Bushfire decisions, benchmarked on public data.

# 1. get the code and create an environment
git clone https://github.com/<owner>/aufirebench.git && cd aufirebench
python -m venv .venv && . .venv/bin/activate
# 2. install with HiGHS; add SCIP and Gurobi interfaces if you have them
pip install -e .
pip install -e ".[solvers]"
# 3. confirm every released instance and stored solution
make check

# build, solve and verify a Tasmanian road-protection instance
aufirebench generate P2 --region TAS --budget_km 50 --out inst
aufirebench solve inst/p2_TAS_K3_B50_S34 --time 60 --out /tmp/sol.json
aufirebench check inst/p2_TAS_K3_B50_S34 /tmp/sol.json --bks
echo $?   # 0 valid, 20 below best known, 21 infeasible, 10 invalid

# install the optional assistant interface (Model Context Protocol)
pip install -e ".[mcp]"
# claude_desktop_config.json (or any MCP client that starts a stdio server)
{
  "mcpServers": {
    "aufirebench": {
      "command": "/path/to/aufirebench/.venv/bin/aufirebench-mcp",
      "env": {
        "AUFIREBENCH_DATA": "/path/to/aufirebench/data/instances",
        "AUFIREBENCH_LEDGER": "/path/to/aufirebench/results/bks_ledger.csv"
      }
    }
  }
}