Datasets:
File size: 281,465 Bytes
535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c df5b7d0 535080c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 | {
"record_type": "publication_metadata",
"record_version": "1.1.0",
"evidence_manifest_v1_0_0_sha256": "a3ecb53728523005648b995871b8389f40cfa997a45b7f58935e9cecc335b936",
"evidence_manifest_v1_1_0_sha256": "4c5328b7ebf2e6f9a037ad0ad09f756c16c794bf2006c465a28ab394c4ee91d4",
"evidence_manifest_v1_1_0_public_half_sha256": "e2a1e2507c6909f7159848d940f54c6be8c4c703621058e346a75582974a3eb1",
"evidence_manifest_note": "Each value is the SHA-256 of the bytes of a checksum manifest file, one per retained evidence package. evidence_manifest_v1_0_0_sha256 is the 1.0.0 package, and is the value 1.0.0 published under the name evidence_manifest_sha256: unchanged, and carried here so the older commitment stays readable. evidence_manifest_v1_1_0_sha256 covers the whole 1.1.0 package, both halves; that manifest lists the filenames of private holdout material, so it is never distributed. evidence_manifest_v1_1_0_public_half_sha256 covers the public half of the same 1.1.0 package, which is the half that can be shown to a reviewer, so it is the only one of the three that a holder of the evidence can verify against. All three packages remain private; these hashes are commitments to their contents, not a means of independent verification.",
"release_id": "CompilingThings/compile-benchmark-v1.1.0",
"release_title": "CompilingThings Compile Benchmark for MQL5\u00ae",
"release_repo_slug": "compile-benchmark",
"metaquotes": {
"trademark_attribution": "MQL5\u00ae and MetaTrader 5\u00ae are registered trademarks of MetaQuotes Ltd.",
"no_endorsement": "CompilingThings is an independent project. No affiliation, sponsorship, certification, endorsement, or approval by MetaQuotes Ltd. is claimed.",
"compiler_use": "Compilation validation was performed using the MetaQuotes Language Compiler supplied with MetaEditor. The compiler is used solely as the MQL5\u00ae compilation and validation instrument.",
"no_redistribution": "This release does not distribute MetaTrader 5\u00ae, MetaEditor, the MetaQuotes Language Compiler, EX5 protection technology, MetaQuotes documentation, or other MetaQuotes-owned software or proprietary materials.",
"source_ownership": "The benchmark prompts published in this release were produced by CompilingThings' own generators and are owned by CompilingThings. No model-generated MQL5 source or other model outputs are distributed. No ownership of the MQL5 language, the MetaTrader platform, MetaEditor, or MetaQuotes technology is claimed.",
"marks_remain": "All MetaQuotes trademarks and MetaQuotes-owned technology remain the property of MetaQuotes Ltd."
},
"item_naming": {
"item_id_purpose": "item_id is a stable public identifier used to join prompts and result rows.",
"names_are_recoverable": "The original item name of every published item is recoverable: each prompt contains a `// NAME:` line, because that line was part of the prompt the models were given. Publishing the prompts as evaluated therefore publishes the names.",
"why_not_removed": "Removing the names would make the published prompt differ from the evaluated prompt. Every hash in this release, and the claim that a reader is running the benchmark we ran, depends on those being the same bytes. The names are published rather than the benchmark being weakened.",
"what_this_discloses": "The names disclose the corpus naming scheme, indicator families, parameter conventions and versioning for these 184 items. The corpus itself, and the names of every item not in this release, remain undistributed."
},
"positive_control": {
"what_it_is": "Before an arm is scored, two fixtures are compiled. One is known to produce a compiler error. One is known to produce a compiler warning and still build.",
"pass_condition": "The error fixture must report at least one error. The warning fixture must report at least one warning and must score as a headline pass and a strict failure - the one outcome that tells the two metrics apart.",
"on_failure": "The harness raises and writes no rows for that arm. A scored arm is therefore an arm whose control passed; there is no run in which the control failed and results were kept anyway.",
"why_every_row_reads_1_and_1": "The recorded counts come from the fixtures, not from the evaluated item, so they are the same on every row of an arm. A row reading 1 and 1 means the control passed for that arm. It is not a defect in that item.",
"what_it_establishes": "That the compiler and the log scorer could still see an error and a warning during this run. It says nothing about any individual item's verdict."
},
"serving": {
"scope": "The two bf16 local arms of the 1.0.0 release, and nothing else. The three local arms that produced every 1.1.0 row were served at Q8_0 through llama.cpp and are described under serving_v1_1. The frontier arm's serving is described under models.frontier.",
"engine": "Hugging Face transformers, AutoModelForCausalLM.from_pretrained, generating with model.generate",
"decoding": "Greedy. do_sample=False and top_p=1.0.",
"temperature_note": "Both local arms record temperature 0.0 and temperature_applied UNSUPPORTED. Greedy decoding does not consume a temperature, so the requested value was recorded and had no effect. This is why the field reads UNSUPPORTED rather than APPLIED.",
"seed": "torch.manual_seed is set per generation. Recorded as seed_applied APPLIED.",
"compute_dtype": "bfloat16, passed as torch_dtype. No quantisation was applied at load time for either local arm.",
"max_new_tokens": 8192,
"library_versions_not_established": "The transformers, torch and tokenizer library versions in force during the evaluation run were not recorded in the run parameters and are not established. The merge-time environment is recorded, but that was a different process.",
"evaluation_hardware": "Both local arms were generated on one AMD Radeon AI PRO R9700 (RDNA 4, gfx1201) under ROCm, in an AMD Ryzen 9 7950X3D host with 64 GB of system memory.",
"hardware_note": "The device is not recorded in the run parameters; it is stated here from the evaluation host's configuration. An earlier version of this note inferred CPU execution from the per-item timing alone, which was wrong.",
"compiler_build": 5836,
"compiler_build_note": "The three 1.0.0 arms compiled against the same compiler build, and so did every arm added in 1.1.0. The build number is published; the compiler binary is not distributed and is not ours to distribute."
},
"attested_not_verifiable": {
"statement": "The following are author attestations. They rely on undistributed evidence and cannot be verified from this package. Each claim is scoped: a claim that names a release, or names its own subject, is about that one alone, and every other claim describes an instrument or a procedure both releases share. 1.0.0 here means the frozen-300 evaluation and the 184-item public set carried forward from the 1.0.0 release; 1.1.0 means the two holdout sets, the Q8_0 bridge and the corpus row-hash manifest this release adds.",
"claims": [
"The per-item verdicts are the compiler outcomes we observed. The generated MQL5, the compiler logs and the compiler binary are not distributed, so a reader can recompute the statistics from our rows but cannot confirm the rows against the artifacts that produced them.",
"The base model identity at the pinned revision.",
"The tuned arm's adapter hash, epoch, global step, merge tool, merge settings and merged-weight hash.",
"The determinism result for the two historical merges.",
"The decontamination collision counts, which are measured against a corpus that is not distributed.",
"1.0.0 scope - that the 1.0.0 evaluation set was frozen at seed 42 before any result was seen and was not redrawn. The 1.1.0 holdout sets were selected by a different rule, published with its three digest conventions under holdout_sets; those digests commit to manifests that are not distributed, so the selection behind them is an attestation of the same kind.",
"Everything stated about how the harness behaves: that the positive control aborts an arm rather than writing rows, that the seed is set per generation, that extraction and scoring ran identically on all three arms, and that the scoring rules published here are the rules the harness applied. The harness is not distributed.",
"1.0.0 scope - the full-300 comparison. Of the 1.0.0 frozen 300 only the clean-184 rows are published, so the full-300 figures and the change between the two strata cannot be recomputed from this package. That limit is on the 1.0.0 file set alone. For the 1.1.0 holdouts every projected result row is published - 1,200 rows in holdout_ea300_results.jsonl and 800 in holdout_nonea200_results.jsonl - and their arithmetic is checkable, with the prompts and item names withheld as what_is_verifiable_instead below states.",
"1.0.0 scope - which model produced which arm. The public `arm` is derived from the source directory of the sealed rows; the two local arms' own sealed arm field reads the same value on both and discriminates nothing. We bound the tuned rows to the merged artifact internally by byte-identity to the run directory that served it, but a reader holding only this release cannot check that attribution. The same holds in 1.1.0 and for the same reason: PROJECTION_REPORT.json arm_derivation_v1_1 records that the published arm comes from the source extract, while the sealed rows' own arm field reads the same value, B, on every local arm of this release, and C on the frontier arm."
],
"what_is_verifiable_instead": "From the 1.0.0 file set: everything computed from the published rows - both prompt hashes, the pairing, all three rates, the differences, the confidence intervals, both McNemar results and both paired tables. From the files 1.1.0 adds: in holdout_ea300_results.jsonl, holdout_nonea200_results.jsonl and bridge_q8_184_results.jsonl a reader can check the row counts, the arm ids, the headline and truncation counts per arm and the item pairing across arms; the holdout prompt hashes cannot be recomputed and no holdout row can be regenerated, because the prompts are not published. Across both, a reader can run the 184 published prompts against any model and compare, which is the comparison the release exists to enable."
},
"licence_scope": {
"licensed_under_the_benchmark_evaluation_licence": [
"prompts.jsonl",
"per_item_results.jsonl",
"serving_template.json",
"holdout_ea300_results.jsonl",
"holdout_nonea200_results.jsonl",
"bridge_q8_184_results.jsonl",
"corpus_row_hashes.json"
],
"supporting_files": [
"README.md",
"LICENSE",
"CITATION.cff",
"SHA256SUMS.txt",
".gitattributes",
"publication_metadata.json",
"PROJECTION_REPORT.json",
"verify_public_release.py"
],
"supporting_files_terms": "The supporting files describe and verify the release. They may be reproduced and quoted for the purpose of using, checking or citing this benchmark, including in published results that disagree with ours, and verify_public_release.py may be run as-is to verify a copy of this release. They carry no grant over the benchmark data itself, which is covered by the licence above.",
"note": "Being listed as published means a file ships in the release. It does not by itself say which terms cover it; this block does."
},
"error_counts": {
"errors": "The number of individual error diagnostics printed in the compile log, counted line by line.",
"result_line_errors": "The compiler's own final tally, read from its `Result: N errors, M warnings` line.",
"why_they_differ": "The compiler stops printing individual diagnostics past a limit while its final tally keeps counting. A log can list 99 errors by line and end with `Result: 103 errors`. Both numbers are recorded and neither is corrected against the other.",
"measured_disagreement": "They disagree on 12 of the 368 local-arm rows of per_item_results.jsonl and on 14 of the 552 local-arm rows of bridge_q8_184_results.jsonl - 26 of the 920 local-arm rows - all in the base arm, all at 99 or 100 printed diagnostics. Both counts are published on 1,104 rows, not 920: those 920 local-arm rows and the 184 frontier rows of per_item_results.jsonl, which carry both counts and disagree on none, in errors or in warnings. The two holdout row files publish `errors` and no `result_line_errors`, so no disagreement is computable on them.",
"effect_on_the_verdict": "None. A verdict turns on whether there were any errors at all, and both counts are far above zero on every row where they differ."
},
"bootstrap_caveat": "The bootstrap interval is reported for completeness and is not independently reproducible from this package: the resampling implementation, its RNG and the percentile convention are not published, and a seed alone does not determine identical endpoints across implementations. The Wilson and Newcombe intervals, the McNemar result and both rates do recompute from the published rows, and those are the intervals the headline rests on.",
"scoring": {
"verdicts": {
"verdict_headline": "TRUE when the compile log reports zero errors AND the .ex5 artifact was produced. This is the headline metric.",
"verdict_strict": "TRUE when the compile log reports zero errors AND zero warnings AND the .ex5 artifact was produced.",
"note": "Both are computed for every item that reached compilation, and published side by side. Neither is derived from the other. Where generation failed outright nothing reached the compiler and both fields are null: three rows, all in holdout_ea300_results.jsonl, all on the frontier arm, each with bucket INFRASTRUCTURE-FAILURE and finish_reason `error`. Those are the only null verdicts in the release. A null is not a failure and must not be read as one - it is an item on which nothing about the model was measured, which is why INFRASTRUCTURE-FAILURE is excluded from the denominator above."
},
"buckets": {
"values": [
"compile-pass",
"compile-fail",
"TRUNCATED",
"INFRASTRUCTURE-FAILURE"
],
"precedence": "Applied in this order, and the order is load-bearing: a generation error gives INFRASTRUCTURE-FAILURE; else truncation gives TRUNCATED; else a compile-side infrastructure reason gives INFRASTRUCTURE-FAILURE; else compile-pass if verdict_headline is true, otherwise compile-fail.",
"why_that_order": "Generation errors, truncations, and compile infrastructure failures are excluded before assigning compile pass or fail. Each of the three is the harness's failure or the platform's rather than the model's, so only once all three are out does a pass or a fail describe the model.",
"excluded_from_the_denominator": [
"TRUNCATED",
"INFRASTRUCTURE-FAILURE"
]
},
"extraction": {
"version": "three_way.v4",
"step_1": "Truncate the response at the first serving end-token found, if any: <|end|>, <|endoftext|>, </s>, <|im_end|>.",
"step_2": "Find every fenced code block, accepting both CommonMark fence characters (``` and ~~~). An unterminated final fence still counts as a block, because that is what a truncated generation looks like and dropping it would hide the truncation.",
"step_3": "If there are no fenced blocks, use the whole response, stripped.",
"step_4": "Otherwise take the first block that defines a real MQL5 event handler - a recognised handler name with a matching return type and parameter signature. If no block defines one, take the first block. The chosen index is published per row as used_block_index, beside n_fenced_blocks.",
"why_the_entry_point_rule": "First-wins handed the compiler whatever a response opened with. Measured against the live compiler on an identical EA body: a leading plan block scored 2 errors and a leading block of helper includes scored 1, each published as the model's own compile failure, while the same EA sent alone passed. The cost is asymmetric - only arms that write prose and plans pay it - so the rule exists to stop the extractor manufacturing failures.",
"applied_identically_to_every_arm": "The same extraction runs on every arm. Stripping fences for some arms and not others is the bias this rule exists to prevent."
},
"compile_log_reading": {
"encoding": "The compiler writes its logs as UTF-16LE with a byte-order mark. They must be decoded before matching. Matching the raw bytes returns zero hits and exits cleanly, which silently collapses the strict metric onto the headline metric with nothing in the output saying so.",
"diagnostic_rule": "A diagnostic is a printed line matching `(line,col): error N:` or `(line,col): warning N:`, not the bare substring 'error' anywhere in the log.",
"completion_rule": "A log without the compiler's own `Result: N errors, M warnings` line is an INFRASTRUCTURE-FAILURE, never a model result. That line is the compiler's statement that a compilation ran to completion; a partial log carries zero printed diagnostics and would otherwise read as a clean compile."
},
"what_is_not_published": "The harness, the extractor and the scorer are not distributed, and neither is the compiler. These rules describe what they do at the level published here; load-bearing details below that level are not published, so an independent implementation may score individual items differently while following every published rule. A reader also cannot confirm that our published rows were produced by an implementation that follows these rules. Two of these instruments have no retained known-answer validation record: the compile verdict, whose two control fixtures and compiler output were not kept, and the extractor, whose rule is stated above and whose source is retained but for which no demonstration that it selects the right block on a known input and rejects a falsifying one was kept. Every verdict in the release rests on both, and on the holdout rows used_block_index is not published either, so the extractor's choice is not row-auditable there."
},
"interval_conventions": {
"z": 1.959963984540054,
"z_1_0_0": 1.96,
"z_note": "Two conventions are in force, one per release, because every 1.0.0 figure is carried forward exactly as 1.0.0 published it rather than recomputed. Every interval in statistics_v1_1 uses z = 1.959963984540054, the two-sided 95% normal quantile at full precision, which is the value the code that produced that block used: all 170 intervals in it reproduce with that z - 108 Wilson intervals and 62 Newcombe difference intervals - and none reproduce with the rounded value. The fields carried forward from 1.0.0 - the headline rates, the headline difference and frontier_result - were computed with the rounded z = 1.96 published here as z_1_0_0, and reproduce only with it. The two values move these endpoints by up to about two parts in 100,000. Nothing is recomputed by this scoping: it records which z produced which block, so a reader matching an endpoint uses the z that belongs to the field's release.",
"rate_intervals": "Wilson score interval, applied to each arm's successes over its denominator.",
"difference_interval": "Newcombe's square-and-add MOVER interval for the difference between two paired proportions.",
"phi": "Newcombe's phi-hat for the paired 2x2 table, the association term the MOVER interval needs. It is published because the interval cannot be recomputed without it. With cells e (both pass), f (first arm only), g (second arm only), h (both fail): A = e*h - f*g and D = sqrt((e+f)(g+h)(e+g)(f+h)). The rule, as computed: phi is A/D, and 0 when D is zero because a marginal is zero. No continuity correction is applied.",
"mcnemar": "Exact binomial test on the discordant pairs."
},
"what_is_not_measured": {
"specification_adherence": "The verdict is a compiler outcome. It does not establish that the generated program implements the specification, or any part of it. A minimal Expert Advisor that compiles and does nothing scores exactly as a correct one does. This is a stronger limitation than 'compiling is not trading correctly': adherence to the prompt is not measured at all.",
"why_it_cannot_be_ruled_out_here": "Because the generated source is not distributed, this benchmark does not distinguish semantic implementation from compilable boilerplate.",
"what_would_settle_it": "Structural or functional conformance checks against each specification, or publication of the generated source so others can judge. Neither is in this release."
},
"serving_template_confound": {
"what_was_done": "The two local arms were served with the same prompt template, the one the tuned model was trained on: <|system|>...<|end|>\\n<|user|>...<|end|>\\n<|assistant|>. The published prompt_sha256 is identical for the two local arms on every item, so this is verifiable from the release rather than asserted.",
"the_confound": "That template is not the base model's native chat format. The base model ships a ChatML chat template using <|im_start|> and <|im_end|>. Serving an instruction model a format it was not tuned for can depress its output quality, so part of the measured difference could in principle reflect template mismatch rather than capability.",
"why_it_was_done_this_way": "Holding the prompt byte-identical across arms removes prompt wording as a variable. That choice trades one confound for another: identical inputs, but an input that suits one arm better than the other.",
"corroborating_observation": "The published rows show a clean split in response convention: every base row records one fenced code block and every tuned row records none. That is consistent with two models following different output conventions, and it is what the extraction rule exists to handle fairly.",
"template_control": "The base model was run twice over the same 184 items with identical generation settings and identical scoring, once under the shared training-format template and once under its own native ChatML template. Among the 183 items with scoreable outputs under both templates, one verdict changed: it compiled under the shared training template and failed under native ChatML. The remaining item was truncated under the shared-template condition and could not be compared. The observed difference on the 183 jointly scoreable items was one compile pass in favour of the shared template. Under the shared template the control reproduced the published base arm's two passing items.",
"template_control_rows": "The control's per-item rows are not distributed. They are identified by SHA-256: training-format condition c9e0a70a99f90785ed4e7f9213ad63fecd9ddbf2319c1394885ecc09b82ce866, native ChatML condition 47f8b9eed8c1e0c369af26ca9606f48a72399f9c1cfa52b29918062964c64e33.",
"template_control_caveats": "The control served a Q8_0 quantisation of the base model on a different inference stack than the published arms, which were served at bf16. The one-item effect is measured in that control configuration; carrying it to the published bf16 arms assumes the template effect transfers across quantisation and serving stack, and that transfer was not measured. Neither control condition is a published arm and neither replaces one. The shared-template Q8 control produced the same two passing item identities as the published base arm among the 183 jointly scoreable items. One item was truncated, so full 184-item replication is not claimed.",
"status": "Measured in a control configuration: one changed verdict among 183 jointly scoreable items, in the shared template's favour. Transfer to the published bf16 arms was not measured. The headline remains a difference between the two local arms as served, under one shared template."
},
"licence": {
"title": "CompilingThings Benchmark Evaluation Licence v1.0",
"notice": "Copyright \u00a9 2026 CompilingThings. All rights reserved.\n\nThe grant covers the published benchmark files: prompts.jsonl, per_item_results.jsonl, serving_template.json, holdout_ea300_results.jsonl, holdout_nonea200_results.jsonl, bridge_q8_184_results.jsonl and corpus_row_hashes.json.\n\nYou may: run the 184 published prompts against any model; check, cite and compare the published result rows and the corpus row-hash manifest, including the holdout row files, whose prompts are not published; implement and use the published scoring contract; publish and compare benchmark results, including results that disagree with ours; cite the benchmark and the release identifier.\n\nYou may not: use the prompts for training, fine-tuning, continued pretraining, distillation, or reinforcement learning; incorporate the prompts or result rows into another dataset; create or sell derivative datasets; claim any right to the private training corpus; represent the benchmark as your own work; treat this grant as extending to future releases.\n\nNo other rights are granted. Nothing here grants any right to the private training corpus, the generated MQL5 outputs, the model weights, the production tooling, or any unreleased material. This grant does not extend to future releases or later versions of this benchmark. Commercial licences, evaluation access, research collaboration and partnership are arranged individually under written agreement. Enquiries are welcome via a discussion on the Hugging Face repository.",
"permitted": [
"run the 184 published prompts against any model",
"check, cite and compare the published result rows and the corpus row-hash manifest, including the holdout row files, whose prompts are not published",
"implement and use the published scoring contract",
"publish and compare benchmark results, including results that disagree with ours",
"cite the benchmark and the release identifier"
],
"prohibited": [
"use the prompts for training, fine-tuning, continued pretraining, distillation, or reinforcement learning",
"incorporate the prompts or result rows into another dataset",
"create or sell derivative datasets",
"claim any right to the private training corpus",
"represent the benchmark as your own work",
"treat this grant as extending to future releases"
],
"hugging_face_fields": {
"license": "other",
"license_name": "compilingthings-benchmark-evaluation-licence-v1.0",
"license_link": "LICENSE"
},
"published": [
"this dataset card (README.md)",
"the 184 benchmark prompts (prompts.jsonl)",
"the serving template and system prompt (serving_template.json)",
"sanitised per-item metadata for the three 1.0.0 arms (per_item_results.jsonl)",
"hash-keyed per-item results on the 300 EA holdout (holdout_ea300_results.jsonl)",
"hash-keyed per-item results on the 200 non-EA holdout (holdout_nonea200_results.jsonl)",
"the Q8_0 bridge rows over the 184 public prompts (bridge_q8_184_results.jsonl)",
"the corpus row-hash manifest (corpus_row_hashes.json)",
"the projection report (PROJECTION_REPORT.json)",
"the release metadata, including all aggregate results (publication_metadata.json)",
"the verification script (verify_public_release.py)",
"the checksum manifest (SHA256SUMS.txt)",
"the citation metadata (CITATION.cff)",
"the line-ending pin that keeps the checksums valid on checkout (.gitattributes)",
"the licence (LICENSE)"
],
"not_published": [
"the training corpus",
"generated MQL5 outputs, all arms",
"model weights",
"production tooling",
"the item-id-to-name mapping"
],
"note": "The grant is deliberately narrow and covers only the published benchmark files. Everything outside that list is reserved, and nothing here extends to a later release.",
"platform_terms_caveat": "Hosting-platform terms operate separately from this notice. They have not been read by the authors of this metadata and are deliberately not characterised here; only artifacts intended to be public are uploaded anywhere."
},
"training_corpus": {
"rows": 83155,
"sha256": "5e9881b61f3375d3d575c35950a236375eb5648f5fe55b8d18de321f69659c68",
"row_schema": [
"prompt",
"completion"
],
"measured": "row count and hash computed directly from the corpus file",
"distributed": false,
"why_withheld": "The corpus is not distributed with this release. Claims that depend on it are attested, not independently reproducible from the public package."
},
"evaluation": {
"source_pool_records": 9168,
"source_pool_sha256": "2ca2d669626a44b8a0096ce25873e23d6cbf08f6887295a26179218c7e0d29cf",
"frozen_n": 300,
"frozen_seed": 42,
"selection_sha256": "8684badae73db0b8c69eb2f8c3d62e6b54aed64875430498485d178af3b33ff4",
"manifest_sha256": "2accc593805d0d480e5bd6777971bf37c540a6b2bc2527283b06efc1554bf1d5",
"selection_sha256_convention": "SHA-256 of the UTF-8 bytes of json.dumps(sorted(selected_ea_names), separators=(\",\", \":\"), ensure_ascii=True) - the item names alone, sorted, compact-separated and ASCII-escaped, with no path, no seed and no counts, so it is a set identity that is identical across machines. It is not the sha256 of the manifest file.",
"manifest_sha256_convention": "SHA-256 of the UTF-8 bytes of json.dumps(the whole manifest object with its own manifest_sha256 key removed, sort_keys=True, separators=(\",\", \":\"), ensure_ascii=True), with no trailing newline. It covers the provenance fields as well as the names, and it is not the sha256 of the manifest file on disk, which is a third value.",
"selection_rule": "score(ea_name) = sha256(utf8(str(seed)) + b\"\\x00\" + utf8(ea_name)); order by (lowercase hex of score ASC, ea_name ASC); take first n; manifest lists them sorted by ea_name. seed is encoded as its decimal characters (--seed 42 hashes b\"42\"). No RNG.",
"selection_is_deterministic": "No RNG. The manifest states the full rule."
},
"decontamination": {
"against": "the training corpus named above, by SHA-256",
"methods": [
{
"name": "exact",
"definition": "SHA-256 equality over the UTF-8 encoding of the evaluation item's reference solution and each training row's completion field",
"collisions": 0
},
{
"name": "normalised-equivalence",
"definition": "SHA-256 equality after a domain normaliser is applied to both sides. Compares an evaluation item's reference solution against every training row's completion.",
"normaliser_source_sha256": "789bd40d55a5c38ead9c0951fbe38bf1a4aa0b06bcb2e9fdc6a17907d2486794",
"normaliser_source_hash_scope": "SHA-256 of the _normalize_code function source after universal-newline normalisation to LF, from the def _normalize_code line through the line immediately before the next top-level def, with no trailing newline. It is not a hash of the whole source file, and it is not a hash of the raw on-disk bytes, which carry CRLF line endings.",
"normaliser_substitution_count": 13,
"normaliser_substitution_count_meaning": "the number of regex substitution rules in the normaliser's source at the hash above, counted from that source",
"normalisation_steps": [
"strip line and block comments",
"canonicalise EA name string literals",
"canonicalise magic numbers",
"canonicalise timeframe constants",
"canonicalise applied-price enums",
"canonicalise indicator period arguments",
"canonicalise risk, stop-loss, take-profit and trailing multipliers",
"canonicalise instrument symbol literals",
"collapse all whitespace runs to a single space"
],
"steps_are_bound_to_the_hash_above": "The step list describes the normaliser at that source hash. If the hash moves, treat the list as unverified until it is re-read.",
"direction_of_error": "The normaliser is deliberately aggressive: it collapses parameter variation, so two items that differ only in their parameters normalise to the same string. Removing these items therefore removes more than a strict reading requires. The clean stratum is the conservative one.",
"collisions": 116
}
],
"frozen_n": 300,
"clean_n": 184,
"colliding_n": 116,
"no_similarity_cutoff": "No similarity threshold decides membership. The two methods above are the whole predicate.",
"which_side_is_scored": "Both methods above are scored on the reference-solution side: an evaluation item's reference solution against every training row's completion. The specification side - the text actually put to the models - was scanned as well, and only its exact-SHA result is scored, at 0 collisions. The normalised-exact method cannot be scored there, and this is measured rather than assumed: these specifications are entirely `//` comment lines and the normaliser strips comments as its first step, so every specification normalises to the empty string and collides with every other by construction - 300 of 300 rows empty on the frozen 300, one distinct normalised hash, and the same at 184 of 184 on the public 184 against the 220k corpus. The retained scan records the resulting count as an instrument artefact, escalates it rather than scoring it as a stop, and records that which side the stop conditions bind on is unsettled and was not settled there. So the specification-side evidence in this release is the exact-SHA result and nothing else, and the 116 and the 0 in the table above are reference-solution-side figures.",
"reproduction_gate_passed": true,
"reproduction_gate": "What that boolean records. On 2026-08-31 the 2026-08-19 normalised-exact scan of the frozen 300 against the 83k corpus was re-run by a separate tool, on an archived copy of the same training file - same sha256, 5e9881b61f3375d3d575c35950a236375eb5648f5fe55b8d18de321f69659c68 - and the re-run was gated on reproducing the audit's published figures: 116 normalised-exact collisions against 116 published, 0 exact against 0 published, and all 30 preserved witnesses present with their normaliser hashes reproduced, 30 of 30. All four conditions held, which is the `true` above. The gate record is retained privately and identified by SHA-256, 0359e3522235bc364daaedf27e540eb98b2428848500884379b1dc74cd098ff4. This does not make the scan reproducible by a reader: neither corpus is distributed, and the archived copy that re-run read no longer exists either, so 2026-08-31 was the last occasion on which it could be run at all.",
"known_answer_validation": "The scanner's own detection test, and where its record is. Three known-positive fixtures and one falsifying twin were put through the same matcher at the same settings as the scans - threshold 0.7, 256 permutations, shingle k 5, seed 20260812, over a 400-row training sample and 5 evaluation items - and all four results are retained in the evidence package and not distributed: frozen300_selftest_P1_known_positive_verbatim_20260819.json, sha256 1bb4d66d3b708732d763f5b93a94dafcfa76ef2581fd321a4b678e3c888bdfb3, a verbatim duplicate, 5 of 5 exact and 5 of 5 normalised; frozen300_selftest_P2_known_positive_cosmetic_20260819.json, sha256 e8b3327ddb0c55c66c1e41c3b63e3c5afadda53ae26abed0668e4b5f28253933, a cosmetic variant, 0 exact and 5 of 5 normalised; frozen300_selftest_P3_known_positive_crlf_20260819.json, sha256 87666831547d8d8b2d19e6806bdb441596335c84b5fd40f800b55ad7bac036c5, a line-ending variant, 0 exact and 5 of 5 normalised; and frozen300_selftest_TWIN_falsifying_20260819.json, sha256 f075afd1f628a3de8945a030f464967a1700f63f5384fa989f871de2108e9eed, the falsifying twin, 0 exact and 0 normalised. The exact method therefore detects only the verbatim positive, the normalised method detects all three positives, and both reject the twin. Each hash here is a line of the evidence package's public-half manifest, whose own sha256 is published at the top of this document as evidence_manifest_v1_1_0_public_half_sha256; a reader without the package cannot check them.",
"known_answer_validation_scope": "Which scan each set of fixture records belongs to. Those four are the 2026-08-19 scan's. The two later scan runs each kept four of their own, at the same settings and with the same four results, and both sets are retained in the evidence package and not distributed. From the 2026-09-10 run, which covered the public 184 and the 1,000-item EA pool together: c184hb_selftest_P1_known_positive_verbatim_20260910T182010Z.json, sha256 280ae2c546a977e83528bbb0b5c8bc42442d90fd955921aebed375c294c8b883; c184hb_selftest_P2_known_positive_cosmetic_20260910T182010Z.json, sha256 061252cfe079b40e690d4feb903ce37131039c8b03c335fec32e7350ed2abe1f; c184hb_selftest_P3_known_positive_crlf_20260910T182010Z.json, sha256 602e61be3e7a1cc918725f09c15c0567ee155e1ec5f18a842a840f424dc924e9; and c184hb_selftest_TWIN_falsifying_unrelated_20260910T182010Z.json, sha256 112ede73d8dd62357f4a7f6612fe0f37999638914f2d457f6fa71d76f1c79d77. From the 2026-09-11 scan of the 600-item non-EA pool: holdoutC_selftest_P1_known_positive_verbatim_20260911T074334Z.json, sha256 afca32d2ea0817e17f261023a06ca623c223a6ade9872f6cd98da0f81f5e64ec; holdoutC_selftest_P2_known_positive_cosmetic_20260911T074334Z.json, sha256 e797de3fb0aefa23fcaeea9fe4605754279f03f33667945c195122cf21e40025; holdoutC_selftest_P3_known_positive_crlf_20260911T074334Z.json, sha256 65521e5b647a6db58e91406898e8d34a634d267574a11c99e986df256ddac583; and holdoutC_selftest_TWIN_falsifying_unrelated_20260911T074334Z.json, sha256 3495cd7f6d10def6b8284d2ebe58650560fd6e67ad4772c7f3b5c5091f7fcd20. Each of those eight reads 5 of 5 exact and 5 of 5 normalised on the verbatim positive, 0 exact and 5 of 5 normalised on the cosmetic and the line-ending positive, and 0 and 0 on the twin. Two things differ from the 2026-08-19 four: the training sample behind them is 5 rows rather than 400, and each of the two runs also carried three known positives and a falsifying twin through the same streaming pass as the scan itself, against all 215,541 corpus rows, recorded in the scan record as matcher_validation_corpus_canaries and passing on every check. The public-half extract of the 2026-09-10 result, DERIVED_corpus220k_vs_clean184_ONLY_20260910T182010Z.json, sha256 79fe033926ad91929fdd3144dca99033e478f4e742de0bd9cad6993fe699443d, carries that validation as two boolean fields of its own, matcher_validated_fixtures and matcher_validated_corpus_canaries, both true, and names the four fixture records it summarises. A fixture pass shows the matcher catches a known positive and rejects a falsifying twin on a small sample; it does not establish its behaviour at corpus scale, and the witness cap below is a separate and known defect.",
"holdout_pool_scans": "Where the two holdout-pool figures come from. The 1,000-item EA pool and the public 184 were scanned against the 220k corpus in one run on 2026-09-10, retained as corpus220k_vs_clean184_holdoutB_20260910T182010Z.json, sha256 6301ae1f8298ffd96358e03f50dacfac57c9ad3211d4d628039e44ca959523b2: on the EA pool, 0 of 1,000 exact and 352 of 1,000 normalised-exact; on the public 184, 0 exact and 1 of 184 normalised-exact. The 600-item non-EA pool, 150 items in each of its four classes, was scanned on 2026-09-11, retained as corpus220k_vs_holdoutC_20260911T074334Z.json, sha256 bbb28f9bf83fab245d476daf1f745b2159fee66e2af3dfa38bfacd534475b582: 0 of 600 exact and 21 of 600 normalised-exact, the 21 being 20 script items and 1 service item, with none in the include and indicator classes. Those two records are where the 352, the 21 and the split of the 21 across classes are measured. Both also record the specification side, at 0 exact and at 1,000 of 1,000 and 600 of 600 normalised, and neither normalised figure is scored, for the reason under which_side_is_scored above. Neither record removes anything itself: each counts its collisions and halts on them, and the removals in the release card's decontamination table are the two draws', not the scanner's. Both files are retained in the evidence package and are not distributed, and their names carry this project's own working labels for the two pools - holdoutB is the 1,000-item EA pool the 300-item EA holdout is drawn from, holdoutC the 600-item non-EA pool behind the 200-item non-EA holdout - so a counterparty asking for one by name is asking for the file named here. Each hash is a line of the full-package evidence manifest whose own sha256 is published at the top of this document as evidence_manifest_v1_1_0_sha256. That manifest is not distributed either, so these hashes cannot be checked against anything a reader holds; they are what a received file can be checked against.",
"witnesses_preserved": 30,
"witness_cap_is_a_known_defect": {
"preserved": 30,
"collisions_found": 116,
"statement": "The scanner caps preserved witnesses. Every collision was counted; only a subset was preserved as a witness. The count is complete, the witness set is not."
}
},
"models": {
"base": {
"repo_id": "Qwen/Qwen2.5-Coder-14B-Instruct",
"pinned_revision": "aedcc2d42b622764e023cf882b6652e646b95671",
"identity_verdict": "PROVEN_AGAINST_HUGGING_FACE_AT_PINNED_REVISION",
"weight_dtype": "bf16",
"quantisation_method": "none - weights are stored at bf16 precision, not quantised",
"serving_quantisation": "None. The model is loaded with torch_dtype=bfloat16 and no quantisation configuration on this path. The rows' `quantisation` field records the string bf16, which names the dtype, not a quantisation method."
},
"tuned": {
"description": "The base above, fine-tuned on the withheld corpus, then merged to bf16 and served directly. No adapter was applied at serving time.",
"weights_distributed": false,
"identity_note": "The tuned weights are not published. The per-item rows commit to its outputs by hash; they are not a route to the model.",
"weight_dtype": "bf16",
"quantisation_method": "none - weights are stored at bf16 precision, not quantised",
"adapter": {
"adapter_model_sha256": "9cc2d9bc1b75dabca474cb49de2c6aa340ee1185ebf463322cc3341ff0769c3d",
"adapter_config_sha256": "f07d8621efe63994172cc86150fcfcae8bca4959899a90e505258c1334e7d8ce",
"epoch": 1,
"epoch_verdict": "PROVEN",
"epoch_status": "MEASURED",
"epoch_evidence": [
{
"requirement": "trainer_state.json epoch == N.0",
"literal_value": "\"epoch\": 1.0"
},
{
"requirement": "trainer_state.json global_step == max_steps",
"literal_value": "\"global_step\": 4694, \"max_steps\": 4694"
},
{
"requirement": "train_config num_train_epochs consistent with N",
"literal_value": "train_config.json \"epochs\": 1 ; trainer_state.json \"num_train_epochs\": 1"
}
],
"directory_name_disclaimer": "The directory name checkpoint-4694 was not used as evidence. The three literals above were read from the file contents.",
"adapter_selection_note": "A second adapter from a later epoch exists and is the same size. The two are told apart by hash only; a check on the directory name would decide nothing.",
"lora": {
"r": 64,
"lora_alpha": 128,
"lora_dropout": 0.05,
"target_modules": [
"down_proj",
"gate_proj",
"k_proj",
"o_proj",
"q_proj",
"up_proj",
"v_proj"
],
"peft_version_that_wrote_the_adapter": "0.18.1"
}
},
"merge": {
"tool": "merge_adapter.py",
"tool_sha256": "f426179b80cce7c1c9f63fbdc2f058e95562bad9a24e6d57fb4ded07d5ede473",
"tool_sha256_source": "re-derived by this generator from the copy sealed in the evidence package, then checked against the value the merge record wrote",
"dtype": "bfloat16",
"device": "cpu",
"base_identity_asserted_against": "Qwen/Qwen2.5-Coder-14B-Instruct",
"argv_note": "The verbatim argv is four absolute local paths plus the settings above. The paths are withheld; the settings are what determine the output.",
"started_utc": "2026-08-23T03:00:55Z",
"finished_utc": "2026-08-23T03:06:35Z",
"duration_seconds": 339.795
},
"merged_weights": {
"model_safetensors_sha256": "1b8ee53b697e9a600834871a595924b183f8229d6cb8850cd6ba63075ac4c66b",
"model_safetensors_bytes": 29540134824,
"dtype": "bfloat16"
},
"determinism": {
"verdict": "MATCH",
"independent_methods": [
"tools/merge_adapter.py internal hashing",
"GNU coreutils sha256sum, run separately after the merge"
],
"environment": "The 2026-08-20 and 2026-08-23 merges ran on the same interpreter and package set: Python 3.14.0, torch 2.13.0+cpu, transformers 5.15.0, peft 0.20.0, accelerate 1.14.0, safetensors 0.8.0. Byte-identity is therefore evidence of determinism within that environment, not across environments.",
"what_it_establishes": "It establishes that the same base + same adapter + same tool + same package versions + same device and dtype reproduce the same bytes. It does not establish that the artifact is correct for evaluation, and it produces no score.",
"limitation": "Two merges run in 2026 from the same inputs in the same environment produced byte-identical weights. Neither was run at release time, and neither was run in a different environment. Cross-environment reproduction is not established."
},
"quantisation_parity_vs_base_arm": {
"verdict": "MATCH",
"target": "bf16 (bfloat16)",
"comparison_arm_quantisation": "bf16"
},
"served_artifact_binding": "The sealed tuned rows were bound to this merged artifact three ways: the sealed results file is byte-identical to the run directory that served it; that run's parameters name the merged model and its fingerprint, which the run's own console log prints; and the only other tuned run candidate produced no results file at all."
},
"frontier": {
"model_id": "gpt-5.6-sol",
"provider": "openai",
"access": "vendor chat-completions API; weights are vendor-hosted and not independently pinned by this release",
"identity_basis": "model identifier and evaluation date",
"evaluated_on": "2026-09-02",
"temperature": 1.0,
"temperature_note": "the run used temperature 1. This arm is a single run of a sampled decoder.",
"seed": 42,
"seed_note": "sent via the API's seed parameter; determinism is not guaranteed by the vendor",
"applied_settings_note": "What temperature_applied and seed_applied record on this arm. All 184 frontier rows of per_item_results.jsonl carry APPLIED in both fields, and the value is the harness's own: this arm's retained run parameters, at run_params_sha256 below, record temperature_applied APPLIED and seed_applied APPLIED, and the rows carry that forward. The harness derives both from the backend and provider recorded for the arm and from nothing else; it reads no part of the service's response. It is not vendor confirmation, and the retained rows hold none: the only field they carry back from the service is the token usage count. So APPLIED here records how the harness classified its own request, not what the service did with it, and whether the service applied either setting is not recorded - which is what seed_note above and the release card both say. The 1.1.0 local rows record UNKNOWN in these two fields, as serving_v1_1.applied_settings_note explains; the difference between APPLIED here and UNKNOWN there is a difference between two harness classifications and is not evidence about the vendor.",
"max_completion_tokens": 8192,
"generation_cap_note": "The retained run record stores the 8,192-token generation cap under the harness's generic max_new_tokens field; max_completion_tokens is the API parameter it was sent as. max_new_tokens is not the literal API parameter used for this arm.",
"results_rows_sha256": "354f695a34a429d9676279c2a39fbe6c47bd3280054db4c643bac51374fbb4f0",
"run_params_sha256": "1914a9aef38f22cf40b6fead17eae077f584a4a471a5be9b573eaa6b5cfde00b",
"run_params_n_note": "run_params.json records n=300, while the retained frontier results contain 184 rows matching the published clean-184 benchmark. The stale n field is not used in any reported result.",
"served_format": "system and user messages as separate API fields; the local serving template cannot be submitted to the API. The frontier rows' prompt_sha256 hashes the canonical system+user rendering, as defined in PROJECTION_REPORT.json."
},
"arm_field_warning": "The sealed per-item rows of the two local arms carry an internal `arm` field that reads \"A\" on both. It discriminates nothing and is withheld from the public rows, which carry an arm derived from the source directory instead. The frontier arm's sealed rows read \"C\"."
},
"headline": {
"stratum": "CLEAN 184 \u2014 no normalised-exact training twin",
"n": 184,
"base": {
"successes": 2,
"total": 184,
"rate": 0.010869565217391304,
"rate_pct": "1.09%",
"ci95": [
0.0029858430553776835,
0.0387599683090911
],
"ci95_pct": [
"0.30%",
"3.88%"
],
"ci_method": "wilson-score-95"
},
"tuned": {
"successes": 170,
"total": 184,
"rate": 0.9239130434782609,
"rate_pct": "92.39%",
"ci95": [
0.8763482447676944,
0.9541387187164326
],
"ci95_pct": [
"87.63%",
"95.41%"
],
"ci_method": "wilson-score-95"
},
"difference": {
"diff": 0.9130434782608695,
"numerator": "(169 - 1)/184",
"ci95": [
0.8540098171850538,
0.9455338107113938
],
"phi": -0.16760440475893776,
"method": "newcombe-1998-square-and-add-mover-paired",
"diff_pp": "91.30 pp",
"ci95_pp": [
"85.40 pp",
"94.55 pp"
]
},
"mcnemar": {
"b": 1,
"c": 169,
"n_discordant": 170,
"p_value": 2.2852138271288273e-49,
"method": "mcnemar-exact-binomial"
},
"contingency": {
"a_both_pass": 1,
"b_base_only": 1,
"c_tuned_only": 169,
"d_neither": 13,
"n": 184
},
"bootstrap": {
"method": "paired-item-percentile-bootstrap",
"seed": 20260824,
"resamples": 20000,
"n_items": 184,
"point_estimate": 0.9130434782608695,
"ci95": [
0.8695652173913043,
0.9510869565217391
]
},
"denominator_rule": "The denominator is every item of the stratum that both local arms produced a scoreable row for. Measured on these rows: truncation 0, infrastructure failure 0, missing output 0 in both local arms, so the strict and inclusive denominators are identical and the headline is the same number under either reading."
},
"frontier_result": {
"successes": 179,
"total": 184,
"rate": 0.9728260869565217,
"rate_pct": "97.28%",
"ci95": [
0.937973928367375,
0.9883384539803052
],
"ci95_pct": [
"93.80%",
"98.83%"
],
"ci_method": "wilson",
"buckets": {
"truncated": 0,
"infrastructure_failure": 0
},
"paired_vs_tuned": {
"both_pass": 168,
"tuned_only_pass": 2,
"frontier_only_pass": 11,
"both_fail": 3,
"n_discordant": 13,
"mcnemar_p": 0.0224609375,
"method": "mcnemar-exact-binomial",
"gap_pp": "4.89 pp",
"reading": "The frontier arm compiled 11 items the tuned arm missed; the tuned arm compiled 2 items the frontier arm missed; 3 failed under both."
},
"derived": "every figure in this block is derived from the sealed frontier and tuned per-item rows"
},
"full_300_for_comparison": {
"contingency": {
"a_both_pass": 1,
"b_base_only": 1,
"c_tuned_only": 283,
"d_neither": 15,
"n": 300
},
"base": {
"successes": 2,
"total": 300
},
"tuned": {
"successes": 284,
"total": 300
},
"difference": 0.94,
"mcnemar_p": 1.8338173912479853e-83,
"decontamination_cost": -0.027,
"reading": "Decontamination moved the difference from 0.9400 to 0.9130, a change of -0.0270. The uplift is smaller on the decontaminated stratum and the direction is unchanged."
},
"public_release_contents": {
"per_item_rows": {
"rows": 552,
"sha256": "3584aefb4f4226e9ba9ae7453c274eebab6f30b62d0dd89a23f41202c6ec070d",
"distribution": "PUBLIC"
},
"fields_published": [
"item_id",
"verdict_headline",
"verdict_strict",
"bucket",
"ex5_exists",
"errors",
"warnings",
"result_line_errors",
"result_line_warnings",
"compile_infra_reason",
"generation_error",
"positive_control_errors",
"positive_control_warnings",
"truncated",
"finish_reason",
"n_fenced_blocks",
"used_block_index",
"fence_unterminated",
"used_block_unterminated",
"extractor_version",
"harness_version",
"seed",
"seed_applied",
"temperature",
"temperature_applied",
"max_new_tokens",
"quantisation",
"quantisation_mismatch",
"prompt_sha256_public",
"prompt_sha256",
"mq5_sha256",
"ex5_sha256"
],
"fields_withheld": [
"adapter_path",
"adapter_sha256",
"base_model_sha256",
"category",
"compile_log_path",
"compile_log_raw_path",
"compile_log_raw_sha256",
"compile_log_sha256",
"compile_stem",
"ea_name",
"elapsed_seconds",
"ex5_path",
"generated_code_path",
"generated_code_sha256",
"manifest_selection_sha256",
"manifest_sha256",
"metaeditor_build",
"metaeditor_path",
"metaeditor_sha256",
"metaeditor_size_bytes",
"model_id",
"mq5_path",
"source_sha256",
"system_prompt_sha256",
"usage"
],
"item_id_scheme": "Opaque. item_id is a stable public identifier used to join prompts and result rows; it carries none of the item's name.",
"ea_name_not_in_rows_because": "The rows key on item_id so that the row schema does not depend on the corpus's naming convention. This is a schema choice, not anonymisation: every published prompt carries a `// NAME:` line holding that item's original name, so all 184 names are recoverable from this release.",
"generated_mql5_source_distributed": false,
"prompts_distributed": true,
"prompts": {
"rows": 184,
"sha256": "d58391038d39b22d1865e5d874764201f65918fbf92fe837a465f58daee7cfa7",
"distribution": "PUBLIC"
},
"serving_template": {
"sha256": "3601f3ff49381282ee7e3f83e21db9f9e3891910e4092c9461a2a798c9b14756",
"distribution": "PUBLIC"
}
},
"not_distributed": {
"item_id_to_name_mapping": 184,
"training_corpus_rows": 83155,
"generated_mql5_source": "all three arms",
"tuned_model_weights": true,
"release_under": "written agreement only"
},
"contact": {
"route": "Hugging Face - open a discussion on the dataset repository",
"for": "evaluation access, research collaboration, commercial licensing, partnership",
"identity": "CompilingThings"
},
"limitations": [
"The training corpus is not distributed. Any claim about its contents is attested by hash, not independently reproducible from this package.",
"The tuned model weights are not distributed.",
"Generated MQL5 source is not distributed for any arm. The mq5_sha256 and ex5_sha256 values are stable commitments to the withheld artifacts: they permit integrity verification if those artifacts are later disclosed under agreement. mq5_sha256 is present on every row, and on a row where the extractor recovered source it commits to that source; on seven rows it is instead the empty-output sentinel defined under public_release_contents_v1_1, which commits to nothing because no source was recovered. ex5_sha256 is present only on rows where an EX5 artifact was produced and is null on the rest, because a failed compile leaves nothing to hash. They do not let a reader check independently generated output against ours, because two models almost never emit byte-identical source.",
"The result rows carry an opaque item_id rather than the original item name, but the names are not hidden: every published prompt contains a `// NAME:` line, because that line was in the prompt the models were given. The naming scheme for these 184 items is therefore public. An earlier draft of this card claimed otherwise and was wrong.",
"The 184 prompts are now public. They should be treated as a public benchmark from this release onward, not as an uncontaminated evaluation set: once published, prompts enter crawls and downstream corpora regardless of licence. A separate private holdout is reserved for future releases.",
"The compile verdict comes from one compiler on one platform. It measures whether the code compiles, not whether it trades correctly or profitably.",
"The benchmark is a single frozen draw at seed 42 from a larger pool. It was frozen before any result was seen, and was not redrawn.",
"The witness set preserved for the normalised-equivalence collisions is capped and therefore incomplete; the collision count itself is complete.",
"Two historical merges under the same recorded environment produced byte-identical weights: same merge tool, same adapter, same 336 merge targets over 579 base tensors, and an identical model.safetensors hash. A fresh release-time or cross-environment reproduction was not performed.",
"One base-model authority underlies both the shard hash check and the Hub tree check, so those two are not fully independent of each other."
],
"reproducibility": {
"what_a_reader_can_verify": [
"Run the benchmark. The 184 prompts and the serving template are published, so a third party can generate their own arm, compile it, and score it against the same items.",
"Both published hashes: prompt_sha256_public is sha256 of the prompt as published, and prompt_sha256 is arm-specific - the rendered serving template on local-arm rows, the canonical system+user rendering on frontier rows. verify_public_release.py recomputes both under each arm's rule.",
"All three rates, both paired tables, both McNemar p-values and every confidence interval except the bootstrap recompute from the published per-item rows alone.",
"New in 1.1.0: the row counts, arm ids, headline counts, truncation counts and item pairing of holdout_ea300_results.jsonl, holdout_nonea200_results.jsonl and bridge_q8_184_results.jsonl all recompute from those files, and verify_public_release.py checks 9-12 do exactly that. The holdout prompt hashes do not recompute and the holdout rows cannot be regenerated.",
"The frozen selection rule is fully stated and deterministic, so anyone holding the 1.0.0 evaluation pool at the source hash published under `evaluation` can recompute which of its items were drawn. That is the only draw it holds for: each 1.1.0 holdout was taken as a prefix of its pool's frozen order rather than by that rule, and no source hash is published for either holdout pool, so neither holdout draw can be recomputed from this package."
],
"what_a_reader_cannot_do": [
"Inspect the generated MQL5 for any arm.",
"Recompute the decontamination counts, which are measured against a corpus that is not distributed.",
"Reproduce our tuned arm, whose weights are not distributed. A reader can run their own model against the same items and compare.",
"Confirm any per-item verdict against the artifact that produced it. The generated MQL5 and the compiler logs are not distributed, so the rows are an attestation and the statistics computed from them are not."
],
"scope_summary": "The public files support new evaluations on the 184 prompts. The corpus, generated source, tuned weights, and original harness are withheld."
},
"checksum_manifest": {
"file": "SHA256SUMS.txt",
"files_listed": 14,
"files_present": 15,
"reconciliation": "every file except SHA256SUMS.txt, which cannot list itself"
},
"supersedes_release_id": "CompilingThings/compile-benchmark-v1.0.0",
"release_v1_1_note": "Version 1.1.0 adds two private holdout sets (300 EA, 200 non-EA), a Q8_0 bridge over the 184 public prompts, and a row-content hash manifest of the 220k corpus. Every 1.0.0 figure, prompt and result row is carried forward unchanged and nothing from 1.0.0 was re-run.",
"protocol": {
"title": "MQL5 Dataset Evaluation Protocol \u2014 Pre-Registered, Three-Way Compile Benchmark Against a Frontier Model.md",
"sha256": "29e7991af7d7a30120c71a329739de3d29bf4cf4d40c3d343f1327bec80f0323",
"distribution": "retained in the evidence package, not distributed in this release",
"what_rests_on_it": "Every `pre-registered` reference in this document and in PROJECTION_REPORT.json, the strict-denominator rule under statistics_v1_1.denominator_variants, and the prompt byte-identity requirement reported beside each set's spec_identity. Those citations point here rather than each carrying the identity.",
"note": "The title is the file's name in the public half of the retained evidence package and the sha256 is that file's, so a counterparty given the package can check the document they receive is the one cited here against the manifest whose own hash is published above. A reader without the package cannot: this is a commitment, like every other withheld-artifact hash in this release."
},
"public_set_184_composition": {
"prompts_total": 184,
"declare_no_artifact_type_and_carry_a_buy_line": 178,
"declare_an_artifact_type": {
"include": 1,
"indicator": 2,
"script": 3
},
"items_declaring_an_artifact_type": {
"item_009": "script",
"item_089": "script",
"item_093": "indicator",
"item_102": "script",
"item_134": "indicator",
"item_167": "include"
},
"base_arm_passing_items": {
"item_025": "declares no artifact type; carries a `// BUY:` line",
"item_102": "declares `// TYPE: script`"
},
"how_measured": "Read from prompts.jsonl, which ships: an item is counted as declaring a type when its own specification carries a `// TYPE:` header line. No item carries both that header and a `// BUY:` line, and the two counts sum to 184. The passing items are read from per_item_results.jsonl and are the same two in bridge_q8_184_results.jsonl.",
"system_instruction_served": "All 184 were sent the single Expert Advisor system instruction, in 1.0.0 and again in the Q8_0 bridge. None was sent the include, indicator, script or service instruction, and the six that declare another type were neither re-instructed nor excluded. This is checkable from the release: verify_public_release.py check 6 substitutes each item into the published serving template and reproduces prompt_sha256 for all 184, and the bridge rows' prompt_sha256 equal the 1.0.0 local rows' on all 184 items.",
"why_it_is_published": "The set is named for its 1.0.0 lineage, and `EA set` is not what its specifications say. It bears on two published facts. One of the exactly two items the base arm compiles here declares `// TYPE: script`, on a set whose base headline is 2 of 184. And the one specification this set shares with the 200-item non-EA holdout, identified by hash under holdout_sets.holdout_nonea200, is item_093, which declares `// TYPE: indicator`: one specification evaluated under two different system instructions, not two separately authored items."
},
"holdout_sets": {
"manifest_digest_conventions": {
"manifest_file_sha256": "SHA-256 over the bytes of the manifest file itself, the sha256sum convention. This is the value to check a received file against.",
"manifest_sha256_recorded_by_every_arm": "The manifest's own internal manifest_sha256 field, which every arm's run parameters record: SHA-256 of the UTF-8 bytes of json.dumps(the manifest object with its own manifest_sha256 key removed, sort_keys=True, separators=(\",\", \":\"), ensure_ascii=True), with no trailing newline. It covers the provenance fields as well as the item names, and it is not the file hash above.",
"manifest_selection_sha256": "The manifest's own internal selection_sha256 field: SHA-256 of the UTF-8 bytes of json.dumps(sorted(selected_ea_names), separators=(\",\", \":\"), ensure_ascii=True) - the item names alone, sorted, with no path, no seed and no counts, so it is a set identity that is identical across machines. The names are sorted for the hash whether or not the manifest lists them sorted; on the 200 non-EA manifest it lists them in class order and the hash is still over the sorted list.",
"note": "The three are different numbers over different bytes for the same manifest, exactly as model_identities.digest_convention_warning describes for a GGUF. Cite each digest together with its convention."
},
"holdout_ea300": {
"title": "300-item EA holdout: the first 300, in the pool's frozen order, of the 648 items left after removing the 352 of 1,000 with a normalised-exact match to the 220k corpus",
"n_items": 300,
"manifest_file_sha256": "75021a9d8db3fc40931de14239e7e279d37967c109dc0de12c4637473022f64a",
"manifest_sha256_recorded_by_every_arm": "d32a6ca860a6dd7aa0e232396a6b4a1af7cf91140439a4f0d3862cf14f8f89d8",
"manifest_selection_sha256": "20e326804cc23282aed8a5591448b8ccd6d9c51819d732387ac0f89f55d7494f",
"prompts_distributed": false,
"item_names_distributed": false,
"decontamination_scope": "scanned against the 220k corpus only. No scan of this pool against the 83k corpus exists, so the 83k arm's figures on this set carry no decontamination evidence of their own.",
"arms": {
"base": {
"rows": 300,
"sealed_results_sha256": "a9aa3ef6748010e2342b14808856fe4c50f8ea3659b9d45d4293dbe0fc7e3871",
"run_params_sha256": "19fe581c017faad00a27df4169429f8f5025b7264da3da3fabba6600c4c92876"
},
"tuned83k": {
"rows": 300,
"sealed_results_sha256": "b5b4f863d8286330cb395d9ec699bb596c6d4ee22e26b6d02212b129561238b8",
"run_params_sha256": "e1b18ce22466c8103a25b96ef402d59493f1dc0101dfaa3f892da0fc610a0185"
},
"tuned220k": {
"rows": 300,
"sealed_results_sha256": "4b83c2a7f468db83b8544ae0a5cb9d722c952d46790b2b1921cd6b6e8bfd8fe0",
"run_params_sha256": "8cb402cd797b2ce359f92ffd28dfc912bb0796f47543a0b2ba02e63853371496"
},
"frontier": {
"rows": 300,
"sealed_results_sha256": "89f614a8b19c9287df52c9ab338dd11260798cd3d825a049333fa9caa7f2e351",
"run_params_sha256": "b265e7976db2893a5d5f088fea6742db9c84ca83b9967f657d8f942d96fd3e11"
}
}
},
"holdout_nonea200": {
"title": "200-item non-EA holdout, 50 each of include file, custom indicator, script and service: the first 50 of each class, in the pool's own order, after removing the 21 of 600 with a normalised-exact match to the 220k corpus",
"n_items": 200,
"manifest_file_sha256": "db0e0614eed18143e41931353fd47eb4e86422f673b0b44e23ec2eebff090065",
"manifest_sha256_recorded_by_every_arm": "59dcb8aebea6049d7fb16681d485995fb7b699c61b13b89a7fe927060a4fe961",
"manifest_selection_sha256": "655cd809eb68cbd1e3d8d54e5ff883ea525a0f9461728e43e1f5088b89532cc4",
"prompts_distributed": false,
"item_names_distributed": false,
"decontamination_scope": "scanned against the 220k corpus only, on the same terms as the EA holdout.",
"overlap_with_the_public_184": "One specification appears in both the 184 public set and this set. It is one item of 200 and it is reported, not removed. MEASURED on the published rows: the item_sha256 sets of the two holdouts and the prompt_sha256_public set of the 184 intersect in exactly this one value and nowhere else.",
"overlap_with_the_public_184_item_sha256": "8d4946c0c860ecdb2d967b590ae14808e5f748273644b741100b73336aad960d",
"arms": {
"base": {
"rows": 200,
"sealed_results_sha256": "36df1d9a08f30d7b215e07306474ba8322404023788ecab8d6ff51ef0bb06154",
"run_params_sha256": "eb377ab8361659e6af6bc88d4fe862bfc12a2bf1508717f63c7fcf47b34e9c2c"
},
"tuned83k": {
"rows": 200,
"sealed_results_sha256": "2570018f7c374bcfd07adc575b7e38389418f53081a4ccd458ac1175c7204007",
"run_params_sha256": "d008c59615009fd667e070f6e5e1da0169e65b932264d630e54f8b0cbba780ae"
},
"tuned220k": {
"rows": 200,
"sealed_results_sha256": "8b53cca6a6922c946be720c13e46cb9cd09c36bd1fd8f286c35c4589910ca47d",
"run_params_sha256": "335922345f30a546f1115541ce380a98bbb108630e0f8eac4b6c8929ffaadaf4"
},
"tuned220k_q8_replay": {
"rows": 200,
"note": "the stability re-run of the 220k arm on the same 200 items under the same configuration; published as a distinct arm, never merged with or averaged into the first run",
"public_row_extract_sha256": "010489a93192aa19a17bc56950c14ba829c6e14c2395bd73daad8062c775075d"
}
}
},
"attestation": "Attested, not independently reproducible from the public package. The holdout prompts and item names are not published, so a third party can check every holdout arithmetic claim against the published rows and can bind a row to a retained artifact by hash, but cannot regenerate a holdout row from this release."
},
"model_identities": {
"digest_convention_warning": "Two different digests are both called \"the sha256 of a GGUF\" in the retained evidence. gguf_q8_0_file_sha256 is sha256 over the whole file, the sha256sum convention. gguf_q8_0_tensor_rollup_sha256 is the whole-file value reported by llama-gguf-hash, which walks tensors and returns a different number for the same file. Neither is wrong; a citation that does not say which one it means is.",
"row_to_weight_binding": "Every Q8_0 arm in this release records base_model_sha256 as null in its run parameters. The run-time binding between a scored row and the weights that produced it is a recorded model path, not a hash; the GGUF digests below come from the separate merge and quantise logs. This release does not claim that each scored row is hash-bound to its weights.",
"base": {
"role": "base Qwen2.5-Coder-14B-Instruct, served at Q8_0",
"repo_id": "Qwen/Qwen2.5-Coder-14B-Instruct",
"pinned_revision": "aedcc2d42b622764e023cf882b6652e646b95671",
"identity_verdict": "PROVEN_AGAINST_HUGGING_FACE_AT_PINNED_REVISION",
"gguf_bf16_intermediate_sha256": "b3865b9725575f6dfd494c240aa8b414f55b46d72153a0f9b916aec7fc74edc4",
"gguf_q8_0_file_sha256": "adc9f8c0201a910617c60075e6431eb47f4bff37b3a69b5589cbf1c032a83bc6",
"gguf_q8_0_tensor_rollup_sha256": "00d4c901753a6524bc8f2d36bea8c98ef84850fb8a5ba22bfbb169f50b7c88ff",
"weights_distributed": false
},
"tuned_83k": {
"role": "the 1.0.0 fine-tuned model, merged and quantised to Q8_0 for this release",
"training_corpus_rows": 83155,
"training_corpus_sha256": "5e9881b61f3375d3d575c35950a236375eb5648f5fe55b8d18de321f69659c68",
"adapter_model_safetensors_sha256": "9cc2d9bc1b75dabca474cb49de2c6aa340ee1185ebf463322cc3341ff0769c3d",
"adapter_config_sha256": "f07d8621efe63994172cc86150fcfcae8bca4959899a90e505258c1334e7d8ce",
"adapter_epoch": 1,
"adapter_epoch_verdict": "PROVEN",
"adapter_epoch_evidence": {
"trainer_state_epoch": "1.0",
"trainer_state_global_step": 4694,
"trainer_state_max_steps": 4694,
"note": "read from the file contents. The checkpoint directory name was not used as evidence."
},
"lora": {
"r": 64,
"lora_alpha": 128,
"lora_dropout": 0.05,
"peft_version_that_wrote_the_adapter": "0.18.1"
},
"merged_model_safetensors_sha256": "1b8ee53b697e9a600834871a595924b183f8229d6cb8850cd6ba63075ac4c66b",
"gguf_bf16_intermediate_sha256": "504789495c2deb382ddae37d5826837baf028e93dc176b0166724284e6a2cf2c",
"gguf_q8_0_file_sha256": "cc8bf888408c6e5873a530fd933a623515aa0261f666ca05b3e35b752ff1d199",
"gguf_q8_0_tensor_rollup_sha256": "e4c6aac606cf8336501e159e170487ae74ec1cab1a47887046ad48ff8f09cc3e",
"weights_distributed": false
},
"tuned_220k": {
"role": "the same base fine-tuned on the 220k corpus, merged and quantised to Q8_0",
"training_corpus_rows": 215541,
"training_corpus_sha256": "3c2f57db5986eaac49f7ae5ed6732e1ae702b339b1354bf3d94e33c7b611e023",
"adapter_model_safetensors_sha256": "4ea3ebaf9ca45f1ed99c22451d6c6fdb14280513fc76a1acff5468ecd805ba75",
"adapter_config_sha256": "a1a1cefb0d31a3007b2a0b98e777f007b1620610f0f802994ea0cb63b27d97b6",
"adapter_epoch_claimed": 1,
"adapter_epoch_verdict": "NOT ESTABLISHED FROM TRAINER STATE",
"adapter_epoch_note": "The merge record for this adapter states verbatim: \"EPOCH AND STEP ARE NOT ESTABLISHED. This record pins the adapter's bytes; no epoch and no global_step was read, and the directory name is not evidence of either.\" The one-epoch description of this model is the plan it was trained under, not a trainer-state reading. The 83k adapter's epoch is proven from trainer state and that evidence is recorded above.",
"lora": {
"r": 64,
"lora_alpha": 128,
"lora_dropout": 0.05,
"peft_version_that_wrote_the_adapter": "0.18.1"
},
"merged_model_safetensors_sha256": "a42ea6eb126286544778a805afab06d774619f4915273b83296d8a603fd5d86b",
"gguf_bf16_intermediate_sha256": "9f6acbcc1dec43bbf2c06d05eb619e1a0e887c8d6cfaf895b3ded36040a02242",
"gguf_q8_0_file_sha256": "443cb811bc06cf5ed04696233f7cc8fdafb4ad75d18f3f693d8a3848a9be149c",
"merge_utc": {
"started": "2026-09-04T07:28:04Z",
"finished": "2026-09-04T07:30:46Z"
},
"convert_utc": {
"started": "2026-09-04T07:37:07Z",
"finished": "2026-09-04T07:39:41Z"
},
"quantise_utc": {
"started": "2026-09-04T07:40:28Z",
"finished": "2026-09-04T07:41:12Z"
},
"weights_distributed": false
},
"quantisation_parity": {
"target": "Q8_0 on every local arm",
"imatrix_used": false,
"imatrix_rationale": "plain Q8_0 with no calibration data; calibration would make two same-format GGUFs non-comparable",
"intermediate_outtype": "bf16, lossless from a bfloat16 source",
"extra_quantize_flags": "none",
"why_it_matters": "an A/B quantisation mismatch would not degrade the comparison, it would replace the question: B minus A would measure the serving path and read exactly like a fine-tuning result."
}
},
"serving_v1_1": {
"scope": "The three local arms of 1.1.0 - base, tuned 83k, tuned 220k - on every set they were run on: the 300 EA holdout, the 200 non-EA holdout, the non-EA stability re-run and the Q8_0 bridge over the 184 public prompts.",
"engine": "llama.cpp llama-server, Q8_0 GGUF, Vulkan backend, 8 parallel slots, 69,632-token context, all layers offloaded",
"llama_cpp_build": 10441,
"llama_cpp_commit": "0177dcc73",
"llama_cpp_build_note": "Clang 20.1.8, Windows x86_64",
"launch_line": "llama-server.exe -m <gguf> -np 8 -c 69632 -ngl 999 --device Vulkan0 --host 127.0.0.1 --port 8088",
"operating_system": "Windows 11 Pro 10.0.26200",
"graphics_driver": "AMD 32.0.31036.15 (2026-08-12)",
"decoding": "temperature 0.0, top-k 1 and seed 42 sent with every request; greedy",
"max_new_tokens": 8192,
"concurrency": 8,
"decoding_note": "the harness records what it sent. Whether the server applied every setting as sent is not independently verified by this release.",
"applied_settings_note": "Every 1.1.0 local row records temperature_applied and seed_applied as UNKNOWN: all 552 rows of bridge_q8_184_results.jsonl, on all three arms. The holdout row files publish neither field. UNKNOWN is the harness's third value beside APPLIED and UNSUPPORTED and it means the harness makes no claim either way - it classifies applied-status from the generation path recorded in the run parameters, the llama-server runs that produced these rows recorded none, and nothing in that serving path reports back whether a setting took effect. It is not the 1.0.0 arms' UNSUPPORTED, which records a setting that was read and had no effect. What was requested is published on the same rows - temperature 0.0 and seed 42 - and what was sent is under decoding above; whether the server applied it is the gap decoding_note names. The 1.0.0 frontier rows in per_item_results.jsonl read APPLIED in both fields instead; what that value records there, and what it does not, is under models.frontier.applied_settings_note.",
"evaluation_hardware": "one AMD Radeon AI PRO R9700 in an AMD Ryzen 9 7950X3D host with 64 GB of system memory. The device is not recorded in the run parameters; it is stated from the evaluation host's configuration.",
"environment_variables": "none were set for the server. This is recorded, not inferred from a log.",
"library_versions_not_established": "The Python-side library versions in force during the 1.1.0 runs were not recorded and are not established. The server build above is pinned; the client stack is a gap.",
"frontier_arm": {
"scope": "the 300 EA holdout only; no frontier arm ran on the non-EA holdout",
"evaluated_on": "2026-09-11",
"temperature": 1.0,
"seed_requested": 42,
"concurrency": 1,
"max_completion_tokens": 8192,
"note": "served through the vendor API, which applied its own chat framing to the same system-prompt text and item text. A single run of a sampled decoder: a re-run may differ item for item."
}
},
"frontier_evaluations": {
"model_id": "gpt-5.6-sol",
"provider": "openai",
"identity_basis": "model name and evaluation date; the weights are vendor-hosted and are not independently pinned by this release",
"public_184": {
"evaluated_on": "2026-09-02",
"temperature": 1.0,
"seed_requested": 42,
"max_completion_tokens": 8192
},
"holdout_ea300": {
"evaluated_on": "2026-09-11",
"temperature": 1.0,
"seed_requested": 42,
"concurrency": 1,
"max_completion_tokens": 8192
},
"holdout_nonea200": {
"evaluated": false,
"note": "no frontier arm was run on the non-EA holdout, and no frontier figure for that set is estimated or implied"
},
"asymmetry": "the frontier arm sampled at temperature 1.0 one request at a time; the local arms ran greedy at concurrency 8. The head-to-head is published with that asymmetry stated rather than adjusted for."
},
"training_corpus_220k": {
"rows": 215541,
"sha256": "3c2f57db5986eaac49f7ae5ed6732e1ae702b339b1354bf3d94e33c7b611e023",
"distributed": false,
"why_withheld": "the corpus is not distributed. Claims that depend on it are attested, not independently reproducible from the public package."
},
"corpus_row_hash_manifest": {
"file": "corpus_row_hashes.json",
"row_hashes": 219867,
"file_sha256": "4dbabd017b4c50f803a83ec981c99b53b9a2b2247ed1000bde4f571f21e1bcfd",
"source_manifest_internal_sha256": "4c10a18620126e2379bca8c5c03d33c298fa5f4abc7b31665ff99893cfc18647",
"source_manifest_internal_sha256_convention": "The private source manifest's own internal manifest_sha256 field, not the sha256 of any file: SHA-256 of the UTF-8 bytes of json.dumps(the manifest object with its own manifest_sha256 key removed, sort_keys=True, separators=(\",\", \":\"), ensure_ascii=True). It covers the construction counts and internal labels this projection drops as well as the row hashes it keeps, so it cannot be recomputed from corpus_row_hashes.json and is a commitment to the source, not a check on this file.",
"projection_note": "This file is a projection of a retained extract, not a copy of it. The extract's header carried construction counts and internal labels that are not published, and each of its entries carried a positional index into a private build; what survives is the schema version, the committed row count, the hash rule, the trained-corpus pin and the 219,867 row hashes, carried across in the extract's own order with no value altered. The extract is CRLF and this file is LF, like every other file in the release, because .gitattributes pins LF and a CRLF file would be rewritten on checkout and stop matching its manifest hash. The retained extract's own sha256 is 3660bdfb621a9e222fdb2485b906223933edbe8f6e46f7fdd98aac7eef552490.",
"contents_published": false,
"names_published": false,
"relation_to_the_trained_corpus": "The 219,867 hashes are the corpus before length filtering. The corpus the 220k arm trained on is the length-filtered pin of 215,541 rows. The relation is a declared filter - quarantine if and only if the rendered token length exceeds the maximum, and admitted rows are never truncated - recorded in the split manifest. It was not verified row by row and no row-level mapping between the two sets exists in this release.",
"entry_shape": "Every entry carries one key, sha256, and nothing else. The order is the retained extract's order and carries no published meaning.",
"who_can_verify_it": "The length threshold is not published and this release did not check the subset row by row, so the check that holds is one-directional: every row of the 215,541-row trained corpus hashes to a member of this manifest, and the 4,326 remaining hashes are the rows the length rule dropped. A counterparty holding the corpus can make that check; no one else can."
},
"public_release_contents_v1_1": {
"holdout_ea300_results.jsonl": {
"rows": 1200,
"arms": {
"base": 300,
"tuned83k": 300,
"tuned220k": 300,
"frontier": 300
},
"distribution": "PUBLIC"
},
"holdout_nonea200_results.jsonl": {
"rows": 800,
"arms": {
"base": 200,
"tuned83k": 200,
"tuned220k": 200,
"tuned220k_q8_replay": 200
},
"distribution": "PUBLIC"
},
"bridge_q8_184_results.jsonl": {
"rows": 552,
"arms": {
"base": 184,
"tuned83k": 184,
"tuned220k": 184
},
"distribution": "PUBLIC"
},
"corpus_row_hashes.json": {
"row_hashes": 219867,
"distribution": "PUBLIC"
},
"holdout_row_fields_published": [
"arm",
"item_sha256",
"prompt_sha256",
"category",
"verdict_headline",
"verdict_strict",
"bucket",
"errors",
"warnings",
"truncated",
"finish_reason",
"mq5_sha256",
"ex5_exists",
"ex5_sha256"
],
"holdout_row_fields_withheld": [
"adapter_path",
"adapter_sha256",
"base_model_sha256",
"compile_infra_reason",
"compile_log_path",
"compile_log_raw_path",
"compile_log_raw_sha256",
"compile_log_sha256",
"compile_stem",
"ea_name",
"elapsed_seconds",
"ex5_path",
"extractor_version",
"fence_unterminated",
"generated_code_path",
"generated_code_sha256",
"generation_error",
"harness_version",
"manifest_selection_sha256",
"manifest_sha256",
"max_new_tokens",
"metaeditor_build",
"metaeditor_path",
"metaeditor_sha256",
"metaeditor_size_bytes",
"model_id",
"mq5_path",
"n_fenced_blocks",
"positive_control_errors",
"positive_control_warnings",
"quantisation",
"quantisation_mismatch",
"result_line_errors",
"result_line_warnings",
"seed",
"seed_applied",
"source_sha256",
"system_instruction",
"system_instruction_sha256",
"system_prompt_sha256",
"temperature",
"temperature_applied",
"usage",
"used_block_index",
"used_block_unterminated"
],
"holdout_row_fields_withheld_scope": "Measured, not curated: every key present on any sealed row of the eight arms behind the two holdout files, less the fourteen this release publishes. The sealed `arm` field is withheld too and is not in the list, because the published `arm` is derived from the source extract and the two are different fields - see PROJECTION_REPORT.json arm_derivation_v1_1. The prompt text itself is not a row field in any of them and is withheld with the item names.",
"mq5_sha256_empty_output_sentinel": {
"value": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
"meaning": "the SHA-256 of zero bytes. Where mq5_sha256 holds it, no source was generated or recovered for that row and the row binds no generated MQL5 artifact: it is not a commitment and three rows sharing it are not three copies of one output.",
"rows": 7,
"where": "all seven are frontier rows of holdout_ea300_results.jsonl - the three generation failures, which carry null verdicts, and the four generations stopped at the 8,192-token ceiling, which are scored as failures. No row of any other file carries it, and every one of the 55 truncated local-arm rows in this release carries a real source hash, so truncation alone does not produce it.",
"what_those_rows_do_bind": "item_sha256 and prompt_sha256, which identify the item and the request put to the arm, and the row's own finish_reason and bucket, which record how the generation ended. The four truncated rows also carry errors 1 and warnings 0, which is the compile result recorded against a source that is not there rather than evidence of a generated artifact; the three generation failures carry null in errors, in warnings and in both verdict fields."
},
"bridge_row_fields_published": [
"arm",
"item_id",
"prompt_sha256_public",
"verdict_headline",
"verdict_strict",
"bucket",
"ex5_exists",
"errors",
"warnings",
"result_line_errors",
"result_line_warnings",
"compile_infra_reason",
"generation_error",
"positive_control_errors",
"positive_control_warnings",
"truncated",
"finish_reason",
"n_fenced_blocks",
"used_block_index",
"fence_unterminated",
"used_block_unterminated",
"extractor_version",
"harness_version",
"seed",
"seed_applied",
"temperature",
"temperature_applied",
"max_new_tokens",
"quantisation",
"quantisation_mismatch",
"prompt_sha256",
"mq5_sha256",
"ex5_sha256"
],
"field_inventory_is_in": "PROJECTION_REPORT.json"
},
"statistics_v1_1": {
"source": {
"sha256": "027f3775194a7c895bd8de38f4d6dec9a630ce40cbf48b8f50ba0ca175acdb1d",
"note": "Full-precision statistics as the statistician computed them, identified by the sha256 above. Every per-item name list has been removed and every field whose value was a filesystem path has been dropped: the holdout item names are not published. Some prose was rewritten for publication - internal set labels replaced by this release's own terms, emphasis normalised - and no figure was recomputed, rounded or re-derived. One value differs from the source deliberately: the Wilson lower bound of the base arm's 0-of-50 indicator cell on the 200-item non-EA set, which the source carries as 6.938893903907228e-18 and this release publishes as 0.0, in the primary block and in its secondary_spec_conformance mirror. The exact bound at 0 successes is zero and the sibling cells print 0.0; that value is binary floating-point residue. It is the only leaf in this block whose value was changed and the change is 7e-18."
},
"generated_utc": "2026-09-12T07:58:29+00:00",
"nothing_rerun": "no arm was re-run, no item re-scored, no sample re-drawn",
"read_only": true,
"verdict_field": "verdict_headline",
"secondary_field": "verdict_strict",
"denominator_rule": "The denominator is every item present in the frozen manifest and present, by the item's name in the sealed rows, in every arm being compared. An item is excluded only when (a) it is in the contamination exclusion list for that set (the `clean` pool; the `full` pool keeps it), or (b) its preserved row shows truncation, an infrastructure compile failure, or a generation error (the `strict` variant; the `inclusive` variant keeps it and scores it as a failure). Both variants are computed and published side by side for every rate and every paired difference. The strict exclusion is scoped: a per-arm rate drops only that arm's own truncated, infrastructure or generation-error items, and a paired test drops an item truncated in either of the two arms being compared, so the base arm's truncations do not shrink a tuned-vs-tuned pair.",
"denominator_variants": {
"inclusive": "every aligned item counts; truncated and infrastructure items are scored as failures",
"strict": "truncated, infrastructure and generation-error items are removed from the denominator (the pre-registered protocol, identified under the top-level `protocol` field: never silently scored as a model failure)"
},
"bootstrap": {
"seed": 20260911,
"resamples": 20000
},
"precision_checks": {
"full_184_vs_clean_183": {
"per_arm": {
"base_q8": {
"full_strict_kn": "2/182 (1.1%)",
"clean_strict_kn": "2/181 (1.1%)",
"delta_pct_points": 0.006071276789508806
},
"tuned83k_q8": {
"full_strict_kn": "175/184 (95.1%)",
"clean_strict_kn": "174/183 (95.1%)",
"delta_pct_points": -0.026728439059155562
},
"tuned220k_q8": {
"full_strict_kn": "169/184 (91.8%)",
"clean_strict_kn": "168/183 (91.8%)",
"delta_pct_points": -0.044547398431930674
},
"frontier_gpt56sol": {
"full_strict_kn": "179/184 (97.3%)",
"clean_strict_kn": "178/183 (97.3%)",
"delta_pct_points": -0.014849132810638821
}
},
"max_abs_delta_pct_points": 0.044547398431930674,
"MATERIAL_DISAGREEMENT": false,
"note": "Denominator ruling: on the public 184 the release publishes the full-184 figures - the 220k arm at 169/184 - and the clean-183 figures beside them are a precision check only. The one contamination collision is reported, not removed."
},
"replay_19_subset_vs_full_184": {
"subset_n": 19,
"subset_same_items_in_full_run_kn": "18/19 (94.7%)",
"subset_replay_kn": "18/19 (94.7%)",
"full_set_kn": "175/184 (95.1%)",
"n_verdict_flips": 0,
"note": "a 19-item cell is thin and indicative only; it is a stability instrument, not an estimate of the arm's rate"
},
"holdout_300_prefix_vs_clean_648": {
"STATUS": "not computable",
"why": "the 300 EA holdout is the first 300 of the clean 648 in frozen order; the remaining 348 items are held in reserve, unrun and unpublished, so no sample-vs-full-set comparison exists for this set. Reported as not computable rather than silently omitted.",
"reserve_items_unrun": 348
},
"legacy_184_vs_holdout_300_same_arms": {
"FLAG": "not a subset-vs-full precision check - these are different item sets. Published so a reader can see the two sets side by side with counts.",
"per_arm": {
"base_q8": {
"legacy_184_full_strict_kn": "2/182 (1.1%)",
"holdout_300_full_strict_kn": "1/297 (0.3%)",
"delta_pct_points": -0.7622007622007623
},
"tuned83k_q8": {
"legacy_184_full_strict_kn": "175/184 (95.1%)",
"holdout_300_full_strict_kn": "281/300 (93.7%)",
"delta_pct_points": -1.4420289855072355
},
"tuned220k_q8": {
"legacy_184_full_strict_kn": "169/184 (91.8%)",
"holdout_300_full_strict_kn": "282/300 (94.0%)",
"delta_pct_points": 2.152173913043484
},
"frontier_gpt56sol": {
"legacy_184_full_strict_kn": "179/184 (97.3%)",
"holdout_300_full_strict_kn": "286/293 (97.6%)",
"delta_pct_points": 0.32831280605431346
}
},
"max_abs_delta_pct_points": 2.152173913043484
},
"nonea_c200_200_of_600_vs_the_full_non_ea_pool": {
"STATUS": "not computable",
"why": "the 200 non-EA holdout is a derived subset of the 600-item non-EA pool (per class, in the pool's own order, minus the pinned normalised-exact completion collisions, first 50 of each class). The remaining 400 pool items have never been run on any arm, so there is no full-set figure to compare the 200-item draw against. Reported as not computable rather than silently omitted. No re-draw was made and none may be.",
"reserve_items_unrun": 400,
"draw_rule": "the first 50 items of each of the four non-EA classes, in the pool's own order, after removing the items carrying a pinned normalised-exact completion collision with the training corpus."
},
"nonea_c200_per_class_dispersion": {
"FLAG": "not a subset-vs-full precision check - it is the spread of the four 50-item class cells around the 200-item aggregate for the same arm. Published because a balanced aggregate can hide a dead class.",
"per_arm": {
"base_q8": {
"aggregate_strict_kn": "56/186 (30.1%)",
"per_class_strict_kn": {
"include": "20/43 (46.5%)",
"indicator": "0/45 (0.0%)",
"script": "13/48 (27.1%)",
"service": "23/50 (46.0%)"
},
"max_abs_class_minus_aggregate_pct_points": 30.107526881720432,
"worst_class": "indicator"
},
"tuned83k_q8": {
"aggregate_strict_kn": "135/187 (72.2%)",
"per_class_strict_kn": {
"include": "38/42 (90.5%)",
"indicator": "49/50 (98.0%)",
"script": "19/45 (42.2%)",
"service": "29/50 (58.0%)"
},
"max_abs_class_minus_aggregate_pct_points": 29.97029114676173,
"worst_class": "script"
},
"tuned220k_q8": {
"aggregate_strict_kn": "168/188 (89.4%)",
"per_class_strict_kn": {
"include": "31/39 (79.5%)",
"indicator": "50/50 (100.0%)",
"script": "39/49 (79.6%)",
"service": "48/50 (96.0%)"
},
"max_abs_class_minus_aggregate_pct_points": 10.63829787234043,
"worst_class": "include"
}
}
},
"EA_holdout_300_vs_nonEA_C200_same_arms": {
"FLAG": "not a precision check - different item sets and different task families, EA against non-EA. Published with counts so a reader can see both sets side by side; it is not a paired comparison.",
"per_arm": {
"base_q8": {
"EA_holdout_300_strict_kn": "1/297 (0.3%)",
"nonEA_C200_strict_kn": "56/186 (30.1%)",
"delta_pct_points": 29.770826545020096
},
"tuned83k_q8": {
"EA_holdout_300_strict_kn": "281/300 (93.7%)",
"nonEA_C200_strict_kn": "135/187 (72.2%)",
"delta_pct_points": -21.47415329768272
},
"tuned220k_q8": {
"EA_holdout_300_strict_kn": "282/300 (94.0%)",
"nonEA_C200_strict_kn": "168/188 (89.4%)",
"delta_pct_points": -4.638297872340431
}
}
}
},
"q8_bridge": {
"title": "Part 5 \u2014 Q8 protocol bridge (83k BF16 vs 83k Q8_0 on the same 184 items)",
"CAVEAT": "The bf16 arm is the 1.0.0 tuned arm as published, identified by the 1.0.0 weight hashes; this release did not re-verify those weights. That identity is recorded, not verified, and the run is labelled arm 'A' in its own run parameters, so the bridge is indicative rather than attested.",
"bf16_results_sha256": "bbeadfe8ce96e92bf7183e0eddbb94ad087711ccb39a386118b7e61bd3378dd8",
"bf16_rows_total": 300,
"n_items_joined": 184,
"q8_pass_kn": "175/184 (95.1%)",
"bf16_pass_kn": "170/184 (92.4%)",
"marginal_item_difference": 5,
"confounds_named": "the BF16 arm ran on a different serving backend (local transformers) from the Q8 arm (llama-server, -np 8, concurrency 8). A difference here is not attributable to quantisation alone.",
"paired": {
"inclusive": {
"arm_x": "83k Q8_0",
"arm_y": "83k BF16",
"n_pairs": 184,
"contingency": {
"a_both_pass": 169,
"b_x_pass_y_fail": 6,
"c_x_fail_y_pass": 1,
"d_neither": 8
},
"x_rate_kn": "175/184 (95.1%)",
"y_rate_kn": "170/184 (92.4%)",
"diff_x_minus_y": 0.02717391304347826,
"diff_pct_points": 2.717391304347826,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 6,
"c": 1,
"n_discordant": 7,
"p_value": 0.125
},
"ci95_diff_newcombe": [
-0.0026328202453298165,
0.0628229813724678
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.0,
"hi": 0.05434782608695652
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "83k Q8_0",
"arm_y": "83k BF16",
"n_pairs": 184,
"contingency": {
"a_both_pass": 169,
"b_x_pass_y_fail": 6,
"c_x_fail_y_pass": 1,
"d_neither": 8
},
"x_rate_kn": "175/184 (95.1%)",
"y_rate_kn": "170/184 (92.4%)",
"diff_x_minus_y": 0.02717391304347826,
"diff_pct_points": 2.717391304347826,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 6,
"c": 1,
"n_discordant": 7,
"p_value": 0.125
},
"ci95_diff_newcombe": [
-0.0026328202453298165,
0.0628229813724678
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.0,
"hi": 0.05434782608695652
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
}
},
"replay_19": {
"title": "Release-stability instrument \u2014 19-item replay, 83k Q8_0",
"replay_results_sha256": "558ec975001179ec1a2772bf6fe197d2c56359ea2561192816985569ce026ca6",
"n_items_joined": 19,
"replay_pass_kn": "18/19 (94.7%)",
"reference_pass_kn": "18/19 (94.7%)",
"verdict_flips": [],
"n_verdict_flips": 0,
"generation_bytes_identical_kn": "18/19 (94.7%)",
"generation_bytes_differing": 1,
"meaning": "One of 19 generation hashes differed under the same recorded settings, and no verdict flipped; this release does not isolate why. On a row where generation produced source, the output hash is provenance of the observed run, not a promise that a re-generation reproduces it."
},
"secondary_spec_conformance_flag": "Every secondary_spec_conformance block in this statistics section is self-declared and non-independent: the verdict_strict figures in it come from this project's own verifier, not from an external judgement. Each block points back here rather than repeating the caveat.",
"sets": {
"legacy_184": {
"title": "Part 1 \u2014 legacy 184-item public benchmark (historical comparability)",
"manifest_file_sha256": "6d09240ae5a42493516272ab9645fb38afc3381d29ad0bdbd1432b1d63c8da24",
"per_arm": {
"full": {
"base_q8": {
"inclusive": {
"k": 2,
"n": 184,
"rate_pct": 1.0869565217391304,
"kn": "2/184 (1.1%)",
"ci95": [
0.002985905599769044,
0.03875918555276116
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 2,
"n": 184,
"kn": "2/184 (1.1%)",
"ci95": [
0.002985905599769044,
0.03875918555276116
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 2,
"n": 182,
"rate_pct": 1.098901098901099,
"kn": "2/182 (1.1%)",
"ci95": [
0.003018774375770715,
0.039175571656886184
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 2,
"n": 182,
"kn": "2/182 (1.1%)",
"ci95": [
0.003018774375770715,
0.039175571656886184
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned83k_q8": {
"inclusive": {
"k": 175,
"n": 184,
"rate_pct": 95.1086956521739,
"kn": "175/184 (95.1%)",
"ci95": [
0.9096677238294343,
0.9740562476632323
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 175,
"n": 184,
"kn": "175/184 (95.1%)",
"ci95": [
0.9096677238294343,
0.9740562476632323
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 175,
"n": 184,
"rate_pct": 95.1086956521739,
"kn": "175/184 (95.1%)",
"ci95": [
0.9096677238294343,
0.9740562476632323
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 175,
"n": 184,
"kn": "175/184 (95.1%)",
"ci95": [
0.9096677238294343,
0.9740562476632323
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned220k_q8": {
"inclusive": {
"k": 169,
"n": 184,
"rate_pct": 91.84782608695652,
"kn": "169/184 (91.8%)",
"ci95": [
0.8698640135719263,
0.9499762973309088
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 169,
"n": 184,
"kn": "169/184 (91.8%)",
"ci95": [
0.8698640135719263,
0.9499762973309088
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 169,
"n": 184,
"rate_pct": 91.84782608695652,
"kn": "169/184 (91.8%)",
"ci95": [
0.8698640135719263,
0.9499762973309088
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 169,
"n": 184,
"kn": "169/184 (91.8%)",
"ci95": [
0.8698640135719263,
0.9499762973309088
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"frontier_gpt56sol": {
"inclusive": {
"k": 179,
"n": 184,
"rate_pct": 97.28260869565217,
"kn": "179/184 (97.3%)",
"ci95": [
0.9379747965705482,
0.988338281982006
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 176,
"n": 184,
"kn": "176/184 (95.7%)",
"ci95": [
0.9165644224588859,
0.9778068257987526
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 179,
"n": 184,
"rate_pct": 97.28260869565217,
"kn": "179/184 (97.3%)",
"ci95": [
0.9379747965705482,
0.988338281982006
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 176,
"n": 184,
"kn": "176/184 (95.7%)",
"ci95": [
0.9165644224588859,
0.9778068257987526
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
}
},
"clean": {
"base_q8": {
"inclusive": {
"k": 2,
"n": 183,
"rate_pct": 1.092896174863388,
"kn": "2/183 (1.1%)",
"ci95": [
0.003002250027948822,
0.038966266329115845
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 2,
"n": 183,
"kn": "2/183 (1.1%)",
"ci95": [
0.003002250027948822,
0.038966266329115845
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 2,
"n": 181,
"rate_pct": 1.1049723756906078,
"kn": "2/181 (1.1%)",
"ci95": [
0.003035481630522502,
0.03938713757365047
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 2,
"n": 181,
"kn": "2/181 (1.1%)",
"ci95": [
0.003035481630522502,
0.03938713757365047
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned83k_q8": {
"inclusive": {
"k": 174,
"n": 183,
"rate_pct": 95.08196721311475,
"kn": "174/183 (95.1%)",
"ci95": [
0.9091883431019155,
0.9739133038315165
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 174,
"n": 183,
"kn": "174/183 (95.1%)",
"ci95": [
0.9091883431019155,
0.9739133038315165
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 174,
"n": 183,
"rate_pct": 95.08196721311475,
"kn": "174/183 (95.1%)",
"ci95": [
0.9091883431019155,
0.9739133038315165
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 174,
"n": 183,
"kn": "174/183 (95.1%)",
"ci95": [
0.9091883431019155,
0.9739133038315165
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned220k_q8": {
"inclusive": {
"k": 168,
"n": 183,
"rate_pct": 91.80327868852459,
"kn": "168/183 (91.8%)",
"ci95": [
0.8691766392132034,
0.9496994333977973
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 168,
"n": 183,
"kn": "168/183 (91.8%)",
"ci95": [
0.8691766392132034,
0.9496994333977973
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 168,
"n": 183,
"rate_pct": 91.80327868852459,
"kn": "168/183 (91.8%)",
"ci95": [
0.8691766392132034,
0.9496994333977973
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 168,
"n": 183,
"kn": "168/183 (91.8%)",
"ci95": [
0.8691766392132034,
0.9496994333977973
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"frontier_gpt56sol": {
"inclusive": {
"k": 178,
"n": 183,
"rate_pct": 97.26775956284153,
"kn": "178/183 (97.3%)",
"ci95": [
0.93764444316215,
0.9882742533195696
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 175,
"n": 183,
"kn": "175/183 (95.6%)",
"ci95": [
0.9161212614482003,
0.9776846478723037
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 178,
"n": 183,
"rate_pct": 97.26775956284153,
"kn": "178/183 (97.3%)",
"ci95": [
0.93764444316215,
0.9882742533195696
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 175,
"n": 183,
"kn": "175/183 (95.6%)",
"ci95": [
0.9161212614482003,
0.9776846478723037
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
}
}
},
"per_class": null,
"paired": {
"full": {
"tuned83k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 184,
"contingency": {
"a_both_pass": 1,
"b_x_pass_y_fail": 174,
"c_x_fail_y_pass": 1,
"d_neither": 8
},
"x_rate_kn": "175/184 (95.1%)",
"y_rate_kn": "2/184 (1.1%)",
"diff_x_minus_y": 0.9402173913043478,
"diff_pct_points": 94.02173913043478,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 174,
"c": 1,
"n_discordant": 175,
"p_value": 7.350102952753538e-51
},
"ci95_diff_newcombe": [
0.8854460655822902,
0.966085147345344
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.9021739130434783,
"hi": 0.9728260869565217
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 182,
"contingency": {
"a_both_pass": 1,
"b_x_pass_y_fail": 173,
"c_x_fail_y_pass": 1,
"d_neither": 7
},
"x_rate_kn": "174/182 (95.6%)",
"y_rate_kn": "2/182 (1.1%)",
"diff_x_minus_y": 0.945054945054945,
"diff_pct_points": 94.50549450549451,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 173,
"c": 1,
"n_discordant": 174,
"p_value": 1.4616682008316695e-50
},
"ci95_diff_newcombe": [
0.890668012548651,
0.9696911761441854
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.9065934065934066,
"hi": 0.978021978021978
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
},
"tuned220k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 184,
"contingency": {
"a_both_pass": 1,
"b_x_pass_y_fail": 168,
"c_x_fail_y_pass": 1,
"d_neither": 14
},
"x_rate_kn": "169/184 (91.8%)",
"y_rate_kn": "2/184 (1.1%)",
"diff_x_minus_y": 0.907608695652174,
"diff_pct_points": 90.76086956521739,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 168,
"c": 1,
"n_discordant": 169,
"p_value": 4.5437000071567326e-49
},
"ci95_diff_newcombe": [
0.8478098364646453,
0.9412821039960407
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.8586956521739131,
"hi": 0.9510869565217391
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 182,
"contingency": {
"a_both_pass": 1,
"b_x_pass_y_fail": 167,
"c_x_fail_y_pass": 1,
"d_neither": 13
},
"x_rate_kn": "168/182 (92.3%)",
"y_rate_kn": "2/182 (1.1%)",
"diff_x_minus_y": 0.9120879120879121,
"diff_pct_points": 91.20879120879121,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 167,
"c": 1,
"n_discordant": 168,
"p_value": 9.033944720111621e-49
},
"ci95_diff_newcombe": [
0.8524578087835255,
0.944927036904818
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.8681318681318682,
"hi": 0.9560439560439561
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
},
"tuned220k_q8__vs__tuned83k_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 184,
"contingency": {
"a_both_pass": 163,
"b_x_pass_y_fail": 6,
"c_x_fail_y_pass": 12,
"d_neither": 3
},
"x_rate_kn": "169/184 (91.8%)",
"y_rate_kn": "175/184 (95.1%)",
"diff_x_minus_y": -0.03260869565217391,
"diff_pct_points": -3.260869565217391,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 6,
"c": 12,
"n_discordant": 18,
"p_value": 0.237884521484375
},
"ci95_diff_newcombe": [
-0.08185163231317105,
0.013900964632705883
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.07608695652173914,
"hi": 0.010869565217391304
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 184,
"contingency": {
"a_both_pass": 163,
"b_x_pass_y_fail": 6,
"c_x_fail_y_pass": 12,
"d_neither": 3
},
"x_rate_kn": "169/184 (91.8%)",
"y_rate_kn": "175/184 (95.1%)",
"diff_x_minus_y": -0.03260869565217391,
"diff_pct_points": -3.260869565217391,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 6,
"c": 12,
"n_discordant": 18,
"p_value": 0.237884521484375
},
"ci95_diff_newcombe": [
-0.08185163231317105,
0.013900964632705883
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.07608695652173914,
"hi": 0.010869565217391304
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
},
"tuned220k_q8__vs__frontier_gpt56sol": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "C frontier gpt-5.6-sol (vendor-hosted)",
"n_pairs": 184,
"contingency": {
"a_both_pass": 167,
"b_x_pass_y_fail": 2,
"c_x_fail_y_pass": 12,
"d_neither": 3
},
"x_rate_kn": "169/184 (91.8%)",
"y_rate_kn": "179/184 (97.3%)",
"diff_x_minus_y": -0.05434782608695652,
"diff_pct_points": -5.434782608695652,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 2,
"c": 12,
"n_discordant": 14,
"p_value": 0.012939453125
},
"ci95_diff_newcombe": [
-0.10046019777816709,
-0.015470146379077683
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.09239130434782608,
"hi": -0.016304347826086956
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "C frontier gpt-5.6-sol (vendor-hosted)",
"n_pairs": 184,
"contingency": {
"a_both_pass": 167,
"b_x_pass_y_fail": 2,
"c_x_fail_y_pass": 12,
"d_neither": 3
},
"x_rate_kn": "169/184 (91.8%)",
"y_rate_kn": "179/184 (97.3%)",
"diff_x_minus_y": -0.05434782608695652,
"diff_pct_points": -5.434782608695652,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 2,
"c": 12,
"n_discordant": 14,
"p_value": 0.012939453125
},
"ci95_diff_newcombe": [
-0.10046019777816709,
-0.015470146379077683
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.09239130434782608,
"hi": -0.016304347826086956
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
},
"tuned83k_q8__vs__frontier_gpt56sol": {
"inclusive": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "C frontier gpt-5.6-sol (vendor-hosted)",
"n_pairs": 184,
"contingency": {
"a_both_pass": 173,
"b_x_pass_y_fail": 2,
"c_x_fail_y_pass": 6,
"d_neither": 3
},
"x_rate_kn": "175/184 (95.1%)",
"y_rate_kn": "179/184 (97.3%)",
"diff_x_minus_y": -0.021739130434782608,
"diff_pct_points": -2.1739130434782608,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 2,
"c": 6,
"n_discordant": 8,
"p_value": 0.2890625
},
"ci95_diff_newcombe": [
-0.05925509430454449,
0.010795792178232518
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.05434782608695652,
"hi": 0.005434782608695652
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "C frontier gpt-5.6-sol (vendor-hosted)",
"n_pairs": 184,
"contingency": {
"a_both_pass": 173,
"b_x_pass_y_fail": 2,
"c_x_fail_y_pass": 6,
"d_neither": 3
},
"x_rate_kn": "175/184 (95.1%)",
"y_rate_kn": "179/184 (97.3%)",
"diff_x_minus_y": -0.021739130434782608,
"diff_pct_points": -2.1739130434782608,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 2,
"c": 6,
"n_discordant": 8,
"p_value": 0.2890625
},
"ci95_diff_newcombe": [
-0.05925509430454449,
0.010795792178232518
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.05434782608695652,
"hi": 0.005434782608695652
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
}
},
"clean": {
"tuned83k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 183,
"contingency": {
"a_both_pass": 1,
"b_x_pass_y_fail": 173,
"c_x_fail_y_pass": 1,
"d_neither": 8
},
"x_rate_kn": "174/183 (95.1%)",
"y_rate_kn": "2/183 (1.1%)",
"diff_x_minus_y": 0.9398907103825137,
"diff_pct_points": 93.98907103825137,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 173,
"c": 1,
"n_discordant": 174,
"p_value": 1.4616682008316695e-50
},
"ci95_diff_newcombe": [
0.8848375909442479,
0.9658980334464432
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.9016393442622951,
"hi": 0.9726775956284153
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 181,
"contingency": {
"a_both_pass": 1,
"b_x_pass_y_fail": 172,
"c_x_fail_y_pass": 1,
"d_neither": 7
},
"x_rate_kn": "173/181 (95.6%)",
"y_rate_kn": "2/181 (1.1%)",
"diff_x_minus_y": 0.9447513812154696,
"diff_pct_points": 94.47513812154696,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 172,
"c": 1,
"n_discordant": 173,
"p_value": 2.906631622225263e-50
},
"ci95_diff_newcombe": [
0.8900805879545658,
0.9695222065436557
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.9060773480662984,
"hi": 0.9779005524861878
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
},
"tuned220k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 183,
"contingency": {
"a_both_pass": 1,
"b_x_pass_y_fail": 167,
"c_x_fail_y_pass": 1,
"d_neither": 14
},
"x_rate_kn": "168/183 (91.8%)",
"y_rate_kn": "2/183 (1.1%)",
"diff_x_minus_y": 0.907103825136612,
"diff_pct_points": 90.7103825136612,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 167,
"c": 1,
"n_discordant": 168,
"p_value": 9.033944720111621e-49
},
"ci95_diff_newcombe": [
0.8470049904851806,
0.9409568734283333
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.8579234972677595,
"hi": 0.9508196721311475
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 181,
"contingency": {
"a_both_pass": 1,
"b_x_pass_y_fail": 166,
"c_x_fail_y_pass": 1,
"d_neither": 13
},
"x_rate_kn": "167/181 (92.3%)",
"y_rate_kn": "2/181 (1.1%)",
"diff_x_minus_y": 0.9116022099447514,
"diff_pct_points": 91.16022099447514,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 166,
"c": 1,
"n_discordant": 167,
"p_value": 1.7960978851819555e-48
},
"ci95_diff_newcombe": [
0.851668647267405,
0.944618775580622
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.8674033149171271,
"hi": 0.9502762430939227
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
},
"tuned220k_q8__vs__tuned83k_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 183,
"contingency": {
"a_both_pass": 162,
"b_x_pass_y_fail": 6,
"c_x_fail_y_pass": 12,
"d_neither": 3
},
"x_rate_kn": "168/183 (91.8%)",
"y_rate_kn": "174/183 (95.1%)",
"diff_x_minus_y": -0.03278688524590164,
"diff_pct_points": -3.278688524590164,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 6,
"c": 12,
"n_discordant": 18,
"p_value": 0.237884521484375
},
"ci95_diff_newcombe": [
-0.08228406453623008,
0.013972693215261692
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.07650273224043716,
"hi": 0.01092896174863388
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 183,
"contingency": {
"a_both_pass": 162,
"b_x_pass_y_fail": 6,
"c_x_fail_y_pass": 12,
"d_neither": 3
},
"x_rate_kn": "168/183 (91.8%)",
"y_rate_kn": "174/183 (95.1%)",
"diff_x_minus_y": -0.03278688524590164,
"diff_pct_points": -3.278688524590164,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 6,
"c": 12,
"n_discordant": 18,
"p_value": 0.237884521484375
},
"ci95_diff_newcombe": [
-0.08228406453623008,
0.013972693215261692
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.07650273224043716,
"hi": 0.01092896174863388
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
},
"tuned220k_q8__vs__frontier_gpt56sol": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "C frontier gpt-5.6-sol (vendor-hosted)",
"n_pairs": 183,
"contingency": {
"a_both_pass": 166,
"b_x_pass_y_fail": 2,
"c_x_fail_y_pass": 12,
"d_neither": 3
},
"x_rate_kn": "168/183 (91.8%)",
"y_rate_kn": "178/183 (97.3%)",
"diff_x_minus_y": -0.0546448087431694,
"diff_pct_points": -5.46448087431694,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 2,
"c": 12,
"n_discordant": 14,
"p_value": 0.012939453125
},
"ci95_diff_newcombe": [
-0.10098946523101315,
-0.015557162942926687
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.09836065573770492,
"hi": -0.01639344262295082
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "C frontier gpt-5.6-sol (vendor-hosted)",
"n_pairs": 183,
"contingency": {
"a_both_pass": 166,
"b_x_pass_y_fail": 2,
"c_x_fail_y_pass": 12,
"d_neither": 3
},
"x_rate_kn": "168/183 (91.8%)",
"y_rate_kn": "178/183 (97.3%)",
"diff_x_minus_y": -0.0546448087431694,
"diff_pct_points": -5.46448087431694,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 2,
"c": 12,
"n_discordant": 14,
"p_value": 0.012939453125
},
"ci95_diff_newcombe": [
-0.10098946523101315,
-0.015557162942926687
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.09836065573770492,
"hi": -0.01639344262295082
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
},
"tuned83k_q8__vs__frontier_gpt56sol": {
"inclusive": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "C frontier gpt-5.6-sol (vendor-hosted)",
"n_pairs": 183,
"contingency": {
"a_both_pass": 172,
"b_x_pass_y_fail": 2,
"c_x_fail_y_pass": 6,
"d_neither": 3
},
"x_rate_kn": "174/183 (95.1%)",
"y_rate_kn": "178/183 (97.3%)",
"diff_x_minus_y": -0.02185792349726776,
"diff_pct_points": -2.185792349726776,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 2,
"c": 6,
"n_discordant": 8,
"p_value": 0.2890625
},
"ci95_diff_newcombe": [
-0.05956767726362802,
0.010850721392073048
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.0546448087431694,
"hi": 0.00546448087431694
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "C frontier gpt-5.6-sol (vendor-hosted)",
"n_pairs": 183,
"contingency": {
"a_both_pass": 172,
"b_x_pass_y_fail": 2,
"c_x_fail_y_pass": 6,
"d_neither": 3
},
"x_rate_kn": "174/183 (95.1%)",
"y_rate_kn": "178/183 (97.3%)",
"diff_x_minus_y": -0.02185792349726776,
"diff_pct_points": -2.185792349726776,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 2,
"c": 6,
"n_discordant": 8,
"p_value": 0.2890625
},
"ci95_diff_newcombe": [
-0.05956767726362802,
0.010850721392073048
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.0546448087431694,
"hi": 0.00546448087431694
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
}
}
},
"harness_bucket_counts_per_arm": {
"base_q8": {
"compile-fail": 180,
"TRUNCATED": 2,
"compile-pass": 2
},
"tuned83k_q8": {
"compile-pass": 175,
"compile-fail": 9
},
"tuned220k_q8": {
"compile-pass": 169,
"compile-fail": 15
},
"frontier_gpt56sol": {
"compile-pass": 179,
"compile-fail": 5
}
},
"exclusion_counts_by_class_per_arm": {
"scored": {
"base_q8": 181,
"tuned83k_q8": 183,
"tuned220k_q8": 183,
"frontier_gpt56sol": 183
},
"contaminated": {
"base_q8": 1,
"tuned83k_q8": 1,
"tuned220k_q8": 1,
"frontier_gpt56sol": 1
},
"truncated": {
"base_q8": 2
}
},
"exclusion_buckets_by_artifact_class": null,
"denominator_pools": {
"full": 184,
"clean": 183
},
"item_alignment": {
"join_key": "ea_name",
"per_arm_unique_names": {
"base_q8": 184,
"tuned83k_q8": 184,
"tuned220k_q8": 184,
"frontier_gpt56sol": 184
},
"intersection_n": 184,
"per_arm_names_missing_from_intersection": {},
"duplicate_ea_names_per_arm": {},
"ALIGNED": true,
"prompt_identity": {
"items_with_identical_prompt_sha256_across_all_arms": 0,
"items_disagreeing_with_the_majority_prompt_sha256_per_arm": {
"base_q8": 0,
"tuned83k_q8": 0,
"tuned220k_q8": 0,
"frontier_gpt56sol": 184
},
"ALL_ARMS_BYTE_IDENTICAL_PROMPTS": false,
"note": "a differing prompt_sha256 means the arm was served a differently assembled prompt string; the head-to-head with that arm is confounded and is reported as such"
},
"spec_identity": {
"items_with_identical_spec_sha256_across_all_arms": 184,
"items_disagreeing_with_the_majority_spec_sha256_per_arm": {
"base_q8": 0,
"tuned83k_q8": 0,
"tuned220k_q8": 0,
"frontier_gpt56sol": 0
},
"rows_missing_spec_sha256_per_arm": {
"base_q8": 0,
"tuned83k_q8": 0,
"tuned220k_q8": 0,
"frontier_gpt56sol": 0
},
"ALL_ARMS_IDENTICAL_SPEC": true,
"note": "spec_sha256 identity is the pairing proof (the same item was put to every arm); prompt_sha256 identity is the separate byte-identity requirement of the pre-registered protocol (see the top-level `protocol` field) and is reported beside it"
},
"manifest_selection_sha256_identical_across_arms": true
},
"instrument_validation": {
"base_q8": {
"known_positive": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 4,
"row_errors_field": 4,
"PASS": true
},
"negative_control": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 0,
"result_line": "Result: 0 errors, 0 warnings",
"PASS": true
},
"utf16_trap": {
"raw_is_utf16_le_bom": true,
"byte_grep_ascii_hits_in_raw": 0,
"decoded_hits": 4,
"meaning": "a GNU-grep-style byte match on the .raw.log returns 0 while the decoded log yields the true count; the taxonomy is built on the decoded text"
}
},
"tuned83k_q8": {
"known_positive": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 1,
"row_errors_field": 1,
"PASS": true
},
"negative_control": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 0,
"result_line": "Result: 0 errors, 0 warnings",
"PASS": true
},
"utf16_trap": {
"raw_is_utf16_le_bom": true,
"byte_grep_ascii_hits_in_raw": 0,
"decoded_hits": 1,
"meaning": "a GNU-grep-style byte match on the .raw.log returns 0 while the decoded log yields the true count; the taxonomy is built on the decoded text"
}
},
"tuned220k_q8": {
"known_positive": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 7,
"row_errors_field": 7,
"PASS": true
},
"negative_control": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 0,
"result_line": "Result: 0 errors, 0 warnings",
"PASS": true
},
"utf16_trap": {
"raw_is_utf16_le_bom": true,
"byte_grep_ascii_hits_in_raw": 0,
"decoded_hits": 7,
"meaning": "a GNU-grep-style byte match on the .raw.log returns 0 while the decoded log yields the true count; the taxonomy is built on the decoded text"
}
},
"frontier_gpt56sol": {
"known_positive": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 1,
"row_errors_field": 1,
"PASS": true
},
"negative_control": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 0,
"result_line": "Result: 0 errors, 0 warnings",
"PASS": true
},
"utf16_trap": {
"raw_is_utf16_le_bom": true,
"byte_grep_ascii_hits_in_raw": 0,
"decoded_hits": 1,
"meaning": "a GNU-grep-style byte match on the .raw.log returns 0 while the decoded log yields the true count; the taxonomy is built on the decoded text"
}
}
},
"artifact_class_counts": null,
"category_field": "The `category` key is present on every row of this set and its value is the empty string on 100% of rows, so no artifact-class breakdown is computable from the rows. It is computable from the published files: prompts.jsonl carries each item's own specification, and six of the 184 declare an artifact type there on an explicit `// TYPE:` line. They are listed under the top-level public_set_184_composition field.",
"category_key_present_on_all_rows": true,
"inclusive_scoring_anomalies": {
"base_q8": {
"n_anomalous_rows": 0,
"rows": [],
"n_denominator_inclusive": 184,
"inclusive_k_under_the_PREREGISTERED_rule": 2,
"inclusive_k_if_the_row_verdict_were_honoured": 2,
"rule_applied": "an item in an excluded class is scored as a failure under the inclusive denominator (the wording of the pre-registered protocol; see the top-level `protocol` field)"
},
"tuned83k_q8": {
"n_anomalous_rows": 0,
"rows": [],
"n_denominator_inclusive": 184,
"inclusive_k_under_the_PREREGISTERED_rule": 175,
"inclusive_k_if_the_row_verdict_were_honoured": 175,
"rule_applied": "an item in an excluded class is scored as a failure under the inclusive denominator (the wording of the pre-registered protocol; see the top-level `protocol` field)"
},
"tuned220k_q8": {
"n_anomalous_rows": 0,
"rows": [],
"n_denominator_inclusive": 184,
"inclusive_k_under_the_PREREGISTERED_rule": 169,
"inclusive_k_if_the_row_verdict_were_honoured": 169,
"rule_applied": "an item in an excluded class is scored as a failure under the inclusive denominator (the wording of the pre-registered protocol; see the top-level `protocol` field)"
},
"frontier_gpt56sol": {
"n_anomalous_rows": 0,
"rows": [],
"n_denominator_inclusive": 184,
"inclusive_k_under_the_PREREGISTERED_rule": 179,
"inclusive_k_if_the_row_verdict_were_honoured": 179,
"rule_applied": "an item in an excluded class is scored as a failure under the inclusive denominator (the wording of the pre-registered protocol; see the top-level `protocol` field)"
}
}
},
"holdout_300": {
"title": "Part 2 \u2014 300-item EA holdout (clean-648 prefix-300, frozen order)",
"manifest_file_sha256": "75021a9d8db3fc40931de14239e7e279d37967c109dc0de12c4637473022f64a",
"per_arm": {
"full": {
"base_q8": {
"inclusive": {
"k": 1,
"n": 300,
"rate_pct": 0.3333333333333333,
"kn": "1/300 (0.3%)",
"ci95": [
0.0005886577218511581,
0.018636693694531237
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 1,
"n": 300,
"kn": "1/300 (0.3%)",
"ci95": [
0.0005886577218511581,
0.018636693694531237
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 1,
"n": 297,
"rate_pct": 0.3367003367003367,
"kn": "1/297 (0.3%)",
"ci95": [
0.0005946062361728496,
0.018822460957932597
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 1,
"n": 297,
"kn": "1/297 (0.3%)",
"ci95": [
0.0005946062361728496,
0.018822460957932597
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned83k_q8": {
"inclusive": {
"k": 281,
"n": 300,
"rate_pct": 93.66666666666667,
"kn": "281/300 (93.7%)",
"ci95": [
0.9032086875144784,
0.9590831176160847
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 281,
"n": 300,
"kn": "281/300 (93.7%)",
"ci95": [
0.9032086875144784,
0.9590831176160847
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 281,
"n": 300,
"rate_pct": 93.66666666666667,
"kn": "281/300 (93.7%)",
"ci95": [
0.9032086875144784,
0.9590831176160847
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 281,
"n": 300,
"kn": "281/300 (93.7%)",
"ci95": [
0.9032086875144784,
0.9590831176160847
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned220k_q8": {
"inclusive": {
"k": 282,
"n": 300,
"rate_pct": 94.0,
"kn": "282/300 (94.0%)",
"ci95": [
0.9071605548470915,
0.9617136304753079
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 281,
"n": 300,
"kn": "281/300 (93.7%)",
"ci95": [
0.9032086875144784,
0.9590831176160847
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 282,
"n": 300,
"rate_pct": 94.0,
"kn": "282/300 (94.0%)",
"ci95": [
0.9071605548470915,
0.9617136304753079
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 281,
"n": 300,
"kn": "281/300 (93.7%)",
"ci95": [
0.9032086875144784,
0.9590831176160847
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"frontier_gpt56sol": {
"inclusive": {
"k": 286,
"n": 300,
"rate_pct": 95.33333333333333,
"kn": "286/300 (95.3%)",
"ci95": [
0.923202639772079,
0.9720010663176659
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 282,
"n": 300,
"kn": "282/300 (94.0%)",
"ci95": [
0.9071605548470915,
0.9617136304753079
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 286,
"n": 293,
"rate_pct": 97.61092150170649,
"kn": "286/293 (97.6%)",
"ci95": [
0.9515155343949656,
0.9883801294668967
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 282,
"n": 293,
"kn": "282/293 (96.2%)",
"ci95": [
0.9340355857729857,
0.9789096647881782
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
}
}
},
"per_class": null,
"paired": {
"full": {
"tuned83k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 300,
"contingency": {
"a_both_pass": 1,
"b_x_pass_y_fail": 280,
"c_x_fail_y_pass": 0,
"d_neither": 19
},
"x_rate_kn": "281/300 (93.7%)",
"y_rate_kn": "1/300 (0.3%)",
"diff_x_minus_y": 0.9333333333333333,
"diff_pct_points": 93.33333333333333,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 280,
"c": 0,
"n_discordant": 280,
"p_value": 1.0295115178936058e-84
},
"ci95_diff_newcombe": [
0.8967515102434487,
0.9558761830532138
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.9033333333333333,
"hi": 0.96
},
"dominance": "X strictly dominates Y (c=0)"
},
"strict": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 297,
"contingency": {
"a_both_pass": 1,
"b_x_pass_y_fail": 278,
"c_x_fail_y_pass": 0,
"d_neither": 18
},
"x_rate_kn": "279/297 (93.9%)",
"y_rate_kn": "1/297 (0.3%)",
"diff_x_minus_y": 0.936026936026936,
"diff_pct_points": 93.6026936026936,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 278,
"c": 0,
"n_discordant": 278,
"p_value": 4.118046071574423e-84
},
"ci95_diff_newcombe": [
0.8996577628152559,
0.9580903377627212
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.9057239057239057,
"hi": 0.9629629629629629
},
"dominance": "X strictly dominates Y (c=0)"
}
},
"tuned220k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 300,
"contingency": {
"a_both_pass": 1,
"b_x_pass_y_fail": 281,
"c_x_fail_y_pass": 0,
"d_neither": 18
},
"x_rate_kn": "282/300 (94.0%)",
"y_rate_kn": "1/300 (0.3%)",
"diff_x_minus_y": 0.9366666666666666,
"diff_pct_points": 93.66666666666667,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 281,
"c": 0,
"n_discordant": 281,
"p_value": 5.147557589468029e-85
},
"ci95_diff_newcombe": [
0.9006397809140065,
0.9585132560623
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.9066666666666666,
"hi": 0.9633333333333334
},
"dominance": "X strictly dominates Y (c=0)"
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 297,
"contingency": {
"a_both_pass": 1,
"b_x_pass_y_fail": 279,
"c_x_fail_y_pass": 0,
"d_neither": 17
},
"x_rate_kn": "280/297 (94.3%)",
"y_rate_kn": "1/297 (0.3%)",
"diff_x_minus_y": 0.9393939393939394,
"diff_pct_points": 93.93939393939394,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 279,
"c": 0,
"n_discordant": 279,
"p_value": 2.0590230357872116e-84
},
"ci95_diff_newcombe": [
0.9036017469011658,
0.9607331722947036
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.9090909090909091,
"hi": 0.9663299663299664
},
"dominance": "X strictly dominates Y (c=0)"
}
},
"tuned220k_q8__vs__tuned83k_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 300,
"contingency": {
"a_both_pass": 264,
"b_x_pass_y_fail": 18,
"c_x_fail_y_pass": 17,
"d_neither": 1
},
"x_rate_kn": "282/300 (94.0%)",
"y_rate_kn": "281/300 (93.7%)",
"diff_x_minus_y": 0.0033333333333333335,
"diff_pct_points": 0.3333333333333333,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 18,
"c": 17,
"n_discordant": 35,
"p_value": 1.0
},
"ci95_diff_newcombe": [
-0.03657662339661758,
0.04336632758652774
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.03333333333333333,
"hi": 0.043333333333333335
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 300,
"contingency": {
"a_both_pass": 264,
"b_x_pass_y_fail": 18,
"c_x_fail_y_pass": 17,
"d_neither": 1
},
"x_rate_kn": "282/300 (94.0%)",
"y_rate_kn": "281/300 (93.7%)",
"diff_x_minus_y": 0.0033333333333333335,
"diff_pct_points": 0.3333333333333333,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 18,
"c": 17,
"n_discordant": 35,
"p_value": 1.0
},
"ci95_diff_newcombe": [
-0.03657662339661758,
0.04336632758652774
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.03333333333333333,
"hi": 0.043333333333333335
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
},
"tuned220k_q8__vs__frontier_gpt56sol": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "C frontier gpt-5.6-sol (vendor-hosted)",
"n_pairs": 300,
"contingency": {
"a_both_pass": 270,
"b_x_pass_y_fail": 12,
"c_x_fail_y_pass": 16,
"d_neither": 2
},
"x_rate_kn": "282/300 (94.0%)",
"y_rate_kn": "286/300 (95.3%)",
"diff_x_minus_y": -0.013333333333333334,
"diff_pct_points": -1.3333333333333333,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 12,
"c": 16,
"n_discordant": 28,
"p_value": 0.5715881884098053
},
"ci95_diff_newcombe": [
-0.04983361208865994,
0.02242048353135713
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.05,
"hi": 0.02
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "C frontier gpt-5.6-sol (vendor-hosted)",
"n_pairs": 293,
"contingency": {
"a_both_pass": 270,
"b_x_pass_y_fail": 6,
"c_x_fail_y_pass": 16,
"d_neither": 1
},
"x_rate_kn": "276/293 (94.2%)",
"y_rate_kn": "286/293 (97.6%)",
"diff_x_minus_y": -0.034129692832764506,
"diff_pct_points": -3.4129692832764507,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 6,
"c": 16,
"n_discordant": 22,
"p_value": 0.052478790283203125
},
"ci95_diff_newcombe": [
-0.06860584179765444,
-0.0024060942614670816
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.06484641638225255,
"hi": -0.0034129692832764505
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
},
"tuned83k_q8__vs__frontier_gpt56sol": {
"inclusive": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "C frontier gpt-5.6-sol (vendor-hosted)",
"n_pairs": 300,
"contingency": {
"a_both_pass": 268,
"b_x_pass_y_fail": 13,
"c_x_fail_y_pass": 18,
"d_neither": 1
},
"x_rate_kn": "281/300 (93.7%)",
"y_rate_kn": "286/300 (95.3%)",
"diff_x_minus_y": -0.016666666666666666,
"diff_pct_points": -1.6666666666666667,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 13,
"c": 18,
"n_discordant": 31,
"p_value": 0.47312965989112854
},
"ci95_diff_newcombe": [
-0.05486005211734017,
0.02075555052375612
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.05333333333333334,
"hi": 0.02
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "C frontier gpt-5.6-sol (vendor-hosted)",
"n_pairs": 293,
"contingency": {
"a_both_pass": 268,
"b_x_pass_y_fail": 7,
"c_x_fail_y_pass": 18,
"d_neither": 0
},
"x_rate_kn": "275/293 (93.9%)",
"y_rate_kn": "286/293 (97.6%)",
"diff_x_minus_y": -0.03754266211604096,
"diff_pct_points": -3.7542662116040955,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 7,
"c": 18,
"n_discordant": 25,
"p_value": 0.043285250663757324
},
"ci95_diff_newcombe": [
-0.07374828048013808,
-0.00374159703780242
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.07167235494880546,
"hi": -0.006825938566552901
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
}
}
},
"harness_bucket_counts_per_arm": {
"base_q8": {
"compile-fail": 296,
"TRUNCATED": 3,
"compile-pass": 1
},
"tuned83k_q8": {
"compile-pass": 281,
"compile-fail": 19
},
"tuned220k_q8": {
"compile-pass": 282,
"compile-fail": 18
},
"frontier_gpt56sol": {
"compile-pass": 286,
"compile-fail": 7,
"TRUNCATED": 4,
"INFRASTRUCTURE-FAILURE": 3
}
},
"exclusion_counts_by_class_per_arm": {
"scored": {
"base_q8": 297,
"tuned83k_q8": 300,
"tuned220k_q8": 300,
"frontier_gpt56sol": 293
},
"truncated": {
"base_q8": 3,
"frontier_gpt56sol": 4
},
"generation_error": {
"frontier_gpt56sol": 3
}
},
"exclusion_buckets_by_artifact_class": null,
"denominator_pools": {
"full": 300
},
"item_alignment": {
"join_key": "ea_name",
"per_arm_unique_names": {
"base_q8": 300,
"tuned83k_q8": 300,
"tuned220k_q8": 300,
"frontier_gpt56sol": 300
},
"intersection_n": 300,
"per_arm_names_missing_from_intersection": {},
"duplicate_ea_names_per_arm": {},
"ALIGNED": true,
"prompt_identity": {
"items_with_identical_prompt_sha256_across_all_arms": 0,
"items_disagreeing_with_the_majority_prompt_sha256_per_arm": {
"base_q8": 0,
"tuned83k_q8": 0,
"tuned220k_q8": 0,
"frontier_gpt56sol": 300
},
"ALL_ARMS_BYTE_IDENTICAL_PROMPTS": false,
"note": "a differing prompt_sha256 means the arm was served a differently assembled prompt string; the head-to-head with that arm is confounded and is reported as such"
},
"spec_identity": {
"items_with_identical_spec_sha256_across_all_arms": 300,
"items_disagreeing_with_the_majority_spec_sha256_per_arm": {
"base_q8": 0,
"tuned83k_q8": 0,
"tuned220k_q8": 0,
"frontier_gpt56sol": 0
},
"rows_missing_spec_sha256_per_arm": {
"base_q8": 0,
"tuned83k_q8": 0,
"tuned220k_q8": 0,
"frontier_gpt56sol": 0
},
"ALL_ARMS_IDENTICAL_SPEC": true,
"note": "spec_sha256 identity is the pairing proof (the same item was put to every arm); prompt_sha256 identity is the separate byte-identity requirement of the pre-registered protocol (see the top-level `protocol` field) and is reported beside it"
},
"manifest_selection_sha256_identical_across_arms": true
},
"instrument_validation": {
"base_q8": {
"known_positive": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 11,
"row_errors_field": 11,
"PASS": true
},
"negative_control": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 0,
"result_line": "Result: 0 errors, 0 warnings",
"PASS": true
},
"utf16_trap": {
"raw_is_utf16_le_bom": true,
"byte_grep_ascii_hits_in_raw": 0,
"decoded_hits": 11,
"meaning": "a GNU-grep-style byte match on the .raw.log returns 0 while the decoded log yields the true count; the taxonomy is built on the decoded text"
}
},
"tuned83k_q8": {
"known_positive": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 7,
"row_errors_field": 7,
"PASS": true
},
"negative_control": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 0,
"result_line": "Result: 0 errors, 0 warnings",
"PASS": true
},
"utf16_trap": {
"raw_is_utf16_le_bom": true,
"byte_grep_ascii_hits_in_raw": 0,
"decoded_hits": 7,
"meaning": "a GNU-grep-style byte match on the .raw.log returns 0 while the decoded log yields the true count; the taxonomy is built on the decoded text"
}
},
"tuned220k_q8": {
"known_positive": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 2,
"row_errors_field": 2,
"PASS": true
},
"negative_control": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 0,
"result_line": "Result: 0 errors, 0 warnings",
"PASS": true
},
"utf16_trap": {
"raw_is_utf16_le_bom": true,
"byte_grep_ascii_hits_in_raw": 0,
"decoded_hits": 2,
"meaning": "a GNU-grep-style byte match on the .raw.log returns 0 while the decoded log yields the true count; the taxonomy is built on the decoded text"
}
},
"frontier_gpt56sol": {
"known_positive": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 10,
"row_errors_field": 10,
"PASS": true
},
"negative_control": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 0,
"result_line": "Result: 0 errors, 0 warnings",
"PASS": true
},
"utf16_trap": {
"raw_is_utf16_le_bom": true,
"byte_grep_ascii_hits_in_raw": 0,
"decoded_hits": 10,
"meaning": "a GNU-grep-style byte match on the .raw.log returns 0 while the decoded log yields the true count; the taxonomy is built on the decoded text"
}
}
},
"artifact_class_counts": null,
"category_field": "The `category` key is present on every row of this set and its value is the empty string on 100% of rows, so no artifact-class breakdown is computable here and none was invented. The key exists; the value is empty.",
"category_key_present_on_all_rows": true,
"contaminated_items_excluded_in_clean_pool": [],
"inclusive_scoring_anomalies": {
"base_q8": {
"n_anomalous_rows": 0,
"rows": [],
"n_denominator_inclusive": 300,
"inclusive_k_under_the_PREREGISTERED_rule": 1,
"inclusive_k_if_the_row_verdict_were_honoured": 1,
"rule_applied": "an item in an excluded class is scored as a failure under the inclusive denominator (the wording of the pre-registered protocol; see the top-level `protocol` field)"
},
"tuned83k_q8": {
"n_anomalous_rows": 0,
"rows": [],
"n_denominator_inclusive": 300,
"inclusive_k_under_the_PREREGISTERED_rule": 281,
"inclusive_k_if_the_row_verdict_were_honoured": 281,
"rule_applied": "an item in an excluded class is scored as a failure under the inclusive denominator (the wording of the pre-registered protocol; see the top-level `protocol` field)"
},
"tuned220k_q8": {
"n_anomalous_rows": 0,
"rows": [],
"n_denominator_inclusive": 300,
"inclusive_k_under_the_PREREGISTERED_rule": 282,
"inclusive_k_if_the_row_verdict_were_honoured": 282,
"rule_applied": "an item in an excluded class is scored as a failure under the inclusive denominator (the wording of the pre-registered protocol; see the top-level `protocol` field)"
},
"frontier_gpt56sol": {
"n_anomalous_rows": 0,
"rows": [],
"n_denominator_inclusive": 300,
"inclusive_k_under_the_PREREGISTERED_rule": 286,
"inclusive_k_if_the_row_verdict_were_honoured": 286,
"rule_applied": "an item in an excluded class is scored as a failure under the inclusive denominator (the wording of the pre-registered protocol; see the top-level `protocol` field)"
}
}
},
"nonea_c200": {
"title": "Part 3 \u2014 balanced 200-item non-EA holdout, 50 per class (include / indicator / script / service)",
"manifest_file_sha256": "db0e0614eed18143e41931353fd47eb4e86422f673b0b44e23ec2eebff090065",
"per_arm": {
"full": {
"base_q8": {
"inclusive": {
"k": 56,
"n": 200,
"rate_pct": 28.0,
"kn": "56/200 (28.0%)",
"ci95": [
0.22236889773949026,
0.3459230458665071
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 48,
"n": 200,
"kn": "48/200 (24.0%)",
"ci95": [
0.1860661599241369,
0.30373340979204166
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 56,
"n": 186,
"rate_pct": 30.107526881720432,
"kn": "56/186 (30.1%)",
"ci95": [
0.23972277016842466,
0.3704782867572216
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 48,
"n": 186,
"kn": "48/186 (25.8%)",
"ci95": [
0.20052355007764888,
0.3253966542913804
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned83k_q8": {
"inclusive": {
"k": 135,
"n": 200,
"rate_pct": 67.5,
"kn": "135/200 (67.5%)",
"ci95": [
0.6073198511602996,
0.7360842846076571
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 126,
"n": 200,
"kn": "126/200 (63.0%)",
"ci95": [
0.5612262992405018,
0.6938739159014088
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 135,
"n": 187,
"rate_pct": 72.19251336898395,
"kn": "135/187 (72.2%)",
"ci95": [
0.653733203391701,
0.7811827761491125
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 126,
"n": 187,
"kn": "126/187 (67.4%)",
"ci95": [
0.6036913813439719,
0.736905470103653
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned220k_q8": {
"inclusive": {
"k": 168,
"n": 200,
"rate_pct": 84.0,
"kn": "168/200 (84.0%)",
"ci95": [
0.7828592983793289,
0.8843258796841297
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 168,
"n": 200,
"kn": "168/200 (84.0%)",
"ci95": [
0.7828592983793289,
0.8843258796841297
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 168,
"n": 188,
"rate_pct": 89.36170212765957,
"kn": "168/188 (89.4%)",
"ci95": [
0.8413986236498567,
0.9300717400025025
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 168,
"n": 188,
"kn": "168/188 (89.4%)",
"ci95": [
0.8413986236498567,
0.9300717400025025
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
}
}
},
"per_class": {
"full": {
"include": {
"INDICATIVE": true,
"cell_note": "Per-class cell, n=50, indicative. Cell counts are printed everywhere so nobody over-reads a k-of-n; a per-class CI is given but it settles nothing on its own.",
"per_arm": {
"base_q8": {
"inclusive": {
"k": 20,
"n": 50,
"rate_pct": 40.0,
"kn": "20/50 (40.0%)",
"ci95": [
0.27608389730256544,
0.5381856225241064
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 16,
"n": 50,
"kn": "16/50 (32.0%)",
"ci95": [
0.20758216260656845,
0.45810297308144077
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 20,
"n": 43,
"rate_pct": 46.51162790697674,
"kn": "20/43 (46.5%)",
"ci95": [
0.32511057704121815,
0.6108435952768602
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 16,
"n": 43,
"kn": "16/43 (37.2%)",
"ci95": [
0.24376339009650572,
0.5214019084031153
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned83k_q8": {
"inclusive": {
"k": 38,
"n": 50,
"rate_pct": 76.0,
"kn": "38/50 (76.0%)",
"ci95": [
0.6258731624205708,
0.8570260860300828
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 38,
"n": 50,
"kn": "38/50 (76.0%)",
"ci95": [
0.6258731624205708,
0.8570260860300828
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 38,
"n": 42,
"rate_pct": 90.47619047619048,
"kn": "38/42 (90.5%)",
"ci95": [
0.7793489419142897,
0.9623377511944261
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 38,
"n": 42,
"kn": "38/42 (90.5%)",
"ci95": [
0.7793489419142897,
0.9623377511944261
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned220k_q8": {
"inclusive": {
"k": 31,
"n": 50,
"rate_pct": 62.0,
"kn": "31/50 (62.0%)",
"ci95": [
0.4815044693099298,
0.7413721068980641
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 31,
"n": 50,
"kn": "31/50 (62.0%)",
"ci95": [
0.4815044693099298,
0.7413721068980641
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 31,
"n": 39,
"rate_pct": 79.48717948717949,
"kn": "31/39 (79.5%)",
"ci95": [
0.644660319864177,
0.8922028176928961
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 31,
"n": 39,
"kn": "31/39 (79.5%)",
"ci95": [
0.644660319864177,
0.8922028176928961
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
}
},
"paired": {
"tuned83k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 19,
"b_x_pass_y_fail": 19,
"c_x_fail_y_pass": 1,
"d_neither": 11
},
"x_rate_kn": "38/50 (76.0%)",
"y_rate_kn": "20/50 (40.0%)",
"diff_x_minus_y": 0.36,
"diff_pct_points": 36.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 19,
"c": 1,
"n_discordant": 20,
"p_value": 4.00543212890625e-05
},
"ci95_diff_newcombe": [
0.20631088196171574,
0.4866281416931669
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.22,
"hi": 0.5
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions",
"n_discordant": 20,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
},
"strict": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 41,
"contingency": {
"a_both_pass": 19,
"b_x_pass_y_fail": 18,
"c_x_fail_y_pass": 1,
"d_neither": 3
},
"x_rate_kn": "37/41 (90.2%)",
"y_rate_kn": "20/41 (48.8%)",
"diff_x_minus_y": 0.4146341463414634,
"diff_pct_points": 41.46341463414634,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 18,
"c": 1,
"n_discordant": 19,
"p_value": 7.62939453125e-05
},
"ci95_diff_newcombe": [
0.2352526212275229,
0.5626130585283183
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.24390243902439024,
"hi": 0.5853658536585366
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions",
"n_discordant": 19,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
}
},
"tuned220k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 20,
"b_x_pass_y_fail": 11,
"c_x_fail_y_pass": 0,
"d_neither": 19
},
"x_rate_kn": "31/50 (62.0%)",
"y_rate_kn": "20/50 (40.0%)",
"diff_x_minus_y": 0.22,
"diff_pct_points": 22.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 11,
"c": 0,
"n_discordant": 11,
"p_value": 0.0009765625
},
"ci95_diff_newcombe": [
0.10248668667863278,
0.3242051136202031
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.12,
"hi": 0.34
},
"dominance": "X strictly dominates Y (c=0)",
"n_discordant": 11,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 38,
"contingency": {
"a_both_pass": 20,
"b_x_pass_y_fail": 10,
"c_x_fail_y_pass": 0,
"d_neither": 8
},
"x_rate_kn": "30/38 (78.9%)",
"y_rate_kn": "20/38 (52.6%)",
"diff_x_minus_y": 0.2631578947368421,
"diff_pct_points": 26.31578947368421,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 10,
"c": 0,
"n_discordant": 10,
"p_value": 0.001953125
},
"ci95_diff_newcombe": [
0.11904686012522916,
0.39311537251842865
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.13157894736842105,
"hi": 0.42105263157894735
},
"dominance": "X strictly dominates Y (c=0)",
"n_discordant": 10,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
}
},
"tuned220k_q8__vs__tuned83k_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 26,
"b_x_pass_y_fail": 5,
"c_x_fail_y_pass": 12,
"d_neither": 7
},
"x_rate_kn": "31/50 (62.0%)",
"y_rate_kn": "38/50 (76.0%)",
"diff_x_minus_y": -0.14,
"diff_pct_points": -14.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 5,
"c": 12,
"n_discordant": 17,
"p_value": 0.143463134765625
},
"ci95_diff_newcombe": [
-0.2892260707501137,
0.018293044636248756
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.3,
"hi": 0.02
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions",
"n_discordant": 17,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 38,
"contingency": {
"a_both_pass": 26,
"b_x_pass_y_fail": 4,
"c_x_fail_y_pass": 8,
"d_neither": 0
},
"x_rate_kn": "30/38 (78.9%)",
"y_rate_kn": "34/38 (89.5%)",
"diff_x_minus_y": -0.10526315789473684,
"diff_pct_points": -10.526315789473685,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 4,
"c": 8,
"n_discordant": 12,
"p_value": 0.3876953125
},
"ci95_diff_newcombe": [
-0.28096575476226193,
0.07714223086316904
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.2894736842105263,
"hi": 0.07894736842105263
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions",
"n_discordant": 12,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
}
}
},
"exclusion_classes_within_this_class": {
"base_q8": {
"scored": 43,
"truncated": 7
},
"tuned83k_q8": {
"scored": 42,
"truncated": 8
},
"tuned220k_q8": {
"scored": 39,
"truncated": 11
}
}
},
"indicator": {
"INDICATIVE": true,
"cell_note": "Per-class cell, n=50, indicative. Cell counts are printed everywhere so nobody over-reads a k-of-n; a per-class CI is given but it settles nothing on its own.",
"per_arm": {
"base_q8": {
"inclusive": {
"k": 0,
"n": 50,
"rate_pct": 0.0,
"kn": "0/50 (0.0%)",
"ci95": [
0.0,
0.07134759913335872
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 0,
"n": 50,
"kn": "0/50 (0.0%)",
"ci95": [
0.0,
0.07134759913335872
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 0,
"n": 45,
"rate_pct": 0.0,
"kn": "0/45 (0.0%)",
"ci95": [
0.0,
0.07865159873288836
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 0,
"n": 45,
"kn": "0/45 (0.0%)",
"ci95": [
0.0,
0.07865159873288836
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned83k_q8": {
"inclusive": {
"k": 49,
"n": 50,
"rate_pct": 98.0,
"kn": "49/50 (98.0%)",
"ci95": [
0.8950455641036219,
0.9964607407283539
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 48,
"n": 50,
"kn": "48/50 (96.0%)",
"ci95": [
0.8653990931249297,
0.9889611156723802
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 49,
"n": 50,
"rate_pct": 98.0,
"kn": "49/50 (98.0%)",
"ci95": [
0.8950455641036219,
0.9964607407283539
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 48,
"n": 50,
"kn": "48/50 (96.0%)",
"ci95": [
0.8653990931249297,
0.9889611156723802
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned220k_q8": {
"inclusive": {
"k": 50,
"n": 50,
"rate_pct": 100.0,
"kn": "50/50 (100.0%)",
"ci95": [
0.9286524008666414,
1.0
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 50,
"n": 50,
"kn": "50/50 (100.0%)",
"ci95": [
0.9286524008666414,
1.0
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 50,
"n": 50,
"rate_pct": 100.0,
"kn": "50/50 (100.0%)",
"ci95": [
0.9286524008666414,
1.0
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 50,
"n": 50,
"kn": "50/50 (100.0%)",
"ci95": [
0.9286524008666414,
1.0
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
}
},
"paired": {
"tuned83k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 0,
"b_x_pass_y_fail": 49,
"c_x_fail_y_pass": 0,
"d_neither": 1
},
"x_rate_kn": "49/50 (98.0%)",
"y_rate_kn": "0/50 (0.0%)",
"diff_x_minus_y": 0.98,
"diff_pct_points": 98.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 49,
"c": 0,
"n_discordant": 49,
"p_value": 3.552713678800501e-15
},
"ci95_diff_newcombe": [
0.8690597634734527,
0.9964607407283539
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.94,
"hi": 1.0
},
"dominance": "X strictly dominates Y (c=0)",
"n_discordant": 49,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
},
"strict": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 45,
"contingency": {
"a_both_pass": 0,
"b_x_pass_y_fail": 44,
"c_x_fail_y_pass": 0,
"d_neither": 1
},
"x_rate_kn": "44/45 (97.8%)",
"y_rate_kn": "0/45 (0.0%)",
"diff_x_minus_y": 0.9777777777777777,
"diff_pct_points": 97.77777777777777,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 44,
"c": 0,
"n_discordant": 44,
"p_value": 1.1368683772161603e-13
},
"ci95_diff_newcombe": [
0.855638650532319,
0.9960664011171667
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.9333333333333333,
"hi": 1.0
},
"dominance": "X strictly dominates Y (c=0)",
"n_discordant": 44,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
}
},
"tuned220k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 0,
"b_x_pass_y_fail": 50,
"c_x_fail_y_pass": 0,
"d_neither": 0
},
"x_rate_kn": "50/50 (100.0%)",
"y_rate_kn": "0/50 (0.0%)",
"diff_x_minus_y": 1.0,
"diff_pct_points": 100.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 50,
"c": 0,
"n_discordant": 50,
"p_value": 1.7763568394002505e-15
},
"ci95_diff_newcombe": [
0.8990992576628453,
1.0
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 1.0,
"hi": 1.0
},
"dominance": "X strictly dominates Y (c=0)",
"n_discordant": 50,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 45,
"contingency": {
"a_both_pass": 0,
"b_x_pass_y_fail": 45,
"c_x_fail_y_pass": 0,
"d_neither": 0
},
"x_rate_kn": "45/45 (100.0%)",
"y_rate_kn": "0/45 (0.0%)",
"diff_x_minus_y": 1.0,
"diff_pct_points": 100.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 45,
"c": 0,
"n_discordant": 45,
"p_value": 5.684341886080802e-14
},
"ci95_diff_newcombe": [
0.8887698423696228,
1.0
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 1.0,
"hi": 1.0
},
"dominance": "X strictly dominates Y (c=0)",
"n_discordant": 45,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
}
},
"tuned220k_q8__vs__tuned83k_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 49,
"b_x_pass_y_fail": 1,
"c_x_fail_y_pass": 0,
"d_neither": 0
},
"x_rate_kn": "50/50 (100.0%)",
"y_rate_kn": "49/50 (98.0%)",
"diff_x_minus_y": 0.02,
"diff_pct_points": 2.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 1,
"c": 0,
"n_discordant": 1,
"p_value": 1.0
},
"ci95_diff_newcombe": [
-0.053221826578012416,
0.1049544358963781
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.0,
"hi": 0.06
},
"dominance": "X strictly dominates Y (c=0)",
"n_discordant": 1,
"COMPUTABLE": true,
"DISCORDANCE_THIN": true
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 49,
"b_x_pass_y_fail": 1,
"c_x_fail_y_pass": 0,
"d_neither": 0
},
"x_rate_kn": "50/50 (100.0%)",
"y_rate_kn": "49/50 (98.0%)",
"diff_x_minus_y": 0.02,
"diff_pct_points": 2.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 1,
"c": 0,
"n_discordant": 1,
"p_value": 1.0
},
"ci95_diff_newcombe": [
-0.053221826578012416,
0.1049544358963781
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.0,
"hi": 0.06
},
"dominance": "X strictly dominates Y (c=0)",
"n_discordant": 1,
"COMPUTABLE": true,
"DISCORDANCE_THIN": true
}
}
},
"exclusion_classes_within_this_class": {
"base_q8": {
"scored": 45,
"truncated": 5
},
"tuned83k_q8": {
"scored": 50
},
"tuned220k_q8": {
"scored": 50
}
}
},
"script": {
"INDICATIVE": true,
"cell_note": "Per-class cell, n=50, indicative. Cell counts are printed everywhere so nobody over-reads a k-of-n; a per-class CI is given but it settles nothing on its own.",
"per_arm": {
"base_q8": {
"inclusive": {
"k": 13,
"n": 50,
"rate_pct": 26.0,
"kn": "13/50 (26.0%)",
"ci95": [
0.15871527493552393,
0.39553157264848837
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 12,
"n": 50,
"kn": "12/50 (24.0%)",
"ci95": [
0.1429739139699173,
0.3741268375794292
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 13,
"n": 48,
"rate_pct": 27.083333333333332,
"kn": "13/48 (27.1%)",
"ci95": [
0.1656595762745644,
0.40996964982240613
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 12,
"n": 48,
"kn": "12/48 (25.0%)",
"ci95": [
0.1492063367379791,
0.3878437280950796
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned83k_q8": {
"inclusive": {
"k": 19,
"n": 50,
"rate_pct": 38.0,
"kn": "19/50 (38.0%)",
"ci95": [
0.25862789310193596,
0.5184955306900703
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 17,
"n": 50,
"kn": "17/50 (34.0%)",
"ci95": [
0.2243694922250009,
0.478461739497674
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 19,
"n": 45,
"rate_pct": 42.22222222222222,
"kn": "19/45 (42.2%)",
"ci95": [
0.289687024672703,
0.5669921129079686
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 17,
"n": 45,
"kn": "17/45 (37.8%)",
"ci95": [
0.25108084672137915,
0.5237006551911048
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned220k_q8": {
"inclusive": {
"k": 39,
"n": 50,
"rate_pct": 78.0,
"kn": "39/50 (78.0%)",
"ci95": [
0.6475845041874634,
0.8724608402978559
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 39,
"n": 50,
"kn": "39/50 (78.0%)",
"ci95": [
0.6475845041874634,
0.8724608402978559
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 39,
"n": 49,
"rate_pct": 79.59183673469387,
"kn": "39/49 (79.6%)",
"ci95": [
0.6636299908048744,
0.8851815039421786
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 39,
"n": 49,
"kn": "39/49 (79.6%)",
"ci95": [
0.6636299908048744,
0.8851815039421786
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
}
},
"paired": {
"tuned83k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 5,
"b_x_pass_y_fail": 14,
"c_x_fail_y_pass": 8,
"d_neither": 23
},
"x_rate_kn": "19/50 (38.0%)",
"y_rate_kn": "13/50 (26.0%)",
"diff_x_minus_y": 0.12,
"diff_pct_points": 12.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 14,
"c": 8,
"n_discordant": 22,
"p_value": 0.28627872467041016
},
"ci95_diff_newcombe": [
-0.0614237176657442,
0.29111832884188804
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.06,
"hi": 0.3
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions",
"n_discordant": 22,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
},
"strict": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 44,
"contingency": {
"a_both_pass": 5,
"b_x_pass_y_fail": 14,
"c_x_fail_y_pass": 8,
"d_neither": 17
},
"x_rate_kn": "19/44 (43.2%)",
"y_rate_kn": "13/44 (29.5%)",
"diff_x_minus_y": 0.13636363636363635,
"diff_pct_points": 13.636363636363637,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 14,
"c": 8,
"n_discordant": 22,
"p_value": 0.28627872467041016
},
"ci95_diff_newcombe": [
-0.06906839603660886,
0.3269501414557799
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.06818181818181818,
"hi": 0.3409090909090909
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions",
"n_discordant": 22,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
}
},
"tuned220k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 11,
"b_x_pass_y_fail": 28,
"c_x_fail_y_pass": 2,
"d_neither": 9
},
"x_rate_kn": "39/50 (78.0%)",
"y_rate_kn": "13/50 (26.0%)",
"diff_x_minus_y": 0.52,
"diff_pct_points": 52.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 28,
"c": 2,
"n_discordant": 30,
"p_value": 8.67992639541626e-07
},
"ci95_diff_newcombe": [
0.33970846660140913,
0.6505168266317711
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.36,
"hi": 0.68
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions",
"n_discordant": 30,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 47,
"contingency": {
"a_both_pass": 11,
"b_x_pass_y_fail": 26,
"c_x_fail_y_pass": 2,
"d_neither": 8
},
"x_rate_kn": "37/47 (78.7%)",
"y_rate_kn": "13/47 (27.7%)",
"diff_x_minus_y": 0.5106382978723404,
"diff_pct_points": 51.06382978723404,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 26,
"c": 2,
"n_discordant": 28,
"p_value": 3.032386302947998e-06
},
"ci95_diff_newcombe": [
0.32348040174335807,
0.6461002126173133
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.3404255319148936,
"hi": 0.6808510638297872
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions",
"n_discordant": 28,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
}
},
"tuned220k_q8__vs__tuned83k_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 15,
"b_x_pass_y_fail": 24,
"c_x_fail_y_pass": 4,
"d_neither": 7
},
"x_rate_kn": "39/50 (78.0%)",
"y_rate_kn": "19/50 (38.0%)",
"diff_x_minus_y": 0.4,
"diff_pct_points": 40.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 24,
"c": 4,
"n_discordant": 28,
"p_value": 0.00017999112606048584
},
"ci95_diff_newcombe": [
0.2101100714093677,
0.5512558966727544
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.22,
"hi": 0.58
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions",
"n_discordant": 28,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 44,
"contingency": {
"a_both_pass": 15,
"b_x_pass_y_fail": 19,
"c_x_fail_y_pass": 4,
"d_neither": 6
},
"x_rate_kn": "34/44 (77.3%)",
"y_rate_kn": "19/44 (43.2%)",
"diff_x_minus_y": 0.3409090909090909,
"diff_pct_points": 34.09090909090909,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 19,
"c": 4,
"n_discordant": 23,
"p_value": 0.002599477767944336
},
"ci95_diff_newcombe": [
0.14041635147589288,
0.5054228250754605
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.1590909090909091,
"hi": 0.5227272727272727
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions",
"n_discordant": 23,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
}
}
},
"exclusion_classes_within_this_class": {
"base_q8": {
"scored": 48,
"truncated": 2
},
"tuned83k_q8": {
"scored": 45,
"truncated": 5
},
"tuned220k_q8": {
"scored": 49,
"truncated": 1
}
}
},
"service": {
"INDICATIVE": true,
"cell_note": "Per-class cell, n=50, indicative. Cell counts are printed everywhere so nobody over-reads a k-of-n; a per-class CI is given but it settles nothing on its own.",
"per_arm": {
"base_q8": {
"inclusive": {
"k": 23,
"n": 50,
"rate_pct": 46.0,
"kn": "23/50 (46.0%)",
"ci95": [
0.3296965219222435,
0.5960112860084253
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 20,
"n": 50,
"kn": "20/50 (40.0%)",
"ci95": [
0.27608389730256544,
0.5381856225241064
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 23,
"n": 50,
"rate_pct": 46.0,
"kn": "23/50 (46.0%)",
"ci95": [
0.3296965219222435,
0.5960112860084253
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 20,
"n": 50,
"kn": "20/50 (40.0%)",
"ci95": [
0.27608389730256544,
0.5381856225241064
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned83k_q8": {
"inclusive": {
"k": 29,
"n": 50,
"rate_pct": 58.0,
"kn": "29/50 (58.0%)",
"ci95": [
0.4423344176857823,
0.7062499664528802
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 23,
"n": 50,
"kn": "23/50 (46.0%)",
"ci95": [
0.3296965219222435,
0.5960112860084253
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 29,
"n": 50,
"rate_pct": 58.0,
"kn": "29/50 (58.0%)",
"ci95": [
0.4423344176857823,
0.7062499664528802
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 23,
"n": 50,
"kn": "23/50 (46.0%)",
"ci95": [
0.3296965219222435,
0.5960112860084253
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
},
"tuned220k_q8": {
"inclusive": {
"k": 48,
"n": 50,
"rate_pct": 96.0,
"kn": "48/50 (96.0%)",
"ci95": [
0.8653990931249297,
0.9889611156723802
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 48,
"n": 50,
"kn": "48/50 (96.0%)",
"ci95": [
0.8653990931249297,
0.9889611156723802
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
},
"strict": {
"k": 48,
"n": 50,
"rate_pct": 96.0,
"kn": "48/50 (96.0%)",
"ci95": [
0.8653990931249297,
0.9889611156723802
],
"ci_method": "Wilson score interval (no continuity correction)",
"secondary_spec_conformance": {
"field": "verdict_strict",
"k": 48,
"n": 50,
"kn": "48/50 (96.0%)",
"ci95": [
0.8653990931249297,
0.9889611156723802
],
"ci_method": "Wilson score interval",
"FLAG": "self-declared, non-independent; see statistics_v1_1.secondary_spec_conformance_flag"
}
}
}
},
"paired": {
"tuned83k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 13,
"b_x_pass_y_fail": 16,
"c_x_fail_y_pass": 10,
"d_neither": 11
},
"x_rate_kn": "29/50 (58.0%)",
"y_rate_kn": "23/50 (46.0%)",
"diff_x_minus_y": 0.12,
"diff_pct_points": 12.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 16,
"c": 10,
"n_discordant": 26,
"p_value": 0.32693958282470703
},
"ci95_diff_newcombe": [
-0.07617870358394077,
0.3039227266392255
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.08,
"hi": 0.32
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions",
"n_discordant": 26,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
},
"strict": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 13,
"b_x_pass_y_fail": 16,
"c_x_fail_y_pass": 10,
"d_neither": 11
},
"x_rate_kn": "29/50 (58.0%)",
"y_rate_kn": "23/50 (46.0%)",
"diff_x_minus_y": 0.12,
"diff_pct_points": 12.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 16,
"c": 10,
"n_discordant": 26,
"p_value": 0.32693958282470703
},
"ci95_diff_newcombe": [
-0.07617870358394077,
0.3039227266392255
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": -0.08,
"hi": 0.32
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions",
"n_discordant": 26,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
}
},
"tuned220k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 23,
"b_x_pass_y_fail": 25,
"c_x_fail_y_pass": 0,
"d_neither": 2
},
"x_rate_kn": "48/50 (96.0%)",
"y_rate_kn": "23/50 (46.0%)",
"diff_x_minus_y": 0.5,
"diff_pct_points": 50.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 25,
"c": 0,
"n_discordant": 25,
"p_value": 5.960464477539063e-08
},
"ci95_diff_newcombe": [
0.3496662223470193,
0.6280461412539109
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.36,
"hi": 0.64
},
"dominance": "X strictly dominates Y (c=0)",
"n_discordant": 25,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 23,
"b_x_pass_y_fail": 25,
"c_x_fail_y_pass": 0,
"d_neither": 2
},
"x_rate_kn": "48/50 (96.0%)",
"y_rate_kn": "23/50 (46.0%)",
"diff_x_minus_y": 0.5,
"diff_pct_points": 50.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 25,
"c": 0,
"n_discordant": 25,
"p_value": 5.960464477539063e-08
},
"ci95_diff_newcombe": [
0.3496662223470193,
0.6280461412539109
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.36,
"hi": 0.64
},
"dominance": "X strictly dominates Y (c=0)",
"n_discordant": 25,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
}
},
"tuned220k_q8__vs__tuned83k_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 29,
"b_x_pass_y_fail": 19,
"c_x_fail_y_pass": 0,
"d_neither": 2
},
"x_rate_kn": "48/50 (96.0%)",
"y_rate_kn": "29/50 (58.0%)",
"diff_x_minus_y": 0.38,
"diff_pct_points": 38.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 19,
"c": 0,
"n_discordant": 19,
"p_value": 3.814697265625e-06
},
"ci95_diff_newcombe": [
0.24158549713890842,
0.5137079862933835
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.24,
"hi": 0.52
},
"dominance": "X strictly dominates Y (c=0)",
"n_discordant": 19,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 50,
"contingency": {
"a_both_pass": 29,
"b_x_pass_y_fail": 19,
"c_x_fail_y_pass": 0,
"d_neither": 2
},
"x_rate_kn": "48/50 (96.0%)",
"y_rate_kn": "29/50 (58.0%)",
"diff_x_minus_y": 0.38,
"diff_pct_points": 38.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 19,
"c": 0,
"n_discordant": 19,
"p_value": 3.814697265625e-06
},
"ci95_diff_newcombe": [
0.24158549713890842,
0.5137079862933835
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.24,
"hi": 0.52
},
"dominance": "X strictly dominates Y (c=0)",
"n_discordant": 19,
"COMPUTABLE": true,
"DISCORDANCE_THIN": false
}
}
},
"exclusion_classes_within_this_class": {
"base_q8": {
"scored": 50
},
"tuned83k_q8": {
"scored": 50
},
"tuned220k_q8": {
"scored": 50
}
}
}
}
},
"paired": {
"full": {
"tuned83k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 200,
"contingency": {
"a_both_pass": 37,
"b_x_pass_y_fail": 98,
"c_x_fail_y_pass": 19,
"d_neither": 46
},
"x_rate_kn": "135/200 (67.5%)",
"y_rate_kn": "56/200 (28.0%)",
"diff_x_minus_y": 0.395,
"diff_pct_points": 39.5,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 98,
"c": 19,
"n_discordant": 117,
"p_value": 5.139554351717117e-14
},
"ci95_diff_newcombe": [
0.2996261196608533,
0.4797735297607776
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.305,
"hi": 0.485
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B1 tuned 83k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 180,
"contingency": {
"a_both_pass": 37,
"b_x_pass_y_fail": 92,
"c_x_fail_y_pass": 19,
"d_neither": 32
},
"x_rate_kn": "129/180 (71.7%)",
"y_rate_kn": "56/180 (31.1%)",
"diff_x_minus_y": 0.40555555555555556,
"diff_pct_points": 40.55555555555556,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 92,
"c": 19,
"n_discordant": 111,
"p_value": 1.1236121882237838e-12
},
"ci95_diff_newcombe": [
0.30191683497659605,
0.4967304007668792
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.3055555555555556,
"hi": 0.5055555555555555
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
},
"tuned220k_q8__vs__base_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 200,
"contingency": {
"a_both_pass": 54,
"b_x_pass_y_fail": 114,
"c_x_fail_y_pass": 2,
"d_neither": 30
},
"x_rate_kn": "168/200 (84.0%)",
"y_rate_kn": "56/200 (28.0%)",
"diff_x_minus_y": 0.56,
"diff_pct_points": 56.0,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 114,
"c": 2,
"n_discordant": 116,
"p_value": 1.6339108165695212e-31
},
"ci95_diff_newcombe": [
0.4824232034252644,
0.6248536127419356
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.49,
"hi": 0.63
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "A base Qwen2.5-Coder-14B-Instruct Q8_0",
"n_pairs": 180,
"contingency": {
"a_both_pass": 54,
"b_x_pass_y_fail": 106,
"c_x_fail_y_pass": 2,
"d_neither": 18
},
"x_rate_kn": "160/180 (88.9%)",
"y_rate_kn": "56/180 (31.1%)",
"diff_x_minus_y": 0.5777777777777777,
"diff_pct_points": 57.77777777777778,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 106,
"c": 2,
"n_discordant": 108,
"p_value": 3.6281438664344504e-29
},
"ci95_diff_newcombe": [
0.4956667524788013,
0.6459876482181227
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.5,
"hi": 0.65
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
},
"tuned220k_q8__vs__tuned83k_q8": {
"inclusive": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 200,
"contingency": {
"a_both_pass": 119,
"b_x_pass_y_fail": 49,
"c_x_fail_y_pass": 16,
"d_neither": 16
},
"x_rate_kn": "168/200 (84.0%)",
"y_rate_kn": "135/200 (67.5%)",
"diff_x_minus_y": 0.165,
"diff_pct_points": 16.5,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 49,
"c": 16,
"n_discordant": 65,
"p_value": 5.084767048790177e-05
},
"ci95_diff_newcombe": [
0.0884623286274878,
0.2396122831867154
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.09,
"hi": 0.24
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
},
"strict": {
"arm_x": "B2 tuned 220k Q8_0",
"arm_y": "B1 tuned 83k Q8_0",
"n_pairs": 182,
"contingency": {
"a_both_pass": 119,
"b_x_pass_y_fail": 43,
"c_x_fail_y_pass": 12,
"d_neither": 8
},
"x_rate_kn": "162/182 (89.0%)",
"y_rate_kn": "131/182 (72.0%)",
"diff_x_minus_y": 0.17032967032967034,
"diff_pct_points": 17.032967032967033,
"mcnemar": {
"test": "McNemar exact (binomial, two-sided)",
"b": 43,
"c": 12,
"n_discordant": 55,
"p_value": 3.3053764732060564e-05
},
"ci95_diff_newcombe": [
0.09348723933106229,
0.24599582265139153
],
"ci95_diff_newcombe_method": "Newcombe (1998) method 10 for the difference of two paired proportions (Wilson intervals + phi)",
"ci95_diff_bootstrap": {
"method": "paired bootstrap over items (percentile, 2.5/97.5)",
"resamples": 20000,
"seed": 20260911,
"lo": 0.09340659340659341,
"hi": 0.24725274725274726
},
"dominance": "NO dominance - each arm fails items the other passes, in both directions"
}
}
}
},
"harness_bucket_counts_per_arm": {
"base_q8": {
"compile-fail": 130,
"compile-pass": 56,
"TRUNCATED": 14
},
"tuned83k_q8": {
"compile-pass": 135,
"compile-fail": 52,
"TRUNCATED": 13
},
"tuned220k_q8": {
"compile-pass": 168,
"compile-fail": 20,
"TRUNCATED": 12
}
},
"exclusion_counts_by_class_per_arm": {
"scored": {
"base_q8": 186,
"tuned83k_q8": 187,
"tuned220k_q8": 188
},
"truncated": {
"base_q8": 14,
"tuned83k_q8": 13,
"tuned220k_q8": 12
}
},
"exclusion_buckets_by_artifact_class": {
"base_q8": {
"truncated": {
"indicator": 5,
"include": 7,
"script": 2
}
},
"tuned83k_q8": {
"truncated": {
"include": 8,
"script": 5
}
},
"tuned220k_q8": {
"truncated": {
"include": 11,
"script": 1
}
}
},
"denominator_pools": {
"full": 200
},
"item_alignment": {
"join_key": "ea_name",
"per_arm_unique_names": {
"base_q8": 200,
"tuned83k_q8": 200,
"tuned220k_q8": 200
},
"intersection_n": 200,
"per_arm_names_missing_from_intersection": {},
"duplicate_ea_names_per_arm": {},
"ALIGNED": true,
"prompt_identity": {
"items_with_identical_prompt_sha256_across_all_arms": 200,
"items_disagreeing_with_the_majority_prompt_sha256_per_arm": {
"base_q8": 0,
"tuned83k_q8": 0,
"tuned220k_q8": 0
},
"ALL_ARMS_BYTE_IDENTICAL_PROMPTS": true,
"note": "a differing prompt_sha256 means the arm was served a differently assembled prompt string; the head-to-head with that arm is confounded and is reported as such"
},
"spec_identity": {
"items_with_identical_spec_sha256_across_all_arms": 200,
"items_disagreeing_with_the_majority_spec_sha256_per_arm": {
"base_q8": 0,
"tuned83k_q8": 0,
"tuned220k_q8": 0
},
"rows_missing_spec_sha256_per_arm": {
"base_q8": 0,
"tuned83k_q8": 0,
"tuned220k_q8": 0
},
"ALL_ARMS_IDENTICAL_SPEC": true,
"note": "spec_sha256 identity is the pairing proof (the same item was put to every arm); prompt_sha256 identity is the separate byte-identity requirement of the pre-registered protocol (see the top-level `protocol` field) and is reported beside it"
},
"manifest_selection_sha256_identical_across_arms": true
},
"instrument_validation": {
"base_q8": {
"known_positive": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 2,
"row_errors_field": 2,
"PASS": true
},
"negative_control": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 0,
"result_line": "Result: 0 errors, 0 warnings",
"PASS": true
},
"utf16_trap": {
"raw_is_utf16_le_bom": true,
"byte_grep_ascii_hits_in_raw": 0,
"decoded_hits": 2,
"meaning": "a GNU-grep-style byte match on the .raw.log returns 0 while the decoded log yields the true count; the taxonomy is built on the decoded text"
}
},
"tuned83k_q8": {
"known_positive": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 3,
"row_errors_field": 3,
"PASS": true
},
"negative_control": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 0,
"result_line": "Result: 0 errors, 0 warnings",
"PASS": true
},
"utf16_trap": {
"raw_is_utf16_le_bom": true,
"byte_grep_ascii_hits_in_raw": 0,
"decoded_hits": 3,
"meaning": "a GNU-grep-style byte match on the .raw.log returns 0 while the decoded log yields the true count; the taxonomy is built on the decoded text"
}
},
"tuned220k_q8": {
"known_positive": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 12,
"row_errors_field": 12,
"PASS": true
},
"negative_control": {
"decoded_as": "utf-8/ascii",
"n_error_lines_found": 0,
"result_line": "Result: 0 errors, 0 warnings",
"PASS": true
},
"utf16_trap": {
"raw_is_utf16_le_bom": true,
"byte_grep_ascii_hits_in_raw": 0,
"decoded_hits": 12,
"meaning": "a GNU-grep-style byte match on the .raw.log returns 0 while the decoded log yields the true count; the taxonomy is built on the decoded text"
}
}
},
"artifact_class_counts": {
"indicator": 50,
"include": 50,
"script": 50,
"service": 50
},
"category_field": "category values observed: ['include', 'indicator', 'script', 'service']",
"category_key_present_on_all_rows": true,
"contaminated_items_excluded_in_clean_pool": [],
"inclusive_scoring_anomalies": {
"base_q8": {
"n_anomalous_rows": 0,
"rows": [],
"n_denominator_inclusive": 200,
"inclusive_k_under_the_PREREGISTERED_rule": 56,
"inclusive_k_if_the_row_verdict_were_honoured": 56,
"rule_applied": "an item in an excluded class is scored as a failure under the inclusive denominator (the wording of the pre-registered protocol; see the top-level `protocol` field)"
},
"tuned83k_q8": {
"n_anomalous_rows": 0,
"rows": [],
"n_denominator_inclusive": 200,
"inclusive_k_under_the_PREREGISTERED_rule": 135,
"inclusive_k_if_the_row_verdict_were_honoured": 135,
"rule_applied": "an item in an excluded class is scored as a failure under the inclusive denominator (the wording of the pre-registered protocol; see the top-level `protocol` field)"
},
"tuned220k_q8": {
"n_anomalous_rows": 1,
"rows": [
{
"category": "include",
"bucket": "TRUNCATED",
"truncated": true,
"finish_reason": "length",
"errors": 0,
"ex5_exists": true,
"verdict_headline": true,
"verdict_strict": true
}
],
"n_denominator_inclusive": 200,
"inclusive_k_under_the_PREREGISTERED_rule": 168,
"inclusive_k_if_the_row_verdict_were_honoured": 169,
"rule_applied": "an item in an excluded class is scored as a failure under the inclusive denominator (the wording of the pre-registered protocol; see the top-level `protocol` field)"
}
},
"NO_FRONTIER_ARM": "No frontier arm exists on this set: gpt-5.6-sol was never run against the 200 non-EA holdout."
}
},
"replay_c200": {
"title": "Stability check - the 220k arm re-run on the same 200 non-EA prompts under the same configuration",
"derived_from": "the published rows in holdout_nonea200_results.jsonl, arms tuned220k and tuned220k_q8_replay",
"pre_registered_rule": "fixed before the second run started: the first run is the headline regardless of outcome, both runs publish, neither is averaged and neither is selected",
"first_run": {
"arm": "tuned220k",
"passes_all_200": 168,
"raw_verdict_headline_true": 169,
"truncated": 12,
"per_class": {
"include": {
"n": 50,
"passes": 31,
"truncated": 11
},
"indicator": {
"n": 50,
"passes": 50,
"truncated": 0
},
"script": {
"n": 50,
"passes": 39,
"truncated": 1
},
"service": {
"n": 50,
"passes": 48,
"truncated": 0
}
}
},
"second_run": {
"arm": "tuned220k_q8_replay",
"passes_all_200": 169,
"raw_verdict_headline_true": 169,
"truncated": 11,
"per_class": {
"include": {
"n": 50,
"passes": 31,
"truncated": 10
},
"indicator": {
"n": 50,
"passes": 50,
"truncated": 0
},
"script": {
"n": 50,
"passes": 40,
"truncated": 1
},
"service": {
"n": 50,
"passes": 48,
"truncated": 0
}
}
},
"verdict_flips_raw": 2,
"verdict_flips_under_the_scoring_rule": 1,
"generated_source_byte_identical_items": 164,
"n_items": 200,
"meaning": "Both runs used the same recorded settings; this release does not isolate the cause of the differences. On a row where generation produced source, the output hash is provenance of the observed run, not a promise that a re-generation reproduces it."
}
}
}
|