bench dense arm · model google/embeddinggemma-300m · split selection · code 2077f098670a · prompts 'task: search result | query: {}' / 'title: none | text: {}'
corpus: 182 live records, 2604 vectors (title+symptom, aliases, triggers on 173 lessons) · queries sha 97ecbcb5d681 · nulls 550: 282 calibrate, 268 test
cost: cold load in a fresh process 2.05 s · load here 1.44 s · index 285.4 s · encode per prompt median 184.2 ms
decision score = shared_words + W * maxsim; line at alpha 0.05 by split conformal

    W    line cap |           recall [95%] briefed/q | null test briefed anything [95%] | paraphrase
  8.0   5.755   3 |   90.4% [86.7%, 94.0%]      2.80 | 12 of 268 (4.5%, [2.6%, 7.7%]) | 68/83
  8.0   5.755   5 |   92.4% [88.8%, 95.6%]      4.39 | 12 of 268 (4.5%, [2.6%, 7.7%]) | 72/83
 12.0   7.848   3 |   90.0% [86.3%, 93.6%]      2.65 | 13 of 268 (4.9%, [2.9%, 8.1%]) | 67/83
 12.0   7.848   5 |   91.2% [87.6%, 94.8%]      3.90 | 13 of 268 (4.9%, [2.9%, 8.1%]) | 69/83
 16.0   9.854   3 |   88.4% [83.9%, 92.4%]      2.51 | 14 of 268 (5.2%, [3.1%, 8.6%]) | 67/83
 16.0   9.854   5 |   89.6% [85.5%, 93.6%]      3.52 | 14 of 268 (5.2%, [3.1%, 8.6%]) | 69/83
 24.0  14.158   3 |   85.5% [80.3%, 90.0%]      2.16 | 12 of 268 (4.5%, [2.6%, 7.7%]) | 62/83
 24.0  14.158   5 |   86.3% [81.5%, 90.8%]      2.81 | 12 of 268 (4.5%, [2.6%, 7.7%]) | 63/83
 32.0  18.506   3 |   83.5% [77.5%, 88.4%]      1.92 | 12 of 268 (4.5%, [2.6%, 7.7%]) | 60/83
 32.0  18.506   5 |   83.9% [77.9%, 88.8%]      2.34 | 12 of 268 (4.5%, [2.6%, 7.7%]) | 61/83
