{
 "schema": 1,
 "suite": "parity",
 "date": "2026-10-24",
 "machine": "rtx-4090",
 "device": "cpu",
 "ollaya": "0.17.0 at b97e96a (decider rows, PR #144) and at b37b867 (d1-omni:600m) and at d4ea37f (strands-decider:e2b, :e4b) and at 59d230f (gliclass-v4:2b, :9b)",
 "protocol": {
  "reference": "goldens from each family's goldens.py: the authors' code in PyTorch fp32 (TF32 off) on the checkpoint the manifest names",
  "gate": "fp32 tags: identical encodings and decisions, option logits within 1e-3, wire answers as the authors' code; -fp16 tags (docs/decisions/0008-fp16-decoder-tags.md): identical encodings, at least 99.5 % of decisions the same, differences reported",
  "runner": "convert/ollaya_convert/parity_run.py with crates/ollaya-runner/examples/parity_<family>.rs"
 },
 "results": [
  {
   "model": "decider:2b",
   "questions": 479,
   "decisions_same": 479,
   "max_logit_diff": 0.0000400543212890625,
   "max_prob_diff": 0.000049741802474884134,
   "p99_prob_diff": 0.0000035690437506508133,
   "pass": true,
   "device": "cpu",
   "kind": "runtime",
   "gate": "fp32",
   "export_sha256": "9aaf0a5a45a7419d447a82cf9532d3f187163d30fb53f65567dc67cb34ef7e7a"
  },
  {
   "model": "decider:4b",
   "questions": 479,
   "decisions_same": 479,
   "max_logit_diff": 0.00005340576171875,
   "max_prob_diff": 0.00018378725862522316,
   "p99_prob_diff": 0.000004310762667492085,
   "pass": true,
   "device": "cpu",
   "kind": "runtime",
   "gate": "fp32",
   "export_sha256": "b7ffca6f95269b6a8ecaa600a2092ade7a20d55d342d89915199cb49962c29c0"
  },
  {
   "model": "d1-omni:600m",
   "questions": 456,
   "decisions_same": 456,
   "max_logit_diff": 0.00014024972915649414,
   "max_prob_diff": 0.000009819286194368004,
   "p99_prob_diff": 0.00000780113757109513,
   "pass": true,
   "device": "cpu",
   "kind": "runtime",
   "gate": "fp32",
   "export_sha256": "fd603c73eba7dcc33648e155efc8fe4dc14a7f8aeca0737efc2be3ea29a03e7f"
  },
  {
   "model": "strands-decider:e2b",
   "questions": 650,
   "decisions_same": 650,
   "max_logit_diff": 0.000021476298570632935,
   "max_prob_diff": 0.000011707158444451693,
   "p99_prob_diff": 0.0000014779889780935918,
   "pass": true,
   "device": "cpu",
   "kind": "runtime",
   "gate": "fp32",
   "export_sha256": "790ecb95e54cf68acbab0d0820032c63bd837c214cc6e51a98c2c19da5761ce3"
  },
  {
   "model": "strands-decider:e4b",
   "questions": 650,
   "decisions_same": 650,
   "max_logit_diff": 0.000017181038856506348,
   "max_prob_diff": 0.000005207979155874565,
   "p99_prob_diff": 0.0000015002537040453845,
   "pass": true,
   "device": "cpu",
   "kind": "runtime",
   "gate": "fp32",
   "export_sha256": "935ac79395c9f9014e4ac201c31898ccb71d873757acc524345e017b4d6d2f5a"
  },
  {
   "model": "gliclass-v4:2b",
   "questions": 483,
   "decisions_same": 483,
   "max_logit_diff": 0.00009393692016601562,
   "max_prob_diff": 0.000024913899255962768,
   "p99_prob_diff": 0.0000053676184151307815,
   "pass": true,
   "device": "cpu",
   "kind": "runtime",
   "gate": "fp32",
   "export_sha256": "75258aa8dc83b422eb43ff03f3c6245e95a8fb255fc5ed89c9e497185501d00e"
  },
  {
   "model": "gliclass-v4:9b",
   "questions": 483,
   "decisions_same": 483,
   "max_logit_diff": 0.000028371810913307982,
   "max_prob_diff": 0.000004151531498075567,
   "p99_prob_diff": 0.0000029835161614588657,
   "pass": true,
   "device": "cpu",
   "kind": "runtime",
   "gate": "fp32",
   "export_sha256": "16ca230cea5d297fa0dc74e7327600652b7bd3aff893ff2d9180415546da1774"
  }
 ]
}
