Revolution 2.80 vs Baseline — Summary for Release Notes
Executive summary
Across three gauntlets at identical test conditions (10+0.1, 1 thread, 32 MB hash, UHO_2024_8mvs_+085_+094), Revolution 2.80 scores a small but robust Elo gain over the baseline. Aggregating all runs (4,260 games), the meta-estimate is +13.9 Elo ± 5.6 in favor of Revolution 2.80. This corresponds to a 52.0% score rate overall (approx.), consistent with a modest but statistically meaningful improvement for practical play and continuous integration benchmarks.
Per-run results
| Run | Games | Score | Elo (±) | Notes |
|---|---|---|---|---|
| A | 2,000 | 52.42% | +16.86 ± 8.37 | LOS 100%, PairsRatio 1.38. |
| B | 2,000 | 51.48% | +10.25 ± 8.03 | LOS 99.38%, PairsRatio 1.26. |
| C | 260 | 53.08% | +21.41 ± 24.43 | LOS 95.77%, PairsRatio 1.55. |
Total games: 4,260 (A+B+C).
Aggregate estimate (inverse-variance meta-analysis): +13.9 Elo ± 5.6.
Methodological note: To summarize across runs with different uncertainties, we used an inverse-variance weighted average of the reported Elo estimates. (A simple game-weighted average gives a similar figure, ~+14.0 Elo.) The consistency across independent gauntlets supports a genuine—albeit modest—performance uplift.
What this means for users
- Practical strength: Expect ~+14 Elo on the tested settings/book. That’s a small but real edge in long self-play testbeds and large batch analysis.
- Stability: The effect reproduces across independent 2k-game matches, with high LOS and narrow error bars on the larger runs.
- Caveat: Several runs show PairsRatio > 1.2, which slightly reduces statistical efficiency. The aggregate meta-estimate mitigates that by weighting by uncertainty.
TL;DR
Revolution 2.80 delivers a ~+14 Elo improvement over the baseline across 4,260 games at 10+0.1, 1 thread, 32 MB with the UHO_2024_8mvs_+085_+094 opening suite. Consistent 2k-game gauntlets report LOS ≥ 99% and a combined meta-Elo of +13.9 ± 5.6, confirming a reliable, incremental strength gain suitable for production use.
Restored returning the pointer from
posix_memalignin the POSIX branch ofstd_aligned_alloc, preventing a successful allocation from falling through without a return value
Jorge Ruiz
connoisseur of both chess and anthropology, a combination that reflects his deep intellectual curiosity and passion for understanding both the art of strategic chess books