Ruthless Efficiency

"Our chief weapons are Ruthless Efficiency! …and warm-start caching."

A general optimisation/search substrate: a pure hexagonal core + pluggable search strategies (random built-in; evolve, optuna via extras) + pluggable compute backends.

Status

0.x — ports are still being validated against real consumers; the API may change until 1.0.

Prerequisites

• Python ≥ 3.10
• pip (or uv for development — see CONTRIBUTING.md)

Install

• pip install ruthless-efficiency — core + random strategy
• pip install "ruthless-efficiency[optuna]" — + Optuna strategy (resumable Bayesian/sampler calibration)
• pip install "ruthless-efficiency[evolve]" — + evolve strategy (our orchestration over OpenEvolve)
• pip install "ruthless-efficiency[backends]" — + SSH / HF-Jobs / Docker compute backends

Quick start

An Objective is the only thing you must implement: any class with an evaluate(candidate) method that returns a dict[str, float] of metrics satisfies the Objective protocol (duck-typed — no base class to inherit). Hand it to a strategy with a backend:

from ruthless import Candidate, InProcessBackend, RandomConfig, RandomSearchStrategy


class Quadratic:
    def evaluate(self, candidate: Candidate) -> dict[str, float]:
        return {"loss": (candidate.params["x"] - 3.0) ** 2}


cfg = RandomConfig.model_validate(
    {
        "kind": "random",
        "metric": "loss",
        "direction": "minimize",
        "n_trials": 200,
        "param_space": {"x": {"kind": "float", "lo": -10.0, "hi": 10.0}},
    }
)
result = RandomSearchStrategy(cfg, seed=42).run(Quadratic(), backend=InProcessBackend())
print(result.best.candidate.params, result.best.metrics)

Expected output (search converges on x = 3, where loss is minimised; exact for seed 42):

{'x': 2.9969320310963994} {'loss': 9.412433193460362e-06}

Or drive it from a YAML config via the CLI:

ruthless --config search.yaml --objective my_package.objectives:my_objective

Architecture

A pure hexagonal core (ruthless/) defines the ports (Objective, SearchStrategy, ComputeBackend) and value types; strategies and backends depend on the core, never the reverse (enforced by import-linter). For contributor-level detail see CONTRIBUTING.md.

To explore the C4 diagrams (System Context, Containers, and per-container Components), download docs/c4/architecture.html and open it in a browser — GitHub does not render HTML files inline.

Learn more

• CHANGELOG.md — versioned history of what changed
• Strategies: RandomSearchStrategy (core), OptunaStrategy ([optuna]), EvolveStrategy ([evolve])
• Compute backends ([backends]): build one with ruthless.backends.create_backend(...)

Contributing & community

• CONTRIBUTING.md — dev setup and the local quality gate (mirrors CI).
• CODE_OF_CONDUCT.md — Contributor Covenant.
• SECURITY.md — how to report a vulnerability and the security surface.
• NOTICE — third-party licenses and methodological references.

License

MIT © 2026 Karsten S. Nielsen