Dev Builds » 20260930-1840

Use this dev build

NCM plays each Stockfish dev build 20,000 times against Stockfish 15. This yields an approximate Elo difference and establishes confidence in the strength of the dev builds.

Summary

Host Duration Avg Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo
ncm-dbt-01 06:43:54 583146 4000 1814 290 1896 +139.4 ± 4.36 0 14 492 1450 44 +330.23 ± 15.32
ncm-dbt-02 06:42:34 591278 4000 1838 250 1912 +145.95 ± 4.31 0 17 425 1511 47 +354.93 ± 16.53
ncm-dbt-03 06:43:15 583734 4000 1828 307 1865 +139.1 ± 4.53 1 16 502 1423 58 +324.17 ± 15.17
ncm-dbt-04 06:43:12 568182 4000 1823 282 1895 +141.13 ± 4.3 0 18 459 1487 36 +340.03 ± 15.9
ncm-dbt-05 06:42:53 603030 4000 1809 263 1928 +141.64 ± 4.43 0 21 460 1471 48 +337.25 ± 15.89
20000 9112 1392 9496 +141.44 ± 1.96 1 86 2338 7342 233 +337.09 ± 7.03

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
486240 ncm-dbt-01 584790 500 228 37 235 +139.81 ± 13.26 0 2 65 173 10 +318.25 ± 42.59 ↓
486239 ncm-dbt-03 580655 500 223 45 232 +129.35 ± 11.81 0 0 75 172 3 +301.33 ± 39.1 ↓
486238 ncm-dbt-05 604667 500 218 30 252 +137.37 ± 12.76 0 5 56 185 4 +327.18 ± 46.19 ↓
486237 ncm-dbt-02 590438 500 230 29 241 +148.02 ± 12.94 0 2 55 183 10 +349.43 ± 46.59 ↓
486236 ncm-dbt-04 568593 500 231 32 237 +146.36 ± 11.05 0 1 51 196 2 +370.41 ± 48.42 ↓
486235 ncm-dbt-01 583740 500 229 33 238 +143.89 ± 12.3 0 2 56 186 6 +346.12 ± 46.15 ↓
486234 ncm-dbt-02 591426 500 227 26 247 +148.02 ± 13.15 0 4 50 187 9 +352.8 ± 49.02 ↓
486233 ncm-dbt-05 600735 500 233 34 233 +146.36 ± 11.96 0 1 55 188 6 +356.21 ± 46.49 ↓
486232 ncm-dbt-03 584748 500 235 33 232 +148.85 ± 12.49 0 2 52 188 8 +359.68 ± 48.01 ↓
486231 ncm-dbt-04 568116 500 227 34 239 +141.44 ± 11.98 0 2 57 187 4 +342.85 ± 45.71 ↓
486230 ncm-dbt-01 583405 500 220 32 248 +137.37 ± 11.73 0 0 66 180 4 +327.17 ± 41.95 ↓
486229 ncm-dbt-05 603859 500 237 27 236 +155.54 ± 11.65 0 1 45 197 7 +393.6 ± 51.79 ↓
486228 ncm-dbt-02 590911 500 230 24 246 +152.18 ± 11.4 0 1 47 197 5 +385.59 ± 50.59 ↓
486227 ncm-dbt-04 569669 500 225 25 250 +147.19 ± 11.47 0 1 52 193 4 +366.78 ± 47.92 ↓
486226 ncm-dbt-03 585126 500 221 41 238 +130.94 ± 12.76 0 3 69 173 5 +301.33 ± 41.33 ↓
486225 ncm-dbt-05 599893 500 222 33 245 +138.18 ± 13.12 0 4 60 179 7 +321.19 ± 44.57 ↓
486224 ncm-dbt-01 583908 500 218 45 237 +125.38 ± 12.32 0 2 76 169 3 +288.06 ± 39.11 ↓
486223 ncm-dbt-02 592331 500 235 29 236 +152.18 ± 11.4 0 3 41 203 3 +393.6 ± 54.43 ↓
486222 ncm-dbt-03 584034 500 235 39 226 +143.89 ± 12.3 0 0 62 180 8 +339.63 ± 43.41 ↓
486221 ncm-dbt-04 566572 500 233 44 223 +138.18 ± 12.33 0 2 62 181 5 +327.18 ± 43.69 ↓
486220 ncm-dbt-05 599760 500 233 33 234 +147.19 ± 12.98 0 2 56 182 10 +346.12 ± 46.15 ↓
486219 ncm-dbt-01 581527 500 225 22 253 +149.68 ± 11.32 0 0 52 193 5 +374.1 ± 47.75 ↓
486218 ncm-dbt-03 582402 500 225 32 243 +141.44 ± 13.59 0 4 59 177 10 +324.17 ± 44.97 ↓
486217 ncm-dbt-02 591685 500 232 28 240 +150.51 ± 11.51 0 1 49 195 5 +377.87 ± 49.47 ↓
486216 ncm-dbt-04 569111 500 226 37 237 +138.18 ± 12.53 0 4 57 185 4 +330.23 ± 45.79 ↓
486215 ncm-dbt-05 604937 500 224 40 236 +134.15 ± 12.86 0 4 63 178 5 +312.48 ± 43.44 ↓
486214 ncm-dbt-01 584958 500 237 42 221 +143.07 ± 11.69 0 2 54 191 3 +352.8 ± 47.05 ↓
486213 ncm-dbt-04 566572 500 226 38 236 +137.37 ± 12.36 0 4 57 186 3 +330.23 ± 45.79 ↓
486212 ncm-dbt-03 582694 500 235 36 229 +146.36 ± 12.4 0 2 54 187 7 +352.8 ± 47.05 ↓
486211 ncm-dbt-02 590782 500 225 31 244 +142.25 ± 12.37 0 2 58 184 6 +339.63 ± 45.29 ↓
486210 ncm-dbt-05 602337 500 218 32 250 +135.76 ± 12.01 0 2 63 182 3 +324.17 ± 43.32 ↓
486209 ncm-dbt-02 593412 500 224 45 231 +130.14 ± 12.78 0 4 67 175 4 +301.33 ± 42.04 ↓
486208 ncm-dbt-03 586012 500 235 34 231 +148.02 ± 12.74 0 2 54 185 9 +352.8 ± 47.05 ↓
486207 ncm-dbt-04 568275 500 230 42 228 +137.37 ± 12.76 0 1 68 173 8 +315.35 ± 41.44 ↓
486206 ncm-dbt-01 580945 500 227 34 239 +141.44 ± 13.59 0 4 59 177 10 +324.17 ± 44.97 ↓
486205 ncm-dbt-02 589240 500 235 38 227 +144.71 ± 11.61 0 0 58 187 5 +352.79 ± 45.01 ↓
486204 ncm-dbt-05 608059 500 224 34 242 +138.99 ± 12.5 0 2 62 180 6 +327.18 ± 43.69 ↓
486203 ncm-dbt-03 584201 500 219 47 234 +124.6 ± 13.97 1 3 77 161 8 +275.45 ± 39.01 ↓
486202 ncm-dbt-04 568554 500 225 30 245 +143.07 ± 12.54 0 3 55 186 6 +342.85 ± 46.64 ↓
486201 ncm-dbt-01 581902 500 230 45 225 +134.95 ± 12.04 0 2 64 181 3 +321.19 ± 42.95 ↓

Commit

Commit ID a35e229ec92815450768cf45b42e1127d20a3426
Author Robert Nurnberg
Date 2026-09-30 18:40:26 UTC
Smoothen seekMate trigger This PR replaces the binary `seekMate` trigger with a smoother version. The motivation for the idea comes from #7179. There it was found that master for some of the 4-ply mate puzzles reached a root eval just below the old trigger of 2000, and stayed at that for many iterations. In multi-threaded analysis this could at times lead to a failure of our matetrack CI. This patch replaces the old trigger of 2000 (from depth 16 onwards) with a depth-dependent value. The smooth curve `750 + 220000 / (d*d)` approaches 750 towards infinity and goes through the points (10, 2950), (16, 1609), (20, 1300), (50, 838). These raw eval values for the current WDL model and at the material value 58 correspond to 970cp, 529cp, 427cp, 276cp, and all have W=100%. Indeed, patch on average and in the worst case finds the mates from our 4-ply CI set faster than master: ``` Using ./sf.master on mates2head.epd with --mate 0 --threads 1 Engine ID: Stockfish dev-20260922-0a215d6c Total FENs: 1000 Found mates: 1000 Best mates: 1000 Best mate statistics: |bm| = 1 - mates found: 4 = 100.0% of 4; nodes (min avg max): 4 69 220, depth (min avg max): 1 1 1 |bm| = 2 - mates found: 996 = 100.0% of 996; nodes (min avg max): 4 127441 53095641, depth (min avg max): 1 9 78 All best mates found: 1000 = 100.0% of 1000; nodes (min avg max): 4 126931 53095641, depth (min avg max): 1 9 78 ``` ``` Using ./sf.patch on mates2head.epd with --mate 0 --threads 1 Engine ID: Stockfish dev-20260926-m-ff6998f6 Total FENs: 1000 Found mates: 1000 Best mates: 1000 Best mate statistics: |bm| = 1 - mates found: 4 = 100.0% of 4; nodes (min avg max): 4 69 220, depth (min avg max): 1 1 1 |bm| = 2 - mates found: 996 = 100.0% of 996; nodes (min avg max): 4 23482 8342063, depth (min avg max): 1 8 63 All best mates found: 1000 = 100.0% of 1000; nodes (min avg max): 4 23389 8342063, depth (min avg max): 1 8 63 ``` So patch finds the mates on average with 23K nodes, master with 127K. And the maximum nodes are about 8M vs 53M in favour of patch. The patch passed STC and LTC non-reg vs. master with `seekMate` disabled. STC: LLR: 2.93 (-2.94,2.94) <-1.75,0.25> Total: 173504 W: 44066 L: 44003 D: 85435 Ptnml(0-2): 215, 18265, 49709, 18368, 195 https://tests.stockfishchess.org/tests/view/6ab230bbe28108115db44437 LTC: LLR: 2.96 (-2.94,2.94) <-1.75,0.25> Total: 375264 W: 96398 L: 96530 D: 182336 Ptnml(0-2): 80, 36928, 113741, 36810, 73 https://tests.stockfishchess.org/tests/view/6ab56fae83e20a0615e67adb And even though that was not the original intention of the patch, it also shows a nice bump in our matetrack stats: ``` Using ./sf.patch on matetrack.epd with --nodes 1000000 Engine ID: Stockfish dev-20260926-m-ff6998f6 Total FENs: 6554 Found mates: 3765 Best mates: 2535 ``` This compares with 3595, 2382 for master, see [matetrack](https://github.com/vondele/matetrack/blob/edee34f4ba4c45c2f7141ec0e7503baf3a1f7ef1/matetrack1000000.csv#L4588). For the suite of classic mate puzzles the difference is minor: 107, 20 vs 107, 12 in favour of patch. Fixes #7179. closes https://github.com/official-stockfish/Stockfish/pull/7201 Bench: 1514301
Copyright 2011–2026 Next Chess Move LLC