Datasets:
File size: 4,462 Bytes
4a0d755 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 | {
"experiment": "aid-definitions-v1",
"status": "prepared_not_run",
"dataset": "aid-test-2000",
"dataset_version": "1000198073523310044",
"case_count": 2000,
"case_manifest_sha256": "4d24a4e088d564d32e8ec16869ef6f82b224d67a5aa45068f2a9efc4efe32c0f",
"class_counts": {
"Airport": 72,
"BareLand": 62,
"BaseballField": 44,
"Beach": 80,
"Bridge": 72,
"Center": 52,
"Church": 48,
"Commercial": 70,
"DenseResidential": 82,
"Desert": 60,
"Farmland": 74,
"Forest": 50,
"Industrial": 78,
"Meadow": 56,
"MediumResidential": 58,
"Mountain": 68,
"Park": 70,
"Parking": 78,
"Playground": 74,
"Pond": 84,
"Port": 76,
"RailwayStation": 52,
"Resort": 58,
"River": 82,
"School": 60,
"SparseResidential": 60,
"Square": 66,
"Stadium": 58,
"StorageTanks": 72,
"Viaduct": 84
},
"labels": [
"Airport",
"BareLand",
"BaseballField",
"Beach",
"Bridge",
"Center",
"Church",
"Commercial",
"DenseResidential",
"Desert",
"Farmland",
"Forest",
"Industrial",
"Meadow",
"MediumResidential",
"Mountain",
"Park",
"Parking",
"Playground",
"Pond",
"Port",
"RailwayStation",
"Resort",
"River",
"School",
"SparseResidential",
"Square",
"Stadium",
"StorageTanks",
"Viaduct"
],
"label_order": "Original alphabetical order, unchanged",
"models": [
"luna",
"jev",
"clef",
"d1",
"d1-ultra"
],
"trials": 1,
"max_concurrency_per_model": 6,
"planned_predictions": 10000,
"retry_policy": "Inherited: up to 3 attempts",
"prompt_file": "prompt.txt",
"prompt_sha256": "12e2b2b58a973300a03c8c307366cefa37488dce365152bb2852d08115a5ea1b",
"baseline_prompt_sha256": "73c54a1d7852ebc4bf18c2363e47cc676af7e7d3580a18c4b667f91c48e9a5b3",
"prompt_words": 559,
"prompt_characters": 3923,
"image_policy": "Original image bytes, detected MIME type; no client resizing; provider preprocessing applies",
"sole_planned_input_change": "Append identical class descriptions to the original instruction for all five models; choice values and descriptions remain the original class names.",
"sources_and_assumptions": "class-guide.json",
"source_versions_sha256": {
"native_eval.py": "816fce1151a70cfd412a221fd79311956c7aacaba3623566bbf0943a824cc5ec",
"eval.py": "8d83811e75d170ca839732a6665e0d4f8a50237ae80d7c8a43caf71eee865cc9",
"cost_accounting.py": "800c448c10683d7efcf8c34f45c3917ad87cc09ad33274ac4e7347793c0c1494",
"config.json": "f4711af0e40718d5919b14b1507775c57520ff3b3116655c7322fdebd27c11f0"
},
"baseline_runs": {
"luna": "luna.jsonl",
"jev": "jev.jsonl",
"clef": "clef.jsonl",
"d1": "d1.jsonl",
"d1-ultra": "d1-ultra.jsonl"
},
"analysis_plan": [
"Preserve original predictions, ground-truth labels, and published baseline plots.",
"Compare per-model accuracy on all 2000 cases, valid-output coverage, macro F1 with its denominator stated, class recalls, and full confusion matrices.",
"Report paired counts of baseline-wrong/new-correct and baseline-correct/new-wrong; use a paired bootstrap for accuracy-difference intervals.",
"Report prompt token usage, actual available provider cost, separately labeled token cost estimates, latency, and requested/served model versions.",
"Inspect remaining shared errors without treating model agreement as proof of annotation error.",
"Do not select prompt revisions by their performance on these 2000 cases. Any later changes require a new version and separate development data."
],
"interpretation_limits": [
"Exploratory follow-up designed after reviewing baseline test errors, not a pristine confirmatory test. No case-specific labels or examples are included in the prompt.",
"Working definitions may introduce bias, especially Center and residential density. Improvement would show sensitivity to category wording, not prove that removed errors were dataset ambiguity.",
"The historical baseline and a later rerun can differ because of provider/model changes. A fresh baseline-control run would be needed for stronger causal attribution.",
"One trial does not measure run-to-run variability.",
"The copied price table is historical. Refresh price evidence before publishing a new cost comparison; do not silently present token estimates as current billed costs."
]
}
|