{
  "unit": "string x locale",
  "aggregation": "median",
  "headline": {
    "metric": "mean",
    "higher_is_better": false,
    "unit": "fraction"
  },
  "secondary": [
    {
      "id": "icu_valid",
      "metric": "mean",
      "higher_is_better": true,
      "unit": "fraction",
      "aggregation": "median"
    },
    {
      "id": "plural_complete",
      "metric": "mean",
      "higher_is_better": true,
      "unit": "fraction",
      "aggregation": "median"
    },
    {
      "id": "glossary_adherence",
      "metric": "mean",
      "higher_is_better": true,
      "unit": "fraction",
      "aggregation": "median"
    },
    {
      "id": "key_set_equal",
      "metric": "mean",
      "higher_is_better": true,
      "unit": "fraction",
      "aggregation": "median"
    },
    {
      "id": "markup_preserved",
      "metric": "mean",
      "higher_is_better": true,
      "unit": "fraction",
      "aggregation": "median"
    },
    {
      "id": "after_retry",
      "metric": "mean",
      "higher_is_better": false,
      "unit": "fraction",
      "aggregation": "median"
    },
    {
      "id": "locale_de",
      "metric": "mean",
      "higher_is_better": false,
      "unit": "fraction",
      "aggregation": "median"
    },
    {
      "id": "locale_ja",
      "metric": "mean",
      "higher_is_better": false,
      "unit": "fraction",
      "aggregation": "median"
    },
    {
      "id": "locale_pl",
      "metric": "mean",
      "higher_is_better": false,
      "unit": "fraction",
      "aggregation": "median"
    },
    {
      "id": "locale_ar",
      "metric": "mean",
      "higher_is_better": false,
      "unit": "fraction",
      "aggregation": "median"
    },
    {
      "id": "locale_pt_BR",
      "metric": "mean",
      "higher_is_better": false,
      "unit": "fraction",
      "aggregation": "median"
    }
  ],
  "repeats": 3,
  "seed": 42,
  "verdict_rule": {
    "bootstrap": "paired-items",
    "resamples": 10000,
    "confidence": 0.95,
    "prng": "mulberry32",
    "quantile": "type-7",
    "range": "absolute-difference-exceeds-wider-run-range",
    "label_tie": "json-lexicographic",
    "zero_division": 0,
    "degenerate_kappa": 1,
    "rounding": "half-away-from-zero",
    "decimals": 4
  },
  "record_shape": [
    "score"
  ],
  "prompt": "You localize application interface strings. Translate the given English string into the requested locale and call record. Preserve every placeholder token exactly as written ({name}, {count}, %s, %1$s and {{name}}). Keep markup tags such as <strong></strong> and <a href=\"...\"> intact, with attributes unchanged. Preserve ICU plural messages: keep {count, plural, ...} with one clause per required category and # for the number. Use the supplied glossary term untranslated when one is given. Return the same key with your translation. Do not add quotes or explanations.",
  "fixture_sha256": "5f4971588347fa747a2b752696afa412c46dbc1f7a8dd34feba7660cf4ac59f9",
  "recipe_sha256": "c7050d65dbfed559333932be93898e7d4115e7fd5da61b61756b659f41a5af1b",
  "checker_sha256": "e12ba97bdd489f2c4165e1d11a5ce884caccd57230736dc1e67bdad942718700",
  "runner_sha256": "40d6076708c0b2b3e6d65149f83104133a9b754da480f00d82d2d9eeafdc44c2",
  "max_tokens": 512,
  "reasoning_effort": "medium",
  "tool_choice": {
    "type": "function",
    "function": {
      "name": "record"
    }
  },
  "planning": {
    "score_bounds": [
      0,
      1
    ]
  },
  "tuning_repeats": 1,
  "retain_attempts": true,
  "fixture_source": {
    "source": "Original synthetic application strings; generator-intended structural labels",
    "license": "CC0-1.0",
    "generators": [
      "Pareto (authoring assistant), deterministic Python renderer"
    ]
  }
}