Skip to content

Latest commit

 

History

History
59 lines (44 loc) · 2.14 KB

getting_started.md

File metadata and controls

59 lines (44 loc) · 2.14 KB

Environment and construction

Use Python 3.10 on Linux for the recorded reference environment. Supported dependency ranges are in requirements.txt; the historical snapshot is in requirements-lock.txt. The setup helper creates .venv/ without changing system Python:

PYTHON=python3.10 bash scripts/setup_environment.sh
source .venv/bin/activate

Add --locked to install the recorded dependency snapshot. The environment also includes the Hugging Face download client.

Configure services

Copy .env.example to .env for your model endpoints. Copy .env.mcp.example to .env.mcp and supply only the credentials needed by the servers you enable. Do not commit either local file. Model names in historical records identify the experiment; availability of the original provider build is not guaranteed.

Live MCP execution additionally needs Node.js (22 or newer recommended) for Node adapters, and uv for isolated Python servers:

python -m pip install uv
python mcp_servers/install.py --install-dependencies

Seven upstream projects have no explicit redistribution license and their sources are omitted. Inspect THIRD_PARTY_NOTICES.md before explicitly opting into their pinned-source fetch:

python mcp_servers/install.py --fetch-external --accept-unlicensed-upstream
python mcp_servers/install.py --install-dependencies

Two workflows

Construct new records: follow the four commands in the root README. Profile generation defaults to artifacts/profile_runs/latest/; the scenario and trajectory commands pass that profile path explicitly. Outputs are local and ignored by Git. Review each stage before proceeding.

Reproduce frozen results: obtain approved external data using docs/data_access.md, then run:

python -m eval.reproduce_results --input-root .

If data were extracted elsewhere, pass that directory as --input-root. It must contain artifacts/profile_pool/, artifacts/scenario/, artifacts/trajectories/, and artifacts/privacy_experiments/.

Generating new model outputs is not equivalent to reproducing the frozen results: provider builds, tool responses, and sampling can change.