File size: 1,483 Bytes
7edf561 c484dad 7edf561 c484dad 7edf561 c484dad 7edf561 c484dad 7edf561 c484dad 7edf561 c484dad 7edf561 c484dad 7edf561 c484dad 7edf561 c484dad 7edf561 | 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 | {
"artifact_sha256": "2dfd7463ae0a8ca7dd2e4abebeed9cc77a584569b9c99dc9e68ad85f1399b55a",
"quantization": "Q8_0",
"scope": "ScribeKey app-boundary, paired against ScribeKey Fast v1 with identical controls",
"controls": {
"styling": "semi-casual",
"structure": "prose",
"context": "general"
},
"benchmark": {
"name": "ScribeKey holdout-v1",
"scored_rows": 7516,
"total_rows": 7524,
"groups_resampled": true,
"hidden_rows_distributed": false
},
"metrics_vs_v1": {
"self_correction_ter": [
-0.036,
-0.046,
-0.026
],
"self_correction_ter_questions": [
-0.048,
-0.065,
-0.03
],
"self_correction_ter_written": [
-0.023,
-0.032,
-0.014
],
"exact_pp": [
3.3,
2.4,
4.3
],
"clean_input_damage_pp": [
-0.9,
-2.1,
0.0
],
"new_values_pp": [
-0.3,
-0.7,
0.1
]
},
"metric_format": "[difference, 95% lower, 95% upper], v1.1 minus v1",
"gates": "pre-registered: correction TER upper < 0, exact lower > -1 pt, new values upper <= +0.5 pt; all pass",
"bf16_quality_benchmark": "not measured",
"final_artifact_phone": "not measured",
"midrange_phone": "not measured",
"limitations": [
"model can change or omit meaning",
"list recall low",
"source/data provenance inherited in part",
"phone latency assumed from v1's identical architecture and size, not measured"
]
}
|