labrecipe 0.1.1

Recipe-driven, reproducible analysis of experimental data

LabRecipe

LabRecipe is an open-source Python 3.12+ package for recipe-driven, reproducible analysis of one local experimental CSV, TSV, or explicitly bounded named-sheet XLSX dataset. A strict YAML recipe names data roles and units, explicit preprocessing, transparent analyses, figures, and output policies; one run produces a traceable evidence bundle.

LabRecipe never silently sorts, drops, interpolates, coerces dtypes, infers units, smooths, chooses a baseline, or tunes a parameter. Raw input bytes are immutable and hashed before parsing. Every derived stage, metric, figure, diagnostic, and artifact remains traceable to the raw hash and fully resolved recipe.

Install

python -m pip install labrecipe

For bounded XLSX input:

python -m pip install "labrecipe[excel]"

Tested quickstart

From a repository checkout, run the committed analytical linear control into a new directory:

labrecipe validate examples/clean_time_series/recipe.yaml
labrecipe inspect examples/clean_time_series/recipe.yaml
labrecipe plan examples/clean_time_series/recipe.yaml
labrecipe run examples/clean_time_series/recipe.yaml --output clean-bundle

The same engine is available from Python:

from labrecipe import run_recipe

result = run_recipe(
    "examples/clean_time_series/recipe.yaml",
    "clean-bundle-api",
)
print(result.manifest.identity.run_id)

Both workflows publish only a complete, hash-validated bundle and refuse to replace an existing target.

What version 0.1 covers

• strict, safely loaded, versioned recipes with complete submitted/resolved identities;
• immutable CSV/TSV and optional bounded XLSX ingestion with structured quality diagnostics;
• explicit group-safe transforms for missingness, structure, units, smoothing, signal processing, normalization, and baseline correction;
• contextual summaries, distributions, scatter/correlation, and conservative outlier/change candidates that are never promoted automatically to physical conclusions;
• evidence-bearing PNG and optional SVG/PDF figures;
• deterministic serialization, artifact-specific numerical tolerances, complete provenance, and atomic no-overwrite publication;
• equivalent validate, inspect, plan/explain, and run behavior through Python and CLI.

Single-dataset local batch analysis is the v1 boundary. Instrument control, arbitrary executable recipe content, external plugins, multi-file cohort workflows, and GUI/cloud services are not part of this release.

Documentation and examples

• Quickstart
• Recipe schema
• CLI and Python API
• Method assumptions
• Troubleshooting
• Reproducibility contract
• Representative example corpus

Contributing and citation

See CONTRIBUTING.md and DEVELOPMENT.md before proposing a change. Security issues follow SECURITY.md. Cite the software using CITATION.cff.

LabRecipe is licensed under Apache-2.0.