{
  "notes": "EPISODE 10 of 10 = WRITE-UP. All 29 batches settled (371 of 400 items used, 29 unspent on purpose), nothing running, nothing refused. finish = true; spawned nothing (no episode 11 could read it). The write-up is next: write it from these notes and the 30 handed results.\n\nROUND 7 READ AND CHECKED (r7-t03-joint 150 items in 15 merges, r7-t03-confirm 4 items). Modes: 41 searches (37 valid), 104 real-only (slow machines, T3 vs HA), 5 blocked (j025, j034, j089, j115, j144: uv Exec format error = Darwin arm64 hosts). I re-ranked every row from the merges' JSON myself and recomputed pooled figures from n, sum, sumsq (never averaged se). I downloaded the winner's artifact: HTTP 200, file sha256 = its id, code_zb64 decodes to the code sha256 below (merge 03's `best` block carries the same code_zb64 and params; I decoded that too, same hash, 27,700 chars, `contract Strategy`).\n\nNEW BEST = J24 = r7-t03-joint/j024 (merge 03). Artifact https://api.imdnotdead.fun/artifacts/829f927c66461935528f22de289983a23fc27b162375853e8f19161a0bb644d5 ; code sha256 226d20b7c388c4e70e8fa0a2c057ed81240a4f0bbd55f1c8443be456c7ea7a25 ; fields code_zb64, params (W1 8x19, b1, W2 2x8, b2, 41 constants, scaling unchanged), gates, curve, holdout, real, gas. Recipe: CMA-ES from T3's frozen params over 219 = 178 head weights + 41 constants (weight step 0.03 x block rms, constants c0 x 2^(0.02 z) clipped, decays in 1-d space), pop 16, 512 fresh markets a generation from seed 1976000000, sigma0 1.0 under 0.5^(min/40), es_seed 824; 889 generations in 45.3 min at 3,780 markets/s (32 processes); monitor over T 2.585 -> 2.762, still rising at the end. Gates: A port vs T3 EVM 5.7e-5 bps on 50 recorded sequences of 2,000 callbacks (seeds 310000..310049); B replica CG minus T on 7232768..7233279 = +1.920 se 0.044 vs real +2.00 (within 0.15); C frozen Solidity vs port 6.4e-5 bps. Holdout 40000000..40008191 (n 8,192): over T +2.710 se 0.0135, over T3 +0.177 se 0.0054 (retail +0.149, arb +0.028). REAL (candidate alone and T3 alone vs 30 bps Vanilla, run_batch(X, Vanilla, configs, 8), MatchRunner configs): block 8792000..8793023 +0.185 se 0.0149 (679 wins of 1,024, worst -1.69), block 8794000..8795023 +0.197 se 0.0150 (692 wins, worst -2.20); pooled n 2,048 +0.191 se 0.0106 (sum 391.254, sumsq 542.144): 18 se over T3, confirmed on two fresh blocks of 1,024 (the requester's two confirmations, one machine). Gas: afterSwap peak 237,452 (block 2; 235,471 block 1), afterInitialize peak 243,234, 0 OOG, 0 reverts in 22.3M calls. The replica (+0.177) again called the real gain (+0.191) within 0.02.\nRunners-up (holdout over T3 / real over T3 pooled on 2,048 fresh seeds; artifact id, code sha256): j116 +0.146 / +0.149 se 0.011 (edb467cbe3b57f2a8c91d57d50729d98ee07a1f5b8adfe683916a1cc923f09b9, a4a71517cd...; 1,585 generations at 4,791 markets/s), j003 +0.128 / +0.150 se 0.011 (b8d6294ef034239e807f591ee9bc4038092bc6ad4c5e6a8f20dd4957668f98c9, 9386b3d1b0...), j008 +0.122 / +0.145 se 0.011 (6a5111c7264cec77eb522d341e255a2ff75fff9223f72a1d934c4dcf9b1657c0), j030 +0.152 / +0.123 one block (a6c3d192c7...), j002 +0.128 / +0.116. Items under about 1,000 markets/s gained +0.02 to +0.08 in 45-55 min: the gain tracks generations (about +0.15 to +0.19 per 900-1,600 generations from T3). Cautions: j050 holdout +0.026 but real -0.374 se 0.007 (its frozen code does not match its port; never submit it); j022 failed gate C (2.3 bps at a fee-routing threshold); j095 returned T3 unchanged; j027 and j036 incomplete blocks. No merge found any top-10 result; merges verified bytes and arithmetic only, not reruns.\n\nT3 CONFIRMATIONS (r7-t03-confirm c1..c4, T3/IB/HA alone on 9500000, 9510000, 9520000, 9530000 blocks; c1-c3 lost T3's first 64 seeds to a KIT decompose assertion (T3 initialises at BASE_FEE, not 4 bps) and c4 ran 896 seeds; nothing rerun, nothing invented): T3 minus IB n 3,776 +0.805 se 0.014 (3,284 wins; blocks +0.784, +0.821, +0.820, +0.796); T3 minus HA n 3,776 +1.448 se 0.016 (3,647 wins); IB minus HA n 3,968 +0.643 se 0.012. Quartiles by IB retail volume: T3 minus IB 0.68-0.91, even. T3's gain over IB is arbitrage (+1.0) with retail -0.2; IB's over HA is retail. Gas: T3 afterSwap 237,689 peak, init 243,234, 0 OOG in 41M calls; IB 237,181; HA 233,781. With note 9's 3,072 seeds, T3 minus IB is about +0.80 on 6,848 fresh seeds. Real-only rows in merges 02, 03, 07-10, 12, 14, 15 give T3 minus HA +1.39 to +1.43 per group (about 60,000 more seeds), consistent.\n\nFINAL LADDER (fresh real seeds; over IB chained across different seed blocks where not paired directly): HA = T + 1.257 (objective). IB = HA + 0.643 to 0.652 (n 3,968 + 9,216). T3 = IB + 0.805 (n 3,776; +0.80 on 3,072 in note 9) = HA + 1.448 (direct, n 3,776). J24 = T3 + 0.191 (direct, n 2,048) = about IB + 1.00 = about HA + 1.64 = about T + 2.90; holdout J24 over T +2.710. Board (note 6, 2026-11-09): IB 522.00 (20th), 19th 522.05, 14th 522.40, 11th 523.82, 10th 524.66. Expected: T3 about 522.8 (12th-14th), J24 about 523.0 (12th-13th, below 11th by about 0.8). Top 10 needs +2.66 over IB: J24 is about 1.66 short. The goal's aim (top 10) was NOT reached; the entry moved from 27th (RS2 520.05) to 20th (IB, on the board) with files in hand for about 12th-13th. Board transfer held within 0.05 for IB (local +2.0, board +1.95).\n\nFILE FOR THE REQUESTER: J24 (code sha256 226d20b7..., >= +0.3 over IB on 2,048 fresh seeds by a wide margin: +0.191 over T3 which is +0.805 over IB; direct J24 vs IB was never run, say so). Fallback T3 (artifact 2333ecf28d4356bd232b85c1e55842e94e1ada05e1cd067830b5625dfdbc5336, code sha256 cb6d945511e1e10af78baa64c2a6cb25806f0998403ce6440654bfc3c70092d9, confirmed +0.805 over IB on 3,776 + 3,072 seeds). J5 (dd53db1f...) +0.570 over IB on 3,072 remains confirmed too.\n\nLINEAGE AND LESSONS for the write-up: HA (13-8-2 head on T) -> IA/IB (19 inputs: +0.36/+0.65 over HA; IB on the board 522.00) -> CG (IA's head continued, +0.18 over IB) -> T3 (CG's head + 41 of T's constants trained jointly from T's values, +0.58 over CG, +0.805 over IB) -> J24 (T3 continued jointly 889 generations, +0.19). Lines: added inputs +0.35 to +0.65 a round; long weight training +0.2 to +0.5 a round; T's constants trained jointly on a good head +0.5 to +0.8 (arb side); later joint continuations +0.1 to +0.2 per 900-1,600 generations, scaling with markets/s; unions, width (12-16 units), gating, weight mixing: nothing. The replica predicted real gains within 0.02-0.1 at every step. Operations: about a third of machines reach 500 markets/s and one in ten 1,300+; one in eight items lost to Darwin arm64 hosts at SETUP; slow machines did confirmations instead of searches (104 of 150 in round 7); the plane's `sources` check flagged nearly every accepted result without naming a page; no item failed otherwise except one path_violation (r5-cg-continue/k10). All gas peaks stayed under 238,500 afterSwap and 243,234 init, 0 OOG on every valid candidate.\n\nWRITE-UP CITATIONS: r7-t03-joint/merge-03 (J24 row, pooled 2,048, best block with code_zb64 and params), merges 01, 12 (runners-up), merge 05 (j050 caution), r7-t03-confirm/c1..c4 (T3 vs IB and HA), r6-cg-joint-t/merge-01 (T3's training row), r6-confirm/j5-a,b,c (J5), r5-cg-confirm/cf-a and r4-confirm/ib-a (IB over HA), r6-j2-joint/merge-01, r5-cg-continue/merge-02..04 (weights-only ceiling), r4-ib-continue/c2.",
  "cancel": [],
  "finish": true
}