Add docker-compose + deploy.sh (Coolify replacement)

Run on mewtwo: bound 0.0.0.0:8081 -> nginx :80.
Rebuilds site from source on every 'deploy.sh up' so new grades
flow through without Coolify. Point Netbird at http://<host>:8081.

Co-Authored-By: Claude <noreply@anthropic.com>

deploy.sh

@@ -0,0 +1,26 @@
+#!/usr/bin/env bash
+# Deploy the benchmark dashboard via docker compose.
+#   ./deploy.sh         -> regenerate site from source, rebuild, restart
+#   ./deploy.sh logs    -> tail container logs
+#   ./deploy.sh down    -> stop & remove container
+#   ./deploy.sh status  -> show running state
+#
+# The site rebuilds from data/benchmark_history.json on every run, so just
+# add a grade to the JSON, then run this. Bound to 0.0.0.0:8081 — point
+# Netbird (or any reverse proxy) at http://<this-host>:8081 for HTTPS/subdomain.
+set -euo pipefail
+cd "$(dirname "$0")"
+
+case "${1:-up}" in
+  up)
+    # --build forces a rebuild so new grades/code take effect
+    docker compose up -d --build
+    echo
+    echo "✓ dashboard up: http://$(hostname -I 2>/dev/null | awk '{print $1}' || echo localhost):8081/dashboard.html"
+    echo "  (bind 0.0.0.0:8081 -> container :80)"
+    ;;
+  logs)   docker compose logs -f --tail=100 ;;
+  down)   docker compose down ;;
+  status) docker compose ps ;;
+  *) echo "usage: $0 [up|logs|down|status]"; exit 1 ;;
+esac

docker-compose.yml

@@ -0,0 +1,13 @@
+services:
+  benchmark:
+    # Builds from the repo's Dockerfile:
+    #   stage 1 (python) runs generate_dashboard.py from data/benchmark_history.json
+    #   stage 2 (nginx) serves dashboard.html + pages/ as static
+    build: .
+    image: llm-benchmark:latest
+    container_name: llm-benchmark
+    restart: unless-stopped
+    ports:
+      - "0.0.0.0:8081:80"   # bind all interfaces:8081 -> container :80
+    # Re-tag the image so `docker compose up` after a code change rebuilds it.
+    # (compose detects Dockerfile/context changes and rebuilds automatically.)