{
  "schema": "bia.vibe-benchmark-rubric/v1",
  "rubric_id": "autonomous-delivery-v1",
  "primary_metric": {
    "id": "autonomous_success_rate",
    "numerator": "trials with completed terminal, build pass, boot pass, every required functional check pass, and zero counted human interventions",
    "denominator": "every trial planned before execution, including failures and timeouts",
    "aggregation": "publish integer numerator, denominator and basis points; never publish a rounded rate alone"
  },
  "dimensions": {
    "build": {
      "pass": "the provider reports a successful production build or exposes a candidate whose immutable build artifact is retrievable",
      "fail": "the provider reports a build error or no build artifact exists at timeout",
      "not_observable": "provider exposes no build state; this cannot satisfy autonomous success"
    },
    "boot": {
      "pass": "the published or preview candidate serves its application shell and health probe without a fatal runtime error",
      "fail": "candidate does not start, returns a fatal response, or never becomes reachable",
      "not_observable": "no candidate endpoint exists; this cannot satisfy autonomous success"
    },
    "functionality": {
      "pass": "every required check in the versioned prompt corpus passes against the candidate",
      "fail": "at least one required check fails, is skipped, or cannot be evaluated"
    },
    "human_intervention": {
      "zero": "only the allowlisted setup, initial submit, native permission and single publish actions occurred",
      "positive": "any followup, edit, retry, regeneration, debugging, model change or deploy repair occurred"
    },
    "cost": {
      "measured": "USD micros backed by an accepted provider usage, invoice or versioned allocation artifact",
      "unavailable": "reported as unavailable and excluded from cost summaries; never coerced to zero"
    },
    "time": {
      "unit": "milliseconds",
      "clock": "monotonic elapsed time captured from prompt submission to terminal evaluation",
      "failed_trial": "retain elapsed duration through failure or timeout"
    }
  },
  "secondary_metrics": [
    "build_pass_rate",
    "boot_pass_rate",
    "all_functional_checks_pass_rate",
    "zero_intervention_rate",
    "median_elapsed_ms",
    "p95_elapsed_ms",
    "median_measured_cost_usd_micros",
    "failure_code_counts"
  ],
  "blind_scoring": {
    "required": true,
    "evaluator_identifier": "one-way hash only",
    "provider_identity_revealed_after_score_is_frozen": true
  }
}