{
  "metadata": {
    "experiment": "cache_policy",
    "created_at": "2026-10-02T08:49:58.826529+00:00",
    "models": {
      "Luna cached": "openai/gpt-6-luna",
      "Luna uncached": "openai/gpt-6-luna",
      "Jev": "typesafe/jev-1.13"
    },
    "seed": 20261002,
    "cases": 24,
    "calls_planned": 73,
    "budget_usd": 0.988362588,
    "charged_usd": 0.012438148999999999,
    "calls_completed": 73,
    "status": "complete",
    "rulebook_sha256": "01006795edc5f9ee13a1e4efc56fc8e19516d74f40cddbce2e72d19b00f0d43b",
    "cases_sha256": "7559a18a2c58c15206a951c2a483cae6eff11969a62b4ab799a87e919ab2c8ea",
    "code_sha256": "40976904f39e343bfcf97ad38815f70cfe9e10cd69ef2260d77bad57060d14f7",
    "rulebook_words": 1903,
    "prompt_cache_key": "luna-jev-policy-1e70847917e24fd092c9b043bb8de051",
    "source": "https://openrouter.ai/docs/guides/best-practices/prompt-caching",
    "luna_reasoning": "none",
    "luna_provider": "OpenAI",
    "protocol": "One distinct Luna cached priming call, then 24 sequential cases with all three arms randomly ordered within case; no automatic retries; actual usage cost and cache counters; exact same Luna message text and segmentation.",
    "finished_at": "2026-10-02T08:50:56.490676+00:00"
  },
  "arms": {
    "Luna cached": {
      "n": 24,
      "valid": 24,
      "cost_observed_n": 24,
      "accuracy_including_invalid_as_wrong": 0.875,
      "brier_valid_only": 0.125,
      "median_latency_s": 0.9179664584953571,
      "p95_latency_s": 1.1803516982632571,
      "known_cost_subtotal_usd": 0.00171132,
      "cost_missing_n": 0,
      "total_cost_usd": 0.00171132,
      "mean_cost_usd": 7.1305e-05,
      "cache_observed_n": 24,
      "cache_hit_calls": 24,
      "cache_read_tokens": 60792,
      "known_cache_read_subtotal_tokens": 60792,
      "cache_write_observed_n": 24,
      "cache_write_tokens": 0,
      "known_cache_write_subtotal_tokens": 0,
      "prompt_tokens": 69426,
      "cache_read_fraction": 0.8756373692852821
    },
    "Luna uncached": {
      "n": 24,
      "valid": 24,
      "cost_observed_n": 24,
      "accuracy_including_invalid_as_wrong": 0.875,
      "brier_valid_only": 0.125,
      "median_latency_s": 1.0602054579940159,
      "p95_latency_s": 1.291354949955712,
      "known_cost_subtotal_usd": 0.0071802,
      "cost_missing_n": 0,
      "total_cost_usd": 0.0071802,
      "mean_cost_usd": 0.000299175,
      "cache_observed_n": 24,
      "cache_hit_calls": 0,
      "cache_read_tokens": 0,
      "known_cache_read_subtotal_tokens": 0,
      "cache_write_observed_n": 24,
      "cache_write_tokens": 0,
      "known_cache_write_subtotal_tokens": 0,
      "prompt_tokens": 69402,
      "cache_read_fraction": 0.0
    },
    "Jev": {
      "n": 24,
      "valid": 24,
      "cost_observed_n": 24,
      "accuracy_including_invalid_as_wrong": 0.8333333333333334,
      "brier_valid_only": 0.08786666666666666,
      "median_latency_s": 0.2544942910026293,
      "p95_latency_s": 0.33727485434792465,
      "known_cost_subtotal_usd": 0.003184104,
      "cost_missing_n": 0,
      "total_cost_usd": 0.003184104,
      "mean_cost_usd": 0.000132671,
      "cache_observed_n": 0,
      "cache_hit_calls": 0,
      "cache_read_tokens": null,
      "known_cache_read_subtotal_tokens": 0,
      "cache_write_observed_n": 0,
      "cache_write_tokens": null,
      "known_cache_write_subtotal_tokens": 0,
      "prompt_tokens": null,
      "cache_read_fraction": null
    }
  },
  "paired": {
    "Luna cached / Luna uncached": {
      "n": 24,
      "median_latency_ratio": 0.9125562576625701,
      "total_cost_ratio": 0.2383387649369098,
      "label_agreement": 0.9166666666666666
    },
    "Luna cached / Jev": {
      "n": 24,
      "median_latency_ratio": 3.617968437782495,
      "total_cost_ratio": 0.5374573192332913,
      "label_agreement": 0.875
    },
    "Luna uncached / Jev": {
      "n": 24,
      "median_latency_ratio": 3.918573737604335,
      "total_cost_ratio": 2.255014283453053,
      "label_agreement": 0.875
    }
  },
  "cache_verification": {
    "cached_arm_has_observed_hits": true,
    "uncached_control_verified": true,
    "controlled_cache_comparison_verified": true,
    "status": "observed cache hits and clean uncached control",
    "prime_cached_tokens": 0,
    "prime_cache_write_tokens": 2533,
    "prime_cost_usd": 0.000362525,
    "prime_latency_s": 1.2044517089962028,
    "prime_was_observed_cold": true,
    "setup_inclusive_cached_cost_usd": 0.002073845,
    "setup_inclusive_cost_per_measured_case_usd": 8.641020833333333e-05
  }
}