The Signal
Qyantram Makes Quantum Compilation Decisions Auditable
A paper submitted on 2 Nov 2026 describes Qyantram, an open-source quantum platform that records compiler decisions and verifies circuit transformations. Its authors report sub-millisecond checks on modeled 127- and 156-qubit layouts, alongside a measurable increase in SWAP gates against Qiskit’s SABRE.

A paper by Sharyu Kekane, Sadique Khatib, Abhushan Bokade, Uday Wad, S. M. Kamalapur and Y. D. Bhise was submitted to arXiv on 2 Nov 2026. It describes Qyantram as an open-source quantum-computing platform built around evidence-backed results and independent checks. The system brings together compilation, error identification, quantum machine learning and an adaptive learning layer.
A compiler that shows its work
Qyantram divides compilation into stages for gate decomposition, qubit layout and routing, and translation into a target device’s gate set. Each stage is restricted to changing a single property, allowing the platform to attribute added gates to a specific transformation. It also records compiler passes, including their names, duration and the circuit size and depth they produce.
The router exposes its decision process rather than returning only a final circuit. For each routing round, it records the candidate SWAP gates, their scores and the selected move, including candidates it rejects. The paper says that record is intended to make routing inspectable and reproducible.
Idle qubits cut the verification cost
The verification method focuses on qubits that a circuit actually uses. Qyantram’s authors say untouched qubits stay in the |0⟩ state, so removing those wires from the state-vector comparison is an exact reduction rather than an approximation. The comparison’s cost therefore scales with the number of occupied qubits, k, as 2k, rather than with the device size, N, as 2N.
In the paper’s evaluation, a fixed five-qubit circuit takes 0.828 milliseconds to verify on both modeled 127-qubit and 156-qubit layouts. The authors report speed-ups of around four orders of magnitude for the restricted method at large device sizes. The check has a boundary: Qyantram skips equivalence verification when a circuit measures or resets, lacks layout information, or uses more than 12 occupied qubits.
The routing record comes with a SWAP penalty
Across 36 device-and-circuit comparisons, Qyantram’s router uses an average of 1.27 times as many SWAP gates as Qiskit’s SABRE. The paper attributes that gap to its identity starting layout and lack of optimization, choices that make each routing decision easier to trace. The authors describe the measured routing gap as indicative rather than precise, since the comparison uses six circuit families, six device graphs and a single seed.
The other reported tests cover error identification and model training. The error module enumerates possible Pauli-error patterns without sampling and matches closed-form predictions within 2.8×10−17. For batched variational-model training, the authors report speeds 25 to 50 times higher than their Qiskit comparison, with numerical agreement within 2.0×10−15 of qiskit.quantum_info.
What the paper does not establish
The authors say they have not measured whether Qyantram’s learning layer improves educational outcomes, and they report no controlled study or cohort data. They also specify that the compilation engine is separate from hardware submission, so the paper does not characterize how compiled circuits perform under real-device noise. The platform, 599 automated tests and four benchmark scripts are listed as open source.