Configuration Overview
SPIDER uses strict nested JSON configuration validated through config_v2.
Core modules:
spider/core/config_v2/validate.pyspider/core/config_v2/load.pyspider/core/config_v2/legacy_bridge.py
Top-level blocks
io
Input/output file paths and checkpoint controls.
model
EikoNet model path and loader settings.
Domain scaling and bounds.
Priors.
Likelihoods for MAP and sampling phases.
Data filters.
inference
Sampler backend and hyperparameters.
Batch settings.
Runtime and safety controls.
observability
W&B run controls.
Diagnostics group toggles.
High-impact settings to validate early
model.likelihoods.sample.typeshould match your intended sampling model.model.likelihoods.sample.shared_event_re.enabledshould be explicitly set.inference.sampler.backendshould be eitherpsgldorsghmc.inference.batching.standard.shuffleshould typically betrueduring production sampling.inference.sampler.freeze_preconditioner_samplingshould match your Phase 4 policy.
For shared-event PCG setup and convergence checks, see PCG Whitening: Convergence and Fallback Control.
For a field-by-field schema reference, see Configuration Reference (config_v2).
For a symbol-level probability-model description with config-to-symbol mapping, see Probability Model and Symbol Mapping.