{
  "schema_version": 1,
  "evaluation_id": "obstacle-v2",
  "status": "frozen_before_inference",
  "created_at": "2026-11-04T01:17:00Z",
  "relationship_to_prior": "Same focus-trap challenge and interpretation. obstacle-v1 ended before browser launch or inference with zero reserved requests.",
  "infrastructure_change": "Add trap to the runner internal allowedVariant list and give the result a versioned obstacle evaluation label.",
  "challenge": {
    "owned_fixture": "trap",
    "task": "Choose Blue, then add it to the practice kit",
    "model_input": "structured text observation only",
    "screenshots_sent_to_model": false,
    "known_mechanism": "The fixture cycles Tab focus inside the choice controls; the completion control is outside that cycle."
  },
  "acceptance": {
    "maximum_model_requests": 8,
    "automatic_retries": 0,
    "unsafe_actions_allowed": false,
    "publish_regardless_of_outcome": true,
    "replay_additional_model_requests": 0
  },
  "interpretation": {
    "bounded_noncompletion": "Supports the claim that the agent can expose an obstacle with an inspectable focus trace.",
    "completion": "Disproves the expected obstacle and must still be published without tuning.",
    "infrastructure_or_provider_error": "Does not test the product claim and must be reported separately."
  },
  "reporting_rule": "Run once after the allowlist fix, do not retry inference, and preserve the result exactly as produced."
}