Dataset of Adaptive Quantum Orchestration with Recurrent-Depth (AQORD)

This dataset contains the empirical validation logs, hardware telemetry, and simulated benchmarks generated for the framework introduced in the paper: "Recurrent-Depth Mixture-of-Experts with Bidirectional Quantum-Classical Routing for Adaptive Circuit Compilation and High-Dimensional Feature Optimization on NISQ Processors".

The data documents the real-time routing decisions, computational latencies, circuit knitting overheads, and error-mitigation scaling of a Recurrent-Depth Transformer (RDT) interfacing with a 108-qubit quantum topology.

---

Dataset Summary

The AQORD dataset consists of 5.3 Million rows tracking the step-by-step token routing dynamics between classical tensor processing elements and a parameterized quantum circuit (PQC) manifold. It merges physical hardware calibration profiles with large-scale structural simulation traces.

• Primary Author: Tunjay P. Akbarli
• Hardware: Rigetti Cepheus-1 (108 Qubits, configured as a $3 \times 4$ cluster of 9-qubit hardware chiplets)
• Total Instances: ~5,300,000 execution traces
• Data Format: Tabular (Parquet / CSV)

---

Dataset Structure

The dataset features are split into three logical categories: classical context metrics, live hardware/error telemetry, and target execution outcomes.

Schema Definitions

Column Name	Data Type	Description
token_id	int64	Unique identifier for the sequence item or operational token.
contextual_complexity	float64	Quantified structural difficulty of the circuit compilation task ($\chi_t$).
error_telemetry_lambda	float64	Live read-out of hardware noise floor and gate instability metrics ($\Lambda_t$).
routing_decision	string	Selected expert pathway: MODE_1 (AI-supported Quantum) or MODE_2 (Quantum-supported AI).
recurrent_depth_iterations	int32	The number of internal recurrent loops executed by the RDT ($H \le 8$).
circuit_knitting_cuts	int32	Number of active graph cuts ($K_c$) required to map the workload onto 9-qubit chiplets.
native_cz_fidelity	float64	Monitored two-qubit gate fidelity across the active hardware registers.
execution_latency_ms	float64	Total turnaround loop latency in milliseconds (including API/network container overhead).
execution_cost_usd	float64	Pro-rated financial footprint of the AWS Braket quantum task execution instance.
reconstruction_error	float64	Discrepancy metric between the knitted circuit output and the mathematical target state.

---

Creation and Source Data

Hardware Environment

Physical hardware constraints were derived directly from a 108-qubit multi-chiplet architecture on AWS hardware. Native gate profiles emphasize cross-chiplet routing configurations using graph-isomorphism mappings to avoid multi-chiplet interconnect noise bottlenecks.

Simulation Scale

The dataset contains the complete log matrix of the 5,116,928-trial hardware-calibrated simulation layer. Large-scale wide states ($n \le 5,000$ qubits) are reconstructed via tensor network approximations (Matrix Product States) to evaluate theoretical scaling ceilings under variable circuit knitting limits.

---

Usage and Intent

Operational Warning: When using this dataset to train downstream routing models, note that MODE_2 (token-by-token PQC evaluation) introduces high non-linear latency spikes if run over public shared queues instead of prioritized hybrid job execution containers.

Key Research Use Cases

1. Dynamic Mixture-of-Experts (MoE) Benchmarking: Training and testing adaptive routers that must weigh the financial and latency costs of quantum processing against classical evaluation.
2. Quantum Error Mitigation (QEM) Modeling: Analyzing how zero-noise extrapolation and active topology routing degrade or sustain state reconstruction under escalating cut counts ($4^{K_c}$).

---

Citation

If you utilize this dataset or its structural definitions in your research, please cite the framework paper:

@article{akbarli2026aqord,
  title={Recurrent-Depth Mixture-of-Experts with Bidirectional Quantum-Classical Routing for Adaptive Circuit Compilation and High-Dimensional Feature Optimization on NISQ Processors},
  author={Akbarli, Tunjay},
  year={2026},
  note={Dataset hosted on Hugging Face: thehekimoghlu/AQORD}
}