{ "cumulative": { "basis": "All seven retained runs, including smoke, development, both baseline repetitions and actual cascades; every physical run counted once", "calls": 268, "cost_usd": 0.190884758, "exclusions": "Unlogged setup/authentication probes and internal CLI HTTP attempts are not measured. Counterfactual routes do not subtract actual calls.", "jev": { "calls": 154, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.015243438 }, "cost_usd": 0.015243438, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 362939, "output_tokens": 18018, "uncached_input_tokens": 362939 } }, "luna": { "calls": 114, "cost_components_usd": { "cached_input": 0.00904192, "output": 0.0336912, "uncached_input": 0.1329082 }, "cost_usd": 0.17564132, "token_ledger": { "cached_input_tokens": 452096, "input_tokens": 1116637, "output_tokens": 28076, "uncached_input_tokens": 664541 } }, "run_count": 7, "runs": [ { "calls": 12, "cost_usd": 0.013875022, "id": "dev-01", "jev": { "calls": 6, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.000562422 }, "cost_usd": 0.000562422, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 13391, "output_tokens": 702, "uncached_input_tokens": 13391 } }, "luna": { "calls": 6, "cost_components_usd": { "cached_input": 0, "output": 0.0015696, "uncached_input": 0.011743 }, "cost_usd": 0.0133126, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 58715, "output_tokens": 1308, "uncached_input_tokens": 58715 } } }, { "calls": 72, "cost_usd": 0.064609916, "id": "pilot-01", "jev": { "calls": 36, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.003464076 }, "cost_usd": 0.003464076, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 82478, "output_tokens": 4212, "uncached_input_tokens": 82478 } }, "luna": { "calls": 36, "cost_components_usd": { "cached_input": 0.00203264, "output": 0.009462, "uncached_input": 0.0496512 }, "cost_usd": 0.06114584, "token_ledger": { "cached_input_tokens": 101632, "input_tokens": 349888, "output_tokens": 7885, "uncached_input_tokens": 248256 } } }, { "calls": 16, "cost_usd": 0.001675464, "id": "trace-dev-01", "jev": { "calls": 16, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.001675464 }, "cost_usd": 0.001675464, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 39892, "output_tokens": 1872, "uncached_input_tokens": 39892 } }, "luna": { "calls": 0, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0 }, "cost_usd": 0, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 0, "output_tokens": 0, "uncached_input_tokens": 0 } } }, { "calls": 64, "cost_usd": 0.040887056, "id": "trace-eval-01", "jev": { "calls": 32, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.003041136 }, "cost_usd": 0.003041136, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 72408, "output_tokens": 3744, "uncached_input_tokens": 72408 } }, "luna": { "calls": 32, "cost_components_usd": { "cached_input": 0.00387072, "output": 0.0102732, "uncached_input": 0.023702 }, "cost_usd": 0.03784592, "token_ledger": { "cached_input_tokens": 193536, "input_tokens": 312046, "output_tokens": 8561, "uncached_input_tokens": 118510 } } }, { "calls": 22, "cost_usd": 0.008862768, "id": "trace-cascade-01", "jev": { "calls": 16, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.001520568 }, "cost_usd": 0.001520568, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 36204, "output_tokens": 1872, "uncached_input_tokens": 36204 } }, "luna": { "calls": 6, "cost_components_usd": { "cached_input": 0.0006912, "output": 0.0018504, "uncached_input": 0.0048006 }, "cost_usd": 0.0073422, "token_ledger": { "cached_input_tokens": 34560, "input_tokens": 58563, "output_tokens": 1542, "uncached_input_tokens": 24003 } } }, { "calls": 64, "cost_usd": 0.056278768, "id": "continuation-eval-01", "jev": { "calls": 32, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.003319848 }, "cost_usd": 0.003319848, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 79044, "output_tokens": 3744, "uncached_input_tokens": 79044 } }, "luna": { "calls": 32, "cost_components_usd": { "cached_input": 0.00230912, "output": 0.010128, "uncached_input": 0.0405218 }, "cost_usd": 0.05295892, "token_ledger": { "cached_input_tokens": 115456, "input_tokens": 318065, "output_tokens": 8440, "uncached_input_tokens": 202609 } } }, { "calls": 18, "cost_usd": 0.004695764, "id": "continuation-cascade-01", "jev": { "calls": 16, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.001659924 }, "cost_usd": 0.001659924, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 39522, "output_tokens": 1872, "uncached_input_tokens": 39522 } }, "luna": { "calls": 2, "cost_components_usd": { "cached_input": 0.00013824, "output": 0.000408, "uncached_input": 0.0024896 }, "cost_usd": 0.00303584, "token_ledger": { "cached_input_tokens": 6912, "input_tokens": 19360, "output_tokens": 340, "uncached_input_tokens": 12448 } } } ] }, "limitations": [ "Small constructed corpora with correlated variants; not production accuracy or clinical-safety validation.", "Initial pilot labels were author assigned; later labels were blinded model reviewed and agent adjudicated, not independent human ground truth.", "Continuation contains eight archived-trace-derived cases from one execution and eight authenticated-source scenario-derived simulations.", "Quality is repetition zero only; repeat calls measure stability and resources, not additional independent cases.", "Latency includes different client paths: HTTP for Jev and CLI startup for Luna. It is not pure server inference latency.", "First-pass modeled costs use recorded cache utilization; a different cache pattern changes the comparison.", "Do not pool quality across phases or confuse reduced frontier calls with reduced total calls." ], "model_identity": { "jev": "jev-1.13.0 requested and resolved in retained records", "luna": "gpt-5.6-luna requested through Codex CLI, medium effort; resolved snapshot not attested by CLI JSONL" }, "phases": [ { "arms": [ { "all_repetitions": null, "first_pass": null, "id": "deterministic", "label": "Recorded checks", "latency_seconds": { "basis": "All two baseline repetitions; adapter wall time", "median": 0.000001187509042210877, "p95": 0.000002732950088102369, "sample_count": 24 }, "quality": { "defect_detection_denominator": 6, "exact_label_cases": null, "false_alarm_denominator": 6, "fn": 5, "fp": 0, "scored_cases": 12, "tn": 6, "tp": 1 } }, { "all_repetitions": { "calls": 24, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.002318568 }, "cost_usd": 0.002318568, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 55204, "output_tokens": 2808, "uncached_input_tokens": 55204 } }, "first_pass": { "calls": 12, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.001159284 }, "cost_usd": 0.001159284, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 27602, "output_tokens": 1404, "uncached_input_tokens": 27602 } }, "id": "jev", "label": "Jev", "latency_seconds": { "basis": "All two baseline repetitions; adapter wall time", "median": 0.3659076254843967, "p95": 0.541127828505705, "sample_count": 24 }, "quality": { "defect_detection_denominator": 6, "exact_label_cases": 9, "false_alarm_denominator": 6, "fn": 0, "fp": 0, "scored_cases": 12, "tn": 6, "tp": 6 } }, { "all_repetitions": { "calls": 24, "cost_components_usd": { "cached_input": 0.0012032, "output": 0.0065136, "uncached_input": 0.034634 }, "cost_usd": 0.0423508, "token_ledger": { "cached_input_tokens": 60160, "input_tokens": 233330, "output_tokens": 5428, "uncached_input_tokens": 173170 } }, "first_pass": { "calls": 12, "cost_components_usd": { "cached_input": 0.000512, "output": 0.0035988, "uncached_input": 0.018215 }, "cost_usd": 0.0223258, "token_ledger": { "cached_input_tokens": 25600, "input_tokens": 116675, "output_tokens": 2999, "uncached_input_tokens": 91075 } }, "id": "luna", "label": "GPT-5.6 Luna", "latency_seconds": { "basis": "All two baseline repetitions; adapter wall time", "median": 10.355938354507089, "p95": 15.135264464300416, "sample_count": 24 }, "quality": { "defect_detection_denominator": 6, "exact_label_cases": 11, "false_alarm_denominator": 6, "fn": 0, "fp": 0, "scored_cases": 12, "tn": 6, "tp": 6 } } ], "baseline_repetitions": 2, "baseline_run": "pilot-01", "cascade": null, "clean_case_count": 6, "defect_case_count": 6, "eval_case_count": 12, "family_count": 6, "id": "synthetic-pilot", "label": "Authored pilot", "label_provenance": "Author-assigned synthetic labels; not independent human ground truth. Each mutation is designed to introduce one named defect, and every control has all labels false. Pair families are held together within dev or eval. These small constructed cases support harness and sensitivity research, not production accuracy or clinical-safety claims.", "quality_basis": "Evaluation repetition zero; no pooling across corpora" }, { "arms": [ { "all_repetitions": null, "first_pass": null, "id": "deterministic", "label": "Recorded checks", "latency_seconds": { "basis": "All two baseline repetitions; adapter wall time", "median": 0.0000017915153875946999, "p95": 0.0000031267481972463425, "sample_count": 32 }, "quality": { "defect_detection_denominator": 9, "exact_label_cases": null, "false_alarm_denominator": 7, "fn": 9, "fp": 0, "scored_cases": 16, "tn": 7, "tp": 0 } }, { "all_repetitions": { "calls": 32, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.003041136 }, "cost_usd": 0.003041136, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 72408, "output_tokens": 3744, "uncached_input_tokens": 72408 } }, "first_pass": { "calls": 16, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.001520568 }, "cost_usd": 0.001520568, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 36204, "output_tokens": 1872, "uncached_input_tokens": 36204 } }, "id": "jev", "label": "Jev", "latency_seconds": { "basis": "All two baseline repetitions; adapter wall time", "median": 0.22223570849746466, "p95": 0.5654019812456681, "sample_count": 32 }, "quality": { "defect_detection_denominator": 9, "exact_label_cases": 6, "false_alarm_denominator": 7, "fn": 0, "fp": 4, "scored_cases": 16, "tn": 3, "tp": 9 } }, { "all_repetitions": { "calls": 32, "cost_components_usd": { "cached_input": 0.00387072, "output": 0.0102732, "uncached_input": 0.023702 }, "cost_usd": 0.03784592, "token_ledger": { "cached_input_tokens": 193536, "input_tokens": 312046, "output_tokens": 8561, "uncached_input_tokens": 118510 } }, "first_pass": { "calls": 16, "cost_components_usd": { "cached_input": 0.00179712, "output": 0.0048768, "uncached_input": 0.0132328 }, "cost_usd": 0.01990672, "token_ledger": { "cached_input_tokens": 89856, "input_tokens": 156020, "output_tokens": 4064, "uncached_input_tokens": 66164 } }, "id": "luna", "label": "GPT-5.6 Luna", "latency_seconds": { "basis": "All two baseline repetitions; adapter wall time", "median": 10.566153770501842, "p95": 19.294175785768306, "sample_count": 32 }, "quality": { "defect_detection_denominator": 9, "exact_label_cases": 13, "false_alarm_denominator": 7, "fn": 0, "fp": 1, "scored_cases": 16, "tn": 6, "tp": 9 } } ], "baseline_repetitions": 2, "baseline_run": "trace-eval-01", "cascade": { "cost_comparison": { "basis": "Same evaluation cases; baseline repetition zero versus fresh actual cascade", "cascade_first_pass_usd": 0.008862768, "luna_first_pass_usd": 0.01990672, "reduction_percent": 55.47851177893697 }, "first_pass": { "calls": 22, "cost_usd": 0.008862768, "jev": { "calls": 16, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.001520568 }, "cost_usd": 0.001520568, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 36204, "output_tokens": 1872, "uncached_input_tokens": 36204 } }, "luna": { "calls": 6, "cost_components_usd": { "cached_input": 0.0006912, "output": 0.0018504, "uncached_input": 0.0048006 }, "cost_usd": 0.0073422, "token_ledger": { "cached_input_tokens": 34560, "input_tokens": 58563, "output_tokens": 1542, "uncached_input_tokens": 24003 } } }, "jev_calls": 16, "latency_seconds": { "basis": "Fresh one-pass cascade; end-to-end wall time", "median": 0.44696777100034524, "p95": 13.567006301731453, "sample_count": 16 }, "luna_call_reduction_percent": 62.5, "luna_calls": 6, "policy_sha256": "02b3c7f3bc880be1993923a006c5a9bc9f92da23551d3733f22790a51804e14a", "quality": { "defect_detection_denominator": 9, "exact_label_cases": null, "false_alarm_denominator": 7, "fn": 0, "fp": 1, "scored_cases": 16, "tn": 6, "tp": 9 }, "routes": { "frontier-clear-for-sampling": 4, "frontier-defect": 2, "jev-clear-for-sampling": 2, "jev-defect": 8 }, "run": "trace-cascade-01" }, "clean_case_count": 7, "defect_case_count": 9, "eval_case_count": 16, "family_count": 8, "id": "trace-pilot", "label": "Trace-derived evaluation", "label_provenance": "Reference labels finalized by state-only blinded model review and primary research-agent adjudication before benchmark inference, not independent human ground truth. The persistent overlay records the review, rationales, ambiguous cases, and all 24 final labels; original author labels remain in the review mapping. Historical checks are excerpted observations, not fresh validator execution. Observed/counterfactual provenance remains unchanged: altered or inserted events did not occur in the archived run.", "quality_basis": "Evaluation repetition zero; no pooling across corpora" }, { "arms": [ { "all_repetitions": null, "first_pass": null, "id": "deterministic", "label": "Recorded checks", "latency_seconds": { "basis": "All two baseline repetitions; adapter wall time", "median": 0.0000014169927453622222, "p95": 0.00000312135525746271, "sample_count": 32 }, "quality": { "defect_detection_denominator": 6, "exact_label_cases": null, "false_alarm_denominator": 10, "fn": 6, "fp": 0, "scored_cases": 16, "tn": 10, "tp": 0 } }, { "all_repetitions": { "calls": 32, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.003319848 }, "cost_usd": 0.003319848, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 79044, "output_tokens": 3744, "uncached_input_tokens": 79044 } }, "first_pass": { "calls": 16, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.001659924 }, "cost_usd": 0.001659924, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 39522, "output_tokens": 1872, "uncached_input_tokens": 39522 } }, "id": "jev", "label": "Jev", "latency_seconds": { "basis": "All two baseline repetitions; adapter wall time", "median": 0.18336806249863002, "p95": 0.4033085418428527, "sample_count": 32 }, "quality": { "defect_detection_denominator": 6, "exact_label_cases": 13, "false_alarm_denominator": 10, "fn": 0, "fp": 0, "scored_cases": 16, "tn": 10, "tp": 6 } }, { "all_repetitions": { "calls": 32, "cost_components_usd": { "cached_input": 0.00230912, "output": 0.010128, "uncached_input": 0.0405218 }, "cost_usd": 0.05295892, "token_ledger": { "cached_input_tokens": 115456, "input_tokens": 318065, "output_tokens": 8440, "uncached_input_tokens": 202609 } }, "first_pass": { "calls": 16, "cost_components_usd": { "cached_input": 0.00078848, "output": 0.0046896, "uncached_input": 0.0239878 }, "cost_usd": 0.02946588, "token_ledger": { "cached_input_tokens": 39424, "input_tokens": 159363, "output_tokens": 3908, "uncached_input_tokens": 119939 } }, "id": "luna", "label": "GPT-5.6 Luna", "latency_seconds": { "basis": "All two baseline repetitions; adapter wall time", "median": 7.565988978982205, "p95": 17.247781222805497, "sample_count": 32 }, "quality": { "defect_detection_denominator": 6, "exact_label_cases": 13, "false_alarm_denominator": 10, "fn": 0, "fp": 0, "scored_cases": 16, "tn": 10, "tp": 6 } } ], "baseline_repetitions": 2, "baseline_run": "continuation-eval-01", "cascade": { "cost_comparison": { "basis": "Same evaluation cases; baseline repetition zero versus fresh actual cascade", "cascade_first_pass_usd": 0.004695764, "luna_first_pass_usd": 0.02946588, "reduction_percent": 84.06372387317127 }, "first_pass": { "calls": 18, "cost_usd": 0.004695764, "jev": { "calls": 16, "cost_components_usd": { "cached_input": 0, "output": 0, "uncached_input": 0.001659924 }, "cost_usd": 0.001659924, "token_ledger": { "cached_input_tokens": 0, "input_tokens": 39522, "output_tokens": 1872, "uncached_input_tokens": 39522 } }, "luna": { "calls": 2, "cost_components_usd": { "cached_input": 0.00013824, "output": 0.000408, "uncached_input": 0.0024896 }, "cost_usd": 0.00303584, "token_ledger": { "cached_input_tokens": 6912, "input_tokens": 19360, "output_tokens": 340, "uncached_input_tokens": 12448 } } }, "jev_calls": 16, "latency_seconds": { "basis": "Fresh one-pass cascade; end-to-end wall time", "median": 0.1757482914981665, "p95": 7.155920385754143, "sample_count": 16 }, "luna_call_reduction_percent": 87.5, "luna_calls": 2, "policy_sha256": "02b3c7f3bc880be1993923a006c5a9bc9f92da23551d3733f22790a51804e14a", "quality": { "defect_detection_denominator": 6, "exact_label_cases": null, "false_alarm_denominator": 10, "fn": 0, "fp": 0, "scored_cases": 16, "tn": 10, "tp": 6 }, "routes": { "frontier-clear-for-sampling": 2, "jev-clear-for-sampling": 8, "jev-defect": 6 }, "run": "continuation-cascade-01" }, "clean_case_count": 10, "defect_case_count": 6, "eval_case_count": 16, "family_count": 2, "id": "continuation-challenge", "label": "Continuation challenge", "label_provenance": "Reference labels finalized through persistent blinded model review and primary adjudication before benchmark inference; not independent human ground truth. Original author labels remain in the review mapping, and per-case hashes link the final overlay.", "quality_basis": "Evaluation repetition zero; no pooling across corpora" } ], "policy": { "calibration": "Eight development cases, two Jev repetitions; frozen before trace evaluation, reused unchanged for continuation", "classification_threshold": 0.5, "clear_max_score": 0.25, "defect_min_score": 0.9, "otherwise": "Escalate to Luna" }, "pricing": { "cache_note": "Recorded cached input is a subset of input. Cache quantities differ between independent baseline and cascade passes.", "formula": "(input - cached input) * uncached rate + cached input * cached rate + output * output rate; divided by 1,000,000", "interpretation": "Modeled API-equivalent token costs, not an observed subscription invoice or cash savings", "jev_cache_note": "Jev has one input rate; no separate cached-input rate is applied.", "jev_price_checked_in_research": "2026-09-21", "jev_source": "https://docs.typesafe.ai/models", "luna_source": "User-supplied comparison assumption, not independently verified current pricing", "output_note": "Recorded output tokens already include reasoning tokens; do not add reasoning_output_tokens again", "rates": { "jev": { "input": 0.042, "output": 0 }, "luna": { "cached_input": 0.02, "output": 1.2, "uncached_input": 0.2 } } }, "provenance": { "catalog": { "path": "results/catalog-after-continuation-01/catalog.json", "sha256": "59dd5336a7bf57e4986d07ee3fa368f2c190d6591b517be587ca5b83289cc042" }, "generator": "scripts/build-jev-blog-data.mjs", "inputs": [ { "dataset_canonical_sha256": "e9e5d4e5b85435f5fc2bdabefae259d7595c159e56d91bfaefac06c0f611c3cc", "files_sha256": { "dataset.json": "401418660cd60c0684235ce6d9d809d5785e617b68cf3e0efd1ccea4a9917be3", "manifest.json": "b44eaa657182106cd34d02309fe77ca4286dcaa6a5cd3abcd58937c62bf39999", "report.md": "d16dcd310945268820521196b812caefd0384ee8fc9d8f7ae50a31e7ccd8302b", "results.jsonl": "9e7307d0021f01e81c3af1234f5ae34be206da5edc3a59a47bbb0c4cdd7600cb", "rubric.json": "7ed18552bb89f704c2905dcf31d4fddd377c4da451b6ade60bad1128776044a4", "summary.json": "31ca00449e09190033685a7ffab70022607de825adf1f8ce929fa3830cd3a5c6" }, "legacy_missing_metrics_source_hash": true, "rubric_canonical_sha256": "c8c5887dcd1f39ae23fae2db5dca1f44e3733b71336e08ff259934ee4a37fe75", "run": "dev-01" }, { "dataset_canonical_sha256": "52b2048beab8de92ab72d13bbcfeeb5339133cbf24b31e69a5daabeca4c393d0", "files_sha256": { "dataset.json": "cc367982bf895010a14cc9af9f1a664516078704f6240a2b4dda6b22f89da085", "manifest.json": "18bef3431dd50b48e621e492f19afa39de34d64b71eed3c34cdf21cf0013c4cf", "report.md": "d1f12b6e4da21905a04cf74c130c4bd37c316b3d328876061cbdaa0b60bb4d99", "results.jsonl": "62a775a0bc4626265a1a6457cb7576ed88502b722ecb21daeabc74e2e860a9ab", "rubric.json": "08cede55a88918be44b076e448e2a824288431a2a7159343727a2179210e3e70", "summary.json": "6d72bf3ac1e83f9b939fe700aa64183de069f498d0831c7ac935a62444f2358a" }, "legacy_missing_metrics_source_hash": false, "rubric_canonical_sha256": "4004334efcb8c38574b7cf7ce1603d1807ebad04ac6f3283e0a350f4c01a44ea", "run": "pilot-01" }, { "dataset_canonical_sha256": "be00abb70508ddf7dbb9801ec24d15f9d744f02c8150caec42a401feaedb7a6c", "files_sha256": { "dataset.json": "2fe49faf40806c39a4cd19ec6a61d34be5968b2955e1b4399ec7661552ed6969", "manifest.json": "9b63d0bb53ad7924f8f297b49feddfe09261652c1e83ae4cc208fbcb15844430", "report.md": "84763601c58262022f57132b878481d1a2178447bd2a2b58b7c8238ede0b39c4", "results.jsonl": "e23ff83a1494e7f517ef00a85b0ff8a3a515752b5c0f1cd404ff298faf6928ce", "rubric.json": "08cede55a88918be44b076e448e2a824288431a2a7159343727a2179210e3e70", "summary.json": "d56a08de43f80c838f003a615fa225e599aeebf344e487776be0bf3b93c9d838" }, "legacy_missing_metrics_source_hash": false, "rubric_canonical_sha256": "4004334efcb8c38574b7cf7ce1603d1807ebad04ac6f3283e0a350f4c01a44ea", "run": "trace-dev-01" }, { "dataset_canonical_sha256": "fed95209b897f40d354a141803c684d1c3fa68e413748233b7c9c4c0f277f5ab", "files_sha256": { "dataset.json": "f56aac325eec3215a934f5742f44fbb59633608246dca27521475c46329f3148", "manifest.json": "18d24dedfe1098f3215fd33d1c31b6bd03198ab024e6fea1f00dbe57a3dd12ec", "report.md": "5d0f32e093eddfd741a5ea8270c2117494e7f9bf0ef6c9bc1d823a5e4cc0f442", "results.jsonl": "1b6fea68c1982db29775a5ff13eb559d12509570f54477ffe30c923b38e4e211", "rubric.json": "08cede55a88918be44b076e448e2a824288431a2a7159343727a2179210e3e70", "summary.json": "a25f5e44491997c22393ec32b2b3a216a4f8d7bd4e105a7638c504bebe8ab4e9" }, "legacy_missing_metrics_source_hash": false, "rubric_canonical_sha256": "4004334efcb8c38574b7cf7ce1603d1807ebad04ac6f3283e0a350f4c01a44ea", "run": "trace-eval-01" }, { "dataset_canonical_sha256": "fed95209b897f40d354a141803c684d1c3fa68e413748233b7c9c4c0f277f5ab", "files_sha256": { "dataset.json": "f56aac325eec3215a934f5742f44fbb59633608246dca27521475c46329f3148", "manifest.json": "db747eeaee36e25d0475650a7a35082a6197c775aef98a6ce3b8d334a4bad668", "policy.json": "0ba7101c17601c289a05b8149a64092614d8d115cfbd72b0f6391c2863336b44", "report.md": "baa1624a0da33c0e092f37369bbd7e0a434f9115ddbbbd40c1ec18a93cffee6d", "results.jsonl": "6f647b14b83327dff26849fd40ebf752bed936e54d0e003d7c0e75004a54bf1c", "rubric.json": "08cede55a88918be44b076e448e2a824288431a2a7159343727a2179210e3e70", "summary.json": "34ae97f8f78beda928277f0a2182e300be35f277e02acc0b25deb2317ad6a072" }, "legacy_missing_metrics_source_hash": false, "rubric_canonical_sha256": "4004334efcb8c38574b7cf7ce1603d1807ebad04ac6f3283e0a350f4c01a44ea", "run": "trace-cascade-01" }, { "dataset_canonical_sha256": "ac1ce60922da9afa16e72c39db6a24968d278f4ecf2c1cc549639f0c562584c0", "files_sha256": { "dataset.json": "a71268428fdcb6101f1ed8594be36eccbf85145ba80116b72373cac17a10b605", "manifest.json": "974b204e54cc081e2ab356f23b3f1421e7d5d2439b62a10f8c9ab2ddbefbca51", "report.md": "0d0b3db3f7b41371746d9ca04a24da1040bf116c7188eda7eebc95f17da4a48c", "results.jsonl": "219aef7acc50b961dd78d6250fbe65bbfc31a5acd98ca5e6ed11df83737e26fe", "rubric.json": "08cede55a88918be44b076e448e2a824288431a2a7159343727a2179210e3e70", "summary.json": "b3bf376997ac080134800a2b925d42dc2bd36a133f49ecac070bb078894cc45f" }, "legacy_missing_metrics_source_hash": false, "rubric_canonical_sha256": "4004334efcb8c38574b7cf7ce1603d1807ebad04ac6f3283e0a350f4c01a44ea", "run": "continuation-eval-01" }, { "dataset_canonical_sha256": "ac1ce60922da9afa16e72c39db6a24968d278f4ecf2c1cc549639f0c562584c0", "files_sha256": { "dataset.json": "a71268428fdcb6101f1ed8594be36eccbf85145ba80116b72373cac17a10b605", "manifest.json": "a1b5f73c4dce662812cb60274159dfe0e9668731aad23f60640cc9c1118a1c2e", "policy.json": "0ba7101c17601c289a05b8149a64092614d8d115cfbd72b0f6391c2863336b44", "report.md": "b8d48c7f92016fa2a076fe88ae1a8b49c25dc05b46d9b4b2fae26c36d08e45d6", "results.jsonl": "f9fa86586f5dc8287e41fb14fb6ea5249320dd59cc48bee497ea65ca83fd0c0f", "rubric.json": "08cede55a88918be44b076e448e2a824288431a2a7159343727a2179210e3e70", "summary.json": "dae01e5eb9eae9da7798ea0dc98b87e494ee53585911f4ab6359432b0d77bc30" }, "legacy_missing_metrics_source_hash": false, "rubric_canonical_sha256": "4004334efcb8c38574b7cf7ce1603d1807ebad04ac6f3283e0a350f4c01a44ea", "run": "continuation-cascade-01" } ], "privacy": "Aggregated measurements and evidence hashes only; no prompts, credentials, local filesystem paths, or raw source transcripts" }, "schema_version": 1, "study_date": "2026-09-21", "title": "Jev in the rh-skills framework: retained research measurements", "units": { "cost": "USD", "latency": "seconds", "rates": "USD per million tokens" } }