Experiments

A labflow experiment is:

1. A @dataclass config (the "parameters" — single source of truth)
2. A function decorated with @experiment(config=...) (the "computation")

Both live in ONE file. No name-coupling. Config fields include a seed: int — required for reproducibility.

Minimal example

from dataclasses import dataclass
from labflow import experiment

@dataclass
class HelloConfig:
    name: str = "world"
    seed: int = 42

@experiment(config=HelloConfig, tags=["hello"])
def hello(cfg):
    return {"greeting": f"Hello, {cfg.name}!"}

Registry

All @experiment-decorated functions are discovered on first labflow invocation via recursive import of experiments/*.py. See labflow.registry.ExperimentRegistry.