resident q4: 0.97 GB · streamed q8: 1.83 GB
baseline: pure q8 greedy …
baseline: pure q4 greedy …
runtime: mode=margin tau=1.0 W=32 …
  fidelity=0.5518 tok/s=99.6 defer=0.38 rollback=0.067 MB/token(logical)=154
runtime: mode=margin tau=2.0 W=32 …
  fidelity=0.6484 tok/s=71.2 defer=1.00 rollback=0.106 MB/token(logical)=216
runtime: mode=verify-all tau=2.0 W=32 …
  fidelity=0.8825 tok/s=63.6 defer=1.04 rollback=0.117 MB/token(logical)=231

wrote /Users/simon-pierreboucher/Desktop/localvm-research/results/candidate_01/20260812T051501Z/results.json
baselines: {
 "pure_q4": {
  "tokens_per_s": 287.5061867410425,
  "fidelity_vs_q8_mean": 0.486328125
 },
 "pure_q8": {
  "tokens_per_s": 200.05224940165664,
  "fidelity_vs_q8_mean": 1.0
 }
}