Luminex - a UCI chess engine written in C++23. All evaluation values are self-engineered from chess first principles.
Author: Cheng Chang
Luminex 5.13.0 what's new?
Forcing Line Deepening
This release introduces a genuinely novel search technique that tracks consecutive check sequences across plies, treating the forcing LINE as a whole rather than extending individual moves independently.
What's New
- Forcing line extension: Tracks consecutive check plies. When 3+ plies are in check, extends depth to see the tactical conclusion — catching combinations that individual check extensions miss (+72.5 Elo).
- Hindsight depth adjustment: Compensates for aggressive LMR reductions by adjusting child depth based on parent reduction and eval feedback (+14 Elo).
- Eval-adaptive LMR: Uses eval-to-beta distance to modulate reductions — reduces more when clearly winning, less in critical positions.
- King flight safety: Penalty for king with limited escape squares in king danger evaluation.
- Prioritized capture checks: Captures that give check get ordering bonus in PMG phase 2.
Test Results
vs StashV20 (2509 Elo), tc=1+0.01:
| Run | W | L | D | Score | Cutechess Elo |
|---|---|---|---|---|---|
| Run 1 | 66 | 127 | 7 | 0.347 | -109.5 ± 49.9 |
| Run 2 | 55 | 137 | 8 | 0.295 | -151.3 ± 52.0 |
| Average | 121 | 264 | 15 | 0.321 | -130.4 |
Estimated Strength
~2380 CCRL Blitz equivalent (StashV20 2509 - 130 = ~2379, +170 Elo over v5.7.0)

Comments
Post a Comment