Config schema

The source of truth for configuration is the Pydantic model tree in arcade/config.py. Invalid YAML fails at arcade.config.load_config with field paths and messages. Prefer validating with arcade validate before a long run.

Artifact

Role

configs/example.yaml

Annotated template checked into the repository

arcade template

Emit a commented skeleton to standard output

Compare classifiers

Fair bake-off pattern for your own traces

Top-level sections

Key

Model

Default intent

data

DataConfig

Required. Path, keys, demodulation, and split

classifiers

ClassifiersConfig

Who runs and per-method options

tuning

TuningConfig

Off by default

sweep

SweepConfig

Duration and truncation sweep

hardware

HardwareConfig

Analytical FPGA estimate

visualization

VisualizationConfig

Plots and summary report

optimization

OptimizationConfig

Prune, distill, or quantize when enabled

qec

QECConfig

Optional logical error and FTQC curves

data highlights

Field

Meaning

path

File path. Required.

format

auto, hdf5, npy, npz, or pickle

num_qubits

Qubit count. Required.

num_levels

2 for qubit labels; 3 for leakage or qutrit labels

qubit_bit_order

lsb0 for the zoo; msb0 in some paper YAMLs. Wrong order destroys joint accuracy.

demodulation

Digital demodulation of ADC data: enabled, frequencies, sampling_rate, and related fields

preprocessing

boxcar_window, truncate_at, normalize

split

mode: ratio or per_class with fixed counts

classifiers

Field

Notes

run

Registry names or dotted pkg.mod.Class paths. The magic value all expands to every registered name.

filter_types

Shared matched-filter bank. Default is [MF].

threshold_mode

per_qubit, best_separation, or otsu

Per-name blocks

For example fnn: { config_file: configs/nn_lienhard_fnn.yaml }

Extra keys are allowed so each method can keep its own option dictionary.

sweep, hardware, and visualization

Section

Key fields

sweep

enabled, n_points, lengths

hardware

enabled, target_fpga, clock_period_ns, run_synthesis

visualization

output_dir, format as pdf, html, or png, plus summary_sections

Validate and run

arcade validate configs/example.yaml
arcade run configs/paper_all_methods_benchmark.yaml
arcade describe fnn
arcade list

Full model docs live under API modules.