{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "$id": "https://raw.githubusercontent.com/api-evangelist/reliability/refs/heads/main/json-schema/reliability-chaos-experiment-schema.json",
  "title": "ChaosExperiment",
  "description": "A chaos engineering experiment that intentionally injects a failure mode into a target system to validate resilience, with explicit hypothesis, scope, and halt conditions to bound blast radius.",
  "type": "object",
  "properties": {
    "name": {
      "type": "string",
      "description": "Human-readable name of the chaos experiment.",
      "example": "Checkout API survives 50% packet loss to payments dependency"
    },
    "description": {
      "type": "string",
      "description": "Description of the experiment and what it validates.",
      "example": "Validates that the checkout API gracefully degrades and triggers fallback when the payments dependency experiences high packet loss."
    },
    "hypothesis": {
      "type": "string",
      "description": "Steady-state hypothesis the experiment intends to validate or disprove.",
      "example": "Checkout API success rate remains above 99% when payments service experiences 50% packet loss."
    },
    "fault_type": {
      "type": "string",
      "description": "Category of fault being injected.",
      "enum": [
        "network_latency",
        "network_loss",
        "network_partition",
        "cpu_stress",
        "memory_stress",
        "disk_stress",
        "io_stress",
        "process_kill",
        "pod_kill",
        "container_kill",
        "shutdown",
        "dependency_failure",
        "dns_failure",
        "time_skew"
      ],
      "example": "network_loss"
    },
    "target": {
      "type": "object",
      "description": "Target of the fault injection.",
      "properties": {
        "kind": {
          "type": "string",
          "enum": ["service", "pod", "node", "container", "host", "instance", "namespace", "region"],
          "example": "pod"
        },
        "selector": {
          "type": "string",
          "description": "Selector expression identifying the target resources.",
          "example": "app=payments,env=staging"
        },
        "percentage": {
          "type": "integer",
          "description": "Percentage of matching targets to affect.",
          "minimum": 1,
          "maximum": 100,
          "example": 25
        }
      },
      "required": ["kind", "selector"]
    },
    "parameters": {
      "type": "object",
      "description": "Parameters specific to the fault type, such as loss percentage or latency in milliseconds.",
      "additionalProperties": true,
      "example": { "loss_percent": 50, "duration": "5m" }
    },
    "halt_conditions": {
      "type": "array",
      "description": "Conditions that, when met, immediately stop the experiment to limit blast radius.",
      "items": {
        "type": "object",
        "properties": {
          "metric": { "type": "string", "example": "checkout_success_rate" },
          "operator": { "type": "string", "enum": ["lt", "lte", "gt", "gte", "eq"], "example": "lt" },
          "threshold": { "type": "number", "example": 0.95 }
        }
      }
    },
    "environment": {
      "type": "string",
      "description": "Environment in which the experiment runs.",
      "enum": ["dev", "staging", "pre-prod", "production"],
      "example": "staging"
    },
    "status": {
      "type": "string",
      "description": "Current status of the experiment.",
      "enum": ["draft", "scheduled", "running", "halted", "completed", "failed"],
      "example": "completed"
    },
    "owner": {
      "type": "string",
      "description": "Team or person responsible for the experiment.",
      "example": "platform-resilience"
    }
  },
  "required": ["name", "fault_type", "target", "environment"]
}