[eval_v22] Running inference on /home/aiserver/works/fingerprint/sifq/checkpoints/v22/last.pt ...
Loading checkpoint: /home/aiserver/works/fingerprint/sifq/checkpoints/v22/last.pt
Model loaded. Device: cuda
After excluding sensors {'R_500_slap', 'S_500_slap', 'R_1000_slap'}: 29691 records remain
Discovered 29691 records
  4000/29691 done
  8000/29691 done
  12000/29691 done
  16000/29691 done
  20000/29691 done
  24000/29691 done
  28000/29691 done
Done. Scores saved to: /home/aiserver/works/fingerprint/sifq/eval_results/sifq_scores_v22.jsonl
[eval_v22] Inference done: /home/aiserver/works/fingerprint/sifq/eval_results/sifq_scores_v22.jsonl
[eval_v22] Running evaluation (Track 2 + Track 4, exclude: R_1000_slap,R_500_slap,S_500_slap)...
Loaded 29691 SIFQ score records from /home/aiserver/works/fingerprint/sifq/eval_results/sifq_scores_v22.jsonl
After excluding sensors {'S_500_slap', 'R_500_slap', 'R_1000_slap'}: 29691 records remain
[Track 2] Computing sensor invariance...
  Sensor invariance done. Mean KS=0.1527, Pearson=0.4433
[Track 1] Skipped (--skip-track1)
[Track 4] Computing concept grounding (crosstalk matrix)...
  Crosstalk matrix saved.

============================================================
EVALUATION SUMMARY
============================================================

Track 2 — Sensor Invariance:
  SIFQ      : Mean KS = 0.1527  Pearson = 0.4433

Plots and JSON saved to: /home/aiserver/works/fingerprint/sifq/eval_results/v22
Full summary: /home/aiserver/works/fingerprint/sifq/eval_results/v22/eval_summary.json
[eval_v22] Eval done. Results in /home/aiserver/works/fingerprint/sifq/eval_results/v22
[eval_v22] Generating score milestone samples...
Saved → /home/aiserver/works/fingerprint/sifq/eval_results/v22/milestone_samples.png