bench dense arm · model google/embeddinggemma-300m-quantized · split selection · code c1cba7f29e88 · 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 1.92 s · load here 0.93 s · index 253.3 s · encode per prompt median 170.7 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
  0.0   2.002   3 |   86.3% [81.9%, 90.4%]      2.89 | 9 of 268 (3.4%, [1.8%, 6.3%]) | 63/83
  0.0   2.002   5 |   88.0% [83.9%, 91.6%]      4.66 | 9 of 268 (3.4%, [1.8%, 6.3%]) | 64/83
  1.0   2.489   3 |   89.2% [85.1%, 92.8%]      2.90 | 10 of 268 (3.7%, [2.0%, 6.7%]) | 66/83
  1.0   2.489   5 |   90.8% [86.7%, 94.4%]      4.68 | 10 of 268 (3.7%, [2.0%, 6.7%]) | 68/83
  2.0   2.977   3 |   89.2% [85.1%, 92.8%]      2.90 | 10 of 268 (3.7%, [2.0%, 6.7%]) | 66/83
  2.0   2.977   5 |   90.8% [86.7%, 94.4%]      4.68 | 10 of 268 (3.7%, [2.0%, 6.7%]) | 68/83
  4.0   3.953   3 |   89.2% [85.1%, 92.8%]      2.90 | 10 of 268 (3.7%, [2.0%, 6.7%]) | 66/83
  4.0   3.953   5 |   90.8% [86.7%, 94.4%]      4.65 | 10 of 268 (3.7%, [2.0%, 6.7%]) | 68/83
  8.0   5.768   3 |   89.6% [85.5%, 93.2%]      2.82 | 12 of 268 (4.5%, [2.6%, 7.7%]) | 66/83
  8.0   5.768   5 |   92.0% [88.4%, 95.6%]      4.43 | 12 of 268 (4.5%, [2.6%, 7.7%]) | 71/83
 12.0   7.856   3 |   90.8% [87.1%, 94.4%]      2.68 | 14 of 268 (5.2%, [3.1%, 8.6%]) | 69/83
 12.0   7.856   5 |   91.6% [88.0%, 94.8%]      4.00 | 14 of 268 (5.2%, [3.1%, 8.6%]) | 70/83

exact lines: W 0: 2.0017172954349105 · W 1: 2.48927424932097 · W 2: 2.9771515118233163 · W 4: 3.952936767255445 · W 8: 5.767925977644397 · W 12: 7.856077788983961
