4 is globally trivial, the controller skips the outer stage and terminates the cycle early. Otherwise it measures only the overlapping outer checks,

$\Lambda_{L|s}$5

where $\Lambda_{L|s}$6 is the set of generators whose support overlaps that of $\Lambda_{L|s}$7 (Berthusen et al., 20 Feb 2025). Periodic “unmasking” by measuring all outer checks every $\Lambda_{L|s}$8 rounds preserves single-shot behavior. On expander and La-cross families, the adaptive concatenated scheme achieves over an order-of-magnitude lower logical error rates in low-$\Lambda_{L|s}$9 regimes while requiring fewer CNOT gates and, in the La-cross comparison, fewer physical qubits than non-concatenated baselines (Berthusen et al., 20 Feb 2025).

A more general recovery-synthesis route appears in syndrome-based Petz recovery. For arbitrary codes and noise processes, an orthogonalization procedure replaces the original Kraus set $H$0 by $H$1 satisfying

$H$2

which makes syndrome projectors onto the images $H$3 mutually orthogonal. The syndrome-based Petz map

$H$4

then becomes measurement-based and hardware-friendly (Biswas et al., 9 Oct 2025). For the $H$5 Leung code under amplitude damping, the syndrome-based Petz construction yields

$H$6

outperforming both Leung’s original recovery and the standard Petz $H$7 on that code (Biswas et al., 9 Oct 2025).

Reliable syndrome acquisition under noisy measurements is the domain of data-syndrome codes. In their base form, these augment the parity-check matrix to $H$8, so that data and syndrome-bit errors are corrected jointly (Kuo et al., 2021). More generally, syndrome measurement codes encode the $H$9-bit stabilizer syndrome with additional redundant stabilizer measurements, producing $GF(4)$0 QDS constructions (Ashikhmin et al., 2019). Primitive narrow-sense BCH syndrome-measurement codes reduce the number of extra measurements needed to protect against $GF(4)$1 syndrome-bit errors from $GF(4)$2 to $GF(4)$3, with redundancy $GF(4)$4 for $GF(4)$5 (Guttentag et al., 2023).

At the hardware level, flag-based syndrome extraction on IBM heavy-hex processors shows that fault-tolerant syndrome acquisition remains viable under sparse connectivity. On ibm_kyoto, repetition-code logical error rates decrease exponentially with distance from three to nine even when one or two flag qubits intervene between data and syndrome qubits, confirming the effectiveness of flag-based syndrome extraction under heavy-hex constraints (Kim et al., 2024). PropHunt pushes this circuit-level perspective further by optimizing syndrome measurement schedules for CSS codes directly against ambiguity in the circuit-level check matrix $GF(4)$6 and logical map $GF(4)$7. On LP and RQT benchmarks it yields $GF(4)$8–$GF(4)$9 lower logical error rate at $e \in \{I,X,Y,Z\}^n$0 than coloration circuits, and it enables Hook-ZNE, which improves zero-noise extrapolation bias by $e \in \{I,X,Y,Z\}^n$1–$e \in \{I,X,Y,Z\}^n$2 relative to distance-scaling ZNE (Viszlai et al., 24 Jan 2026).

5. Logical-layer mitigation, estimation, and probabilistic reweighting

At the logical layer, syndrome-conditioned control often takes the form of feedforward mitigation rather than direct correction. In fine-grained SALEM, each observed syndrome $e \in \{I,X,Y,Z\}^n$3 yields a per-syndrome unbiased estimator $e \in \{I,X,Y,Z\}^n$4 with variance bound $e \in \{I,X,Y,Z\}^n$5, and inverse-variance aggregation gives

$e \in \{I,X,Y,Z\}^n$6

By contrast, unconditional ExtLEM incurs the geometric-mean overhead $e \in \{I,X,Y,Z\}^n$7, so $e \in \{I,X,Y,Z\}^n$8 yields a strict advantage whenever syndrome-conditioned logical error rates are nonuniform (Aharonov et al., 29 Dec 2025). For distance-3 codes, the reported blowup rates are $e \in \{I,X,Y,Z\}^n$9, versus $f_\ell$00 for surface-code MWPM and $f_\ell$01 for the Steane LUT decoder (Aharonov et al., 29 Dec 2025). In a distance-4 surface-code memory example at physical error $f_\ell$02 and logical error $f_\ell$03, SALEM maintains $f_\ell$04 accuracy at circuit volumes about $f_\ell$05 larger than ExtLEM, about $f_\ell$06 larger than EC+PS, and about $f_\ell$07 larger than EC alone (Aharonov et al., 29 Dec 2025).

The estimation-theoretic work on noisy logical observable estimation sharpens the distinction between classical and quantum use of syndrome data. Classical syndrome-aware estimators saturate a Fisher information

$f_\ell$08

but remain subject to the factor-of-two limitation discussed above (Tsubouchi et al., 5 Mar 2026). Quantum syndrome-conditioned control instead tailors the logical measurement operator to the syndrome branch, and for even-distance codes with ambiguous branches the low-error expansion yields $f_\ell$09, leading to the exponential reduction $f_\ell$10 (Tsubouchi et al., 5 Mar 2026).

Syndrome resampling replaces uniform averaging over measured syndromes by the tilted distribution

$f_\ell$11

For surface codes under independent bit-flip noise, this produces a family of thresholds that tracks the Rényi coherent information phase transitions:

• $f_\ell$12,
• $f_\ell$13,
• $f_\ell$14, with corresponding RCI values $f_\ell$15, $f_\ell$16, and $f_\ell$17 (Colmenarez et al., 7 May 2026). Combined with complementary-gap post-selection, syndrome resampling yields up to four orders of magnitude reduction in logical error for rotated surface codes, and when applied to existing experimental QEC data it achieves nearly two orders of magnitude reduction while retaining about $f_\ell$18 of shots, compared with roughly $f_\ell$19 for full post-selection (Colmenarez et al., 7 May 2026).

Detector-region tomography supplies the characterization layer needed for such control. LSD-DRT estimates detector-conditioned logical Pauli channels by fitting

$f_\ell$20

from variable-length sequences of syndrome-extraction gadgets (Girling et al., 11 Aug 2025). On the $f_\ell$21 X-error-detecting code implemented on Quantinuum H1-1, the conditional logical channel for trivial detector parity $f_\ell$22 is approximately logical dephasing, with

$f_\ell$23

and $f_\ell$24 (Girling et al., 11 Aug 2025). This suggests a direct route from detector-conditioned tomography to online, parity-conditioned decoder weights and mitigation policies.

6. System architectures, demonstrations, and open problems

The broadest system-level embodiment is SCOPE, a syndrome-driven control plane for QEC-enabled quantum networks. Endpoints export decoder syndromes and metadata, an inference engine reconstructs a time-varying error map $f_\ell$25, and a decision engine pushes optimal route-and-code tables to sources. In NetSquid and IBM-calibrated simulations, SCOPE reduces estimation error by more than $f_\ell$26 relative to a standard EM baseline and reduces logical error rates by $f_\ell$27–$f_\ell$28 on average, up to $f_\ell$29, against topology-aware baselines (Fan et al., 7 Jun 2026). The key systems claim is that passive syndrome telemetry can replace active tomography without throughput collapse.

At smaller scale, in-situ characterization uses syndrome statistics to estimate channel parameters and even cancel the invertible part of the error channel. For the three-qubit repetition code with coherent rotation $f_\ell$30 and stochastic bit flips at rate $f_\ell$31, syndrome probabilities depend on

$f_\ell$32

and adaptive counter-rotation can suppress the coherent contribution while retaining sensitivity for estimation (Combes et al., 2014). This is syndrome-conditioned control in a literal feedback sense: syndrome statistics drive both diagnosis and real-time compensation.

Hardware demonstrations already span several controller realizations. ANN decoding on IBM heavy-hex processors implements a learned map $f_\ell$33, with two hidden dense ReLU layers and Sigmoid outputs over per-qubit X/Z corrections. For the adjusted heavy-hex code, the ANN X-logical threshold is about $f_\ell$34, close to MWPM on the same layout, and the decoder runs within microseconds for $f_\ell$35 and $f_\ell$36 (Hall et al., 2023). Flagged extraction on IBM ibm_kyoto confirms that repeated syndrome processing remains effective under constrained connectivity, while detector-conditioned tomography on Quantinuum H1-1 and adaptive extraction on concatenated HGP constructions demonstrate that syndrome-conditioned control is not tied to a single hardware modality (Kim et al., 2024, Girling et al., 11 Aug 2025, Berthusen et al., 20 Feb 2025).

The main limitations are consistent across the literature. Several analyses assume depolarizing or circuit-level local Markovian noise, and strong non-Markovian or long-range temporal correlations can violate locality assumptions used in SALEM, SCOPE, and in-situ characterization (Aharonov et al., 29 Dec 2025, Fan et al., 7 Jun 2026, Combes et al., 2014). SAGMS analysis relies on the Uniform Message Approximation and is exact at initialization for $f_\ell$37, not throughout loopy decoding (Cordova et al., 11 May 2026). FG-SALEM is hard to scale because full syndrome-conditioned characterization is at least as hard as ML decoding; CG-SALEM reduces this to fast classification plus a small lookup table, but classifier quality remains a bottleneck (Aharonov et al., 29 Dec 2025). Leakage, readout errors, drift, and rapid nonstationarity remain open issues for detector-conditioned control and network-level telemetry (Girling et al., 11 Aug 2025, Fan et al., 7 Jun 2026).

Taken together, these results define syndrome-conditioned quantum control as a hierarchy of measurement-conditioned policies: adapt message gains with $f_\ell$38 and $f_\ell$39; select which checks to measure or skip; synthesize recovery maps from orthogonalized syndrome subspaces; tailor logical mitigation, measurement bases, and resampling weights to the observed syndrome; and infer network error structure from passive decoder telemetry. The unifying technical theme is that syndrome records are not merely sufficient statistics for a final correction step. They are control variables that reshape dynamics, inference, and architecture across the full stack of fault-tolerant quantum information processing.