MARIS-AI (v0.2.1) — CAIS Experimental Stack + Statistics/Report

End-to-end, self-contained experimental stack for:

• multi-agent maritime-inspired encounters (2D),
• governance-constrained decision execution (G),
• audit trace semantics (Φ) with hash chaining,
• replayability verification (Ψ),
• federated learning (FedAvg),
• Results-ready exports (CSV/JSON/PNG) + statistical comparisons + LaTeX tables.

Install

python -m venv .venv && source .venv/bin/activate
pip install -e .

Sweep governance modes

maris-cais-sweep --episodes 30 --steps 200 --agents 5 --seed 42 --scenario crossing --centralized --projection slsqp

Plots

maris-cais-plots outputs/

Integrity replay (Ψ)

maris-cais-replay outputs/<run_id>/

Federated run

maris-cais-federated --rounds 20 --clients 10 --agents 5 --seed 7 --mode project --projection slsqp

Paper-ready report (stats + LaTeX)

maris-cais-report outputs/ --metric violation_rate

Outputs:

• outputs/reports/summary_with_ci.csv
• outputs/reports/pairwise_tests.csv
• outputs/reports/table_summary.tex
• outputs/reports/table_pairwise.tex