EurekaLab

Budget‑aware sandbox for autonomous scientific discovery with provenance.

---

EurekaLab (sandbox_science) wraps untrusted, agent‑generated code in a budget‑aware execution environment. It estimates and enforces a cost budget before running, captures every run in a Git‑backed provenance store, and audits results for reward‑hacking — turning the execution environment itself into a first‑class, reusable abstraction.

Installation

pip install git+https://github.com/Lumi-node/eureka-lab.git

Requires Python ≥ 3.10. To work on the project locally:

git clone https://github.com/Lumi-node/eureka-lab.git
cd eureka-lab
pip install -e ".[dev]"
pytest -q

Quick Start

import tempfile
from sandbox_science import Sandbox, ExperimentRequest

# A sandbox enforces a cost budget and records provenance
sandbox = Sandbox(workspace=tempfile.mkdtemp())

request = ExperimentRequest(
    code="print('hello from the sandbox')",
    budget=10.0,
    timeout_seconds=5,
    memory_limit_mb=256,
)

result = sandbox.submit(request)
print(result.success)                  # True
print(result.run_log.stdout.strip())   # hello from the sandbox
print(result.cost_actual.total_cost)   # measured cost, <= budget

Features

• Budget‑aware execution — estimate and enforce cost limits before a run starts
• Git‑backed provenance — every run committed for full reproducibility
• Reward‑hacking auditor with cross‑validation
• Pluggable cost model and policy engine

Modules

Module	Description
auditor	—
cost_model	—
executor	—
policy	—
provenance	—
sandbox	—

Documentation

📖 Full documentation: https://lumi-node.github.io/eureka-lab/ 📄 Technical paper: see paper/ for the LaTeX source and compiled PDF.

This is a reference implementation produced by an autonomous research pipeline. It is not published to PyPI; install from source as shown above.

License

MIT © Andrew Young / Automate Capture Research