bench dense arm · model Qwen/Qwen3-Embedding-0.6B-Q · split selection · code 2077f098670a · prompts "Instruct: Given a developer's task, question or bug report, retrieve the engineering lesson that applies to it\nQuery:{}" / '{}'
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 2.49 s · load here 1.15 s · index 1294.3 s · encode per prompt median 1284.3 ms
decision score = evidence + 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   8.250   3 |   83.9% [79.1%, 88.4%]      2.57 | 6 of 268 (2.2%, [1.0%, 4.8%]) | 60/83
  0.0   8.250   5 |   84.3% [79.5%, 88.8%]      3.59 | 6 of 268 (2.2%, [1.0%, 4.8%]) | 60/83
  5.0  10.577   3 |   86.7% [82.3%, 91.2%]      2.65 | 12 of 268 (4.5%, [2.6%, 7.7%]) | 63/83
  5.0  10.577   5 |   88.8% [84.7%, 92.8%]      3.74 | 12 of 268 (4.5%, [2.6%, 7.7%]) | 65/83
 10.0  13.181   3 |   87.6% [83.1%, 91.6%]      2.64 | 8 of 268 (3.0%, [1.5%, 5.8%]) | 65/83
 10.0  13.181   5 |   89.2% [85.1%, 92.8%]      3.67 | 8 of 268 (3.0%, [1.5%, 5.8%]) | 66/83
 20.0  18.732   3 |   89.2% [84.7%, 93.2%]      2.56 | 8 of 268 (3.0%, [1.5%, 5.8%]) | 67/83
 20.0  18.732   5 |   91.2% [87.6%, 94.4%]      3.50 | 8 of 268 (3.0%, [1.5%, 5.8%]) | 69/83
 40.0  30.168   3 |   89.2% [84.7%, 93.2%]      2.45 | 12 of 268 (4.5%, [2.6%, 7.7%]) | 66/83
 40.0  30.168   5 |   90.4% [86.7%, 94.0%]      3.25 | 12 of 268 (4.5%, [2.6%, 7.7%]) | 67/83
