List available operations with their input kinds and output kind. Optionally filter by category (intervals, series, filtering, peri-event, correlation, alignment).

Full documentation and JSON schema of an operation's parameters.

Import a .npy/.npz/.mat/.csv file into the store and return its artifact id.

Series: the array under data_key with its parametric axis at index parametric_axis, paired with stamps_key (default 'stamps' or Collate's 'timestamps'), or stamps generated from sample_rate and start. ticks_per_unit divides raw stamps (e.g. 30000 for sample numbers at 30 kHz). axes labels every axis, the first naming the parametric axis; or pass just axis (e.g. "depth") and its unit. Intervals: ons_key/offs_key, or times_key for point events, on parametric axis. stream selects one struct in a Collate-style .mat file. CSV series use stamp_column (default: first).

List every artifact in the store.

Summarise an artifact (parametric axis and unit, shape, axes, extent, sample rate, NaN fraction) and how it was produced. Accepts a full id or a unique prefix.

Register an experiment: {name, sessions: [{id, subject, attributes: {condition: ..., day: ...}, inputs: {: | {path, kind, options}}}]}. File inputs are imported (options are import_data's keyword arguments). Returns a content-derived dataset id that pins the exact data.

A registered dataset's sessions, subjects, attributes and input artifact ids. With no id, lists datasets.

Check whether two artifacts agree numerically, e.g. the same step run on a laptop and on a cluster. Ids are hashes of exact bytes, so different hardware can give different ids for results that agree to ~1e-15; this compares kind, axes, stamps, coordinates and data within |a - b| <= atol + rtol * |b|.

Statically check a workflow or pipeline (JSON object, YAML/JSON text, or file path): references, cycles, input/output kinds and parameters. Returns the execution order and each step's output kind; pipelines are compiled first (session steps appear as '__').

Run a workflow or pipeline locally. Steps whose (operation, version, params, inputs) already ran are reused from cache unless force, so adding sessions to a dataset only runs the new ones. Returns run_id, per-step outputs (summarised for pipelines) and the workflow outputs.

Draw a workflow or pipeline: Mermaid flowchart text (render it for the user), or the node / edge graph as JSON for a visual editor. Pipelines show their session steps once, inside an 'each of N sessions' block, with the artifact kind on every edge.

Full provenance record of a run (resolved workflow, environment and package versions, every step's params, inputs, outputs and cache key). With no run_id, lists recent runs.

Compile a workflow or pipeline into cluster jobs that each run one step against a shared store (a pipeline's sessions run in parallel; its aggregate steps wait for all of them).

slurm options: store_path, partition, time, mem, cpus, account, setup (shell lines, e.g. module load), sciflow_cmd, extra_sbatch. argo options: image, namespace, store_pvc, cpu, memory.

Write an artifact to disk as .npz, .mat (Collate struct layout, readable in MATLAB), .csv, or 'table': a long-format CSV with one row per element and columns for every axis plus subject / condition / other attributes of stacked sessions (ready for mixed-effects models in R or statsmodels).