{
  "_note": "Trimmed from the running server-health loop. Hostnames, IDs and paths replaced; structure and values otherwise as deployed.",
  "slug": "fleet-health",
  "title": "Fleet health: zero red checks on both servers, one daily card, every repair gated",
  "validated": true,
  "host": "mac",
  "host_reasons": [
    "the health check reaches both servers over SSH; server keys never go to a cloud environment (policy)",
    "the bot token used for the card lives in the local keychain and is read by name"
  ],
  "goal": {
    "statement": "Every day the health check (24 checks across both servers) reports zero red checks, and one decision card is posted by 10:45 local time with decisions listed apart from FYIs.",
    "metric": "red_checks",
    "target": 0,
    "op": "<=",
    "deadline": { "daily_at": "10:45" },
    "source": {
      "command": "python3 {home}/bin/fleet-check.py --dry-run --json",
      "json_path": "red_checks",
      "timeout_s": 180,
      "on_error_value": null
    }
  },
  "sub_goals": [
    {
      "id": "no_red_checks",
      "weight": 0.3,
      "tier": 1,
      "target": 0,
      "op": "<="
    },
    {
      "id": "backups_fresh",
      "description": "Freshness, not presence, on all four backup lanes; metric is the count of lanes outside their age threshold.",
      "weight": 0.15,
      "tier": 2,
      "target": 0,
      "op": "<="
    }
  ],
  "actions": [
    {
      "id": "post_daily_card",
      "cost_turns": 25,
      "cooldown_h": 20,
      "fingerprint": "{date}:card",
      "gate": false,
      "notes": "The only ungated action. The fingerprint is what stops a second card the same day."
    },
    {
      "id": "prune_docker",
      "sub_goal": "disk_headroom",
      "cost_turns": 10,
      "cooldown_h": 24,
      "fingerprint": "{date}:prune:{errors_hash}",
      "gate": true,
      "preconditions": [
        { "type": "sub_gap_unmet", "id": "no_red_checks" },
        { "type": "error_matches", "sub_goal": "no_red_checks", "pattern": "\\b(disk|headroom|disk-forecast)\\b" }
      ]
    }
  ],
  "forbidden_without_gate": [
    "restart (any container, service or daemon, on either host)",
    "prune (image or system prune; volumes are never pruned, gate or no gate)",
    "patch (package installs or upgrades)",
    "reboot (either host, for any reason)",
    "delete (files, dumps, backup objects, platform resources, messages)",
    "stop (containers, services, sync)",
    "firewall changes",
    "DNS changes",
    "a second card the same day (later runs add buttons to the existing card instead)",
    "running the health check without --dry-run (it would post its own card)"
  ],
  "budget": {
    "turns_per_run": 40,
    "runs_per_day": 3,
    "turns_per_day": 120,
    "wall_min_per_run": 30
  },
  "stop": {
    "goal_hit": "continue",
    "goal_hit_mode": "maintenance",
    "plateau_days": 3,
    "consecutive_failures": 3,
    "note": "A day with zero red checks still posts the card: a silent monitor reads like a healthy fleet. Three days with the same red checks and no answered gate is an escalation, not a routine."
  }
}