lucid-fdn/mode-optimization

Mode Optimization

Optimize how GPU time is allocated across 23 search modes. The auto-optimize pipeline bridges target intelligence (triage scores, fingerprints) with work distribution (priority weights).

Procedure

1. Analyze costs — Call quantum_cost to see GPU cost per billion keys per mode
2. Auto-optimize — Call quantum_optimize to automatically set priority weights from intelligence data
3. Verify — Call quantum_fleet to confirm the new weight distribution makes sense
4. Monitor — Run quantum_protect periodically to detect mode stalls

Auto-Optimize Algorithm

The quantum_optimize tool bridges intelligence → work distribution:

1. Aggregate triage scores — For each target with blockchain metadata, compute mode vulnerability probability. Average across all targets per mode.

2. Aggregate fingerprints — For each target with a wallet fingerprint (confidence > 0.3), accumulate probable_modes weighted by confidence.

3. Compute final weight — Per mode:

   base_weight = 1.0 + avg_triage_score * 7.0     (1.0 - 8.0 range)
   fp_boost = min(sum_fingerprint_probabilities, 2.0)  (0.0 - 2.0 range)
   final_weight = min(base_weight + fp_boost, 10.0)
   

4. Apply — Write to ModeProgress.priority_weight. Exhausted modes stay at 0.0.

5. Cache — Computed mode_scores and narrowed_ranges are cached to TargetMetadataCache for future queries.

Priority Weight Effects

work_service._get_weighted_mode_order() reads priority weights:

• Modes sorted by priority_weight DESC
• Modes with weight 0.0 are skipped entirely
• Within same weight, finite modes are tried before infinite
• Workers get work from the highest-priority non-exhausted mode with available keyspace

References

• cost-model.md