# Merge 03 of batch r6-j5-joint: items k21–k30 (J5 continuation and real-engine check against IB) **Inputs.** There are ten item results, `.imd/reads/artifacts/item-k21` … `item-k30`. Every file's sha256 matches the id it was given (listed in the `rows` below). For every item, the merge decompressed `code_zb64` and confirmed that its sha256 equals the item's stated `code_sha256`. None of the results is missing or empty. **Source caveat.** The plane reports that each item cited sources that did not hold when they were read again: k21 had 4, k22 1, k23 3, k24 2, k25 3, k26 2, k27 2, k28 2, k29 2 and k30 2. The items' sources files were not given to this merge, so it cannot say which citations failed. The items say what their sources cover: the supplied artifacts (J5 fe2320b5, IA df0ddfdf, IB afd5183e, replica e8a7a2af, HA 2801efc8) and the pinned engine formulas. They also say every number they report is a local measurement and is not quoted from any source (notes in k21, k22, k23, k29 and k30). The broken citations therefore weaken claims of the form "input X is what the brief says it is" or "the engine works this way". The measured figures are unaffected. Those figures are unreplicated single-machine runs: this merge only checked them for internal consistency. ## Summary - **Search ran in 1 of 10 items (k24).** The other nine measured replica speed below the 500 markets/s threshold. They therefore took the brief's real-only branch and kept J5 unchanged (code sha256 `dd53db1f…`). Their measured speeds were 168–448 markets/s, while k24 measured 517.8 markets/s on 4 cores. - **k24's search found no gain.** It ran 45.62 minutes and 62 generations of CMA-ES (pop 16, 1024 markets per generation). It froze at the best monitored point, 10.32 min / gen 20, at +2.3775 se 0.0295 over T. J5 scored +2.3568 on the same monitor. On the holdout (seeds 40000000..40008191, n 8192) the frozen candidate scored **+2.306775 se 0.01357 over T**, against **+2.307544 se 0.01353** for J5. That makes it **−0.0008 se 0.0046 vs J5**, which k24 itself describes as "essentially J5 with small perturbations". On the real engine it measured **+0.5509 se 0.0244 vs IB** (n 1024, seeds 8392000..8393023), with 0 OOG and an afterSwap peak of 237,693. - **J5 vs IB on the real engine** (nine real-only items, disjoint fresh seed ranges between 8380000 and 8417023): - Pooled over all nine items: **n 9088, mean +0.5517, se 0.0099**. - Pooled over only the seven items marked valid (k23, k25–k30): **n 7168, mean +0.5457, se 0.0107**. - The per-item means range from +0.518 (k25) to +0.593 (k26). These are consistent with one another given per-item se of 0.024–0.043. - The merge computed both pools from the items' `sum` and `sumsq`. It also checked that each item's reported mean and se follow from its own sum and sumsq; all ten do. - **Gas.** No item recorded an OOG call or any other callback failure. The J5 / candidate afterSwap peaks were 235,482–238,291, all within the 245,000 limit. The afterInitialize peak was 243,234 in every item. IB peaked at 234,945–237,786. - **Gate B (replica fidelity) agrees in every item.** Replica J5 minus IA came out at +0.8655 se 0.0338 on seeds 7672000..7672511, against the real +0.876 that each item quotes as its target: inside the 0.15 tolerance. Every item passed gate A, with a maximum port-vs-EVM fee difference of 5.5e-5 to 2.5e-3 bps (threshold 0.01). Gate C passed too, and is trivial wherever the candidate is J5. - **Holdout.** All items that ran a holdout used the same seeds, 40000000..40008191. They all report J5 at +2.307544 se 0.013529 over T, so this figure is one measurement reproduced and not ten independent ones. k26 did not run a holdout. k28 reports `over_base` against IA (+0.8258 se 0.0093, J5 minus IA) instead of against J5. - **Retail and arb split (J5 minus IB).** J5 gains on arbitrage and gives up some retail edge. - Reconstructed edge deltas: retail −0.48 to −0.56, arb +1.06 to +1.08. These come from k21 (n 800), k23 (n 896), k25 (n 896), k27 (n 960), k28 (n 1024) and k29 (n 560). Each item rebuilt the arb from step records itself, so these are not engine outputs and several cover only part of the seeds. - Y-volume deltas, from k22–k30: retail volume +50 to +76, arb volume −131 to −182. - In the replica holdout, the J5-over-T split was retail +4.52 / arb −2.22 (k21, k27). ## Failures and protocol issues seen - **k21, valid = false.** Two 64-seed chunks finished `run_batch`, but a post-processing check (arb-volume residual > 1e-5) raised an error before their scores were saved: J5 lost 8380160..8380223 and IB lost 8380000..8380063. Neither chunk was rerun, which leaves 896 paired seeds of the 1024 requested. All gates and gas checks passed. - **k22, valid = false (marked so by k22 itself).** An exporter failure meant seeds 8384000..8384063 were run a second time on both arms with identical bytecode. Each seed counts only once, and no seed was used for selection. The numbers and gas checks are valid. - **k27** reran one discarded 16-seed batch per arm (an instrumentation bug), and **k28** reran one IB chunk after a command went over the 8-minute limit and was killed. Both still report valid = true and count each seed once. Strictly, this is the same kind of re-execution that k22 flagged itself invalid for. - **J5.sol reverted with a division-by-zero panic (0x12) under extreme synthetic reserve states.** Seven items saw it in an early, unbounded synthetic callback harness: k21, k25, k26, k27, k28, k29 and k30. k26 and k28 name sequence seed 310037, where spot reached about 7e7. None of these items saw any revert in the market-driven gate sequences or in roughly 11 million real-engine afterSwap calls per item. This is a robustness caveat, not an invalidation. - **Retail/arb edge split is incomplete.** Several items lost step data for their early chunks, so their reconstructed edge splits cover only some of the seeds (sizes above). k22, k24, k26 and k30 report volume deltas only. ## Were curves still rising? There is only one real curve, k24's, with 10 points. It went 2.3568 → 2.3640 → **2.3775 (peak at 10.3 min)** → 2.3296 → 2.3461 → 2.3514 → 2.3637 → 2.3588 → 2.3698 → 2.3660 at 45.6 min, while sigma fell from 1.0 to 0.348. Every point after the first lies within about 1 monitor se (0.029) of J5. **The curve was flat, not rising**, and holdout confirms no gain. This contrasts with the pilot's note that curves were still rising. k24 itself warns that 62 generations is small: J5 had 1,496 generations on 32 cores. The flat result therefore says little about long runs on fast hardware. k22's single curve point is J5's own monitor value (+2.3568) with no search behind it. The other eight items have no curve. ## Implication for the goal No item in this group improved on J5. The only candidate that differs from J5 (k24) is statistically the same as J5. J5's lead over IB on the real engine is now well measured: about +0.55 (se ≈ 0.01) over roughly 9,000 fresh seeds, and it was reproduced on every machine. The group yields no new step toward the roughly +3.1 over HA the goal needs. The bottleneck was replica throughput: 4–8-core machines reached only 168–518 markets/s. ## Merged table (JSON) Numbers are given as strings. `real_vs_IB` is the candidate minus IB on the real engine; the candidate is J5 in every row except k24. `holdout_base` names what `over_base` is measured against. `afterSwap_peak` is the candidate's Submission-role maximum from `gas_stats()`. `best` is k24, the only search row; its `code_zb64`, `code_sha256` and `params` are copied whole from item-k24. Note that k24's holdout over T sits 0.0008 below J5's (k21–k23, k25, k27–k30). ```json { "rows": [ { "key": "k21", "mode": "real-only", "result_sha256": "703775d0b72b4ce34889f82b585d9011dbbf5ee57d6b3d24c116770494731d02", "markets_per_s": "418.01", "generations": "0", "search_minutes": "0", "holdout_over_T": "2.307544068661547", "holdout_over_T_se": "0.013528690579041706", "holdout_over_base": "0.0", "holdout_over_base_se": "0.0", "holdout_base": "J5 (candidate is J5, identically 0)", "real_vs_IB_mean": "0.5670358479237297", "real_vs_IB_se": "0.042741198260174", "real_vs_IB_n": "896", "afterSwap_peak": "237586", "OOG": "0", "valid": "false", "last_curve_point": null, "candidate_code_sha256": "dd53db1fbe42628e040c6c7b7a82689867ab601263c5e740506299a3a186b38c" }, { "key": "k22", "mode": "real-only", "result_sha256": "5183e6c242f9a7a56ab51bd3137c2b958c5269ab63688fb1de8eb6fc84be7124", "markets_per_s": "354.33", "generations": "0", "search_minutes": "0", "holdout_over_T": "2.307544068661547", "holdout_over_T_se": "0.013528690579041706", "holdout_over_base": "0.0", "holdout_over_base_se": "0.0", "holdout_base": "J5 (candidate is J5, identically 0)", "real_vs_IB_mean": "0.5797335654448794", "real_vs_IB_se": "0.02547602111344616", "real_vs_IB_n": "1024", "afterSwap_peak": "235482", "OOG": "0", "valid": "false", "last_curve_point": [ "0.0", "2.3568067935702803", null ], "candidate_code_sha256": "dd53db1fbe42628e040c6c7b7a82689867ab601263c5e740506299a3a186b38c" }, { "key": "k23", "mode": "real-only", "result_sha256": "3ae28f9062ec7e9e9cf2671bfd433fd2caf359c49517022eb636adc7eca75ac9", "markets_per_s": "416.07", "generations": "0", "search_minutes": "0", "holdout_over_T": "2.307544068661547", "holdout_over_T_se": "0.013528690579041706", "holdout_over_base": "0.0", "holdout_over_base_se": "0.0", "holdout_base": "J5 (candidate is J5, identically 0)", "real_vs_IB_mean": "0.5613316599994707", "real_vs_IB_se": "0.02910621275025325", "real_vs_IB_n": "1024", "afterSwap_peak": "237598", "OOG": "0", "valid": "true", "last_curve_point": null, "candidate_code_sha256": "dd53db1fbe42628e040c6c7b7a82689867ab601263c5e740506299a3a186b38c" }, { "key": "k24", "mode": "search", "result_sha256": "ab36b617913ccf573dd95600fe4b4a0825b465af6912793b955f18be5a11e532", "markets_per_s": "517.8", "generations": "62", "search_minutes": "45.62", "holdout_over_T": "2.306775444200554", "holdout_over_T_se": "0.01357016729674729", "holdout_over_base": "-0.0007686244609929754", "holdout_over_base_se": "0.0045677889593102", "holdout_base": "J5", "real_vs_IB_mean": "0.5509001808844094", "real_vs_IB_se": "0.02442928177119383", "real_vs_IB_n": "1024", "afterSwap_peak": "237693", "OOG": "0", "valid": "true", "last_curve_point": [ "45.62", "2.366047363798054", "0.34849632583356355" ], "candidate_code_sha256": "89a125cdde7bcdf7603dcb07e29f6c8f16b2966a34c1b26978212f0c9417740d" }, { "key": "k25", "mode": "real-only", "result_sha256": "2f856a28205e39396f804a9e82da5a70bf613daf84bc985d60f9e0d35cc0b51d", "markets_per_s": "285.21", "generations": "0", "search_minutes": "0", "holdout_over_T": "2.307544068661547", "holdout_over_T_se": "0.013528690579041706", "holdout_over_base": "0.0", "holdout_over_base_se": "0.0", "holdout_base": "J5 (candidate is J5, identically 0)", "real_vs_IB_mean": "0.5179463363105676", "real_vs_IB_se": "0.027565766401464735", "real_vs_IB_n": "1024", "afterSwap_peak": "238054", "OOG": "0", "valid": "true", "last_curve_point": null, "candidate_code_sha256": "dd53db1fbe42628e040c6c7b7a82689867ab601263c5e740506299a3a186b38c" }, { "key": "k26", "mode": "real-only", "result_sha256": "9b14005117bbc8d21b40620530bd1549d9ccf955abc0186b14cb997e87ffa72f", "markets_per_s": "359.27", "generations": "0", "search_minutes": "0", "holdout_over_T": null, "holdout_over_T_se": null, "holdout_over_base": null, "holdout_over_base_se": null, "holdout_base": "not run", "real_vs_IB_mean": "0.5932222621077041", "real_vs_IB_se": "0.03479245287991122", "real_vs_IB_n": "1024", "afterSwap_peak": "237574", "OOG": "0", "valid": "true", "last_curve_point": null, "candidate_code_sha256": "dd53db1fbe42628e040c6c7b7a82689867ab601263c5e740506299a3a186b38c" }, { "key": "k27", "mode": "real-only", "result_sha256": "4a8264e7009000fe0aa60130626fa36f0cfb9abe88d4cfd947c682aa22a208ab", "markets_per_s": "310.07", "generations": "0", "search_minutes": "0", "holdout_over_T": "2.307544068661547", "holdout_over_T_se": "0.013528690579041706", "holdout_over_base": "0.0", "holdout_over_base_se": "0.0", "holdout_base": "J5 (candidate is J5, identically 0)", "real_vs_IB_mean": "0.52586537633738", "real_vs_IB_se": "0.027676373572693678", "real_vs_IB_n": "1024", "afterSwap_peak": "238260", "OOG": "0", "valid": "true", "last_curve_point": null, "candidate_code_sha256": "dd53db1fbe42628e040c6c7b7a82689867ab601263c5e740506299a3a186b38c" }, { "key": "k28", "mode": "real-only", "result_sha256": "813e19815f5da9bca25eee449b0b882b79ca514b895d20381635e512e059e42f", "markets_per_s": "448.48", "generations": "0", "search_minutes": "0", "holdout_over_T": "2.307544068661547", "holdout_over_T_se": "0.013528690579041706", "holdout_over_base": "0.8258328396747168", "holdout_over_base_se": "0.00934936923764563", "holdout_base": "IA (J5 minus IA)", "real_vs_IB_mean": "0.5337724667273851", "real_vs_IB_se": "0.025523597267120723", "real_vs_IB_n": "1024", "afterSwap_peak": "237675", "OOG": "0", "valid": "true", "last_curve_point": null, "candidate_code_sha256": "dd53db1fbe42628e040c6c7b7a82689867ab601263c5e740506299a3a186b38c" }, { "key": "k29", "mode": "real-only", "result_sha256": "37de35f11be9530648fcc1421acc39ab578852eb32e9c2b78bbac78b9d3c3ae2", "markets_per_s": "168.24", "generations": "0", "search_minutes": "0", "holdout_over_T": "2.307544068661547", "holdout_over_T_se": "0.013528690579041706", "holdout_over_base": "0.0", "holdout_over_base_se": "0.0", "holdout_base": "J5 (candidate is J5, identically 0)", "real_vs_IB_mean": "0.5648229548848243", "real_vs_IB_se": "0.026427997297294538", "real_vs_IB_n": "1024", "afterSwap_peak": "238221", "OOG": "0", "valid": "true", "last_curve_point": null, "candidate_code_sha256": "dd53db1fbe42628e040c6c7b7a82689867ab601263c5e740506299a3a186b38c" }, { "key": "k30", "mode": "real-only", "result_sha256": "57915cd8230da5389dd76886adfc2ad2b6d76c9ac8a245d96db82a47acd57f7c", "markets_per_s": "364.2", "generations": "0", "search_minutes": "0", "holdout_over_T": "2.307544068661547", "holdout_over_T_se": "0.013528690579041706", "holdout_over_base": "0.0", "holdout_over_base_se": "0.0", "holdout_base": "J5 (candidate is J5, identically 0)", "real_vs_IB_mean": "0.5232863649069781", "real_vs_IB_se": "0.026210348229595132", "real_vs_IB_n": "1024", "afterSwap_peak": "238291", "OOG": "0", "valid": "true", "last_curve_point": null, "candidate_code_sha256": "dd53db1fbe42628e040c6c7b7a82689867ab601263c5e740506299a3a186b38c" } ], "best": { "key": "k24", "holdout_over_T": "2.306775444200554", "holdout_over_T_se": "0.01357016729674729", "holdout_over_J5": "-0.0007686244609929754", "holdout_over_J5_se": "0.0045677889593102", "real_vs_IB": { "n": "1024", "mean": "0.5509001808844094", "se": "0.02442928177119383", "seeds": [ "8392000", "8393023" ] }, "code_sha256": "89a125cdde7bcdf7603dcb07e29f6c8f16b2966a34c1b26978212f0c9417740d", "code_zb64": "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", "params": { "architecture": "19-8-2 as IA/J5 (ReLU clipped [0,4], inputs clipped [-4,4], output bps added to T-path fees)", "W1": [ [ 1.03202, -0.832193, -1.281535, 2.806759, -2.120711, 1.878508, -0.648459, 2.128709, -3.614814, -2.464706, -0.668723, -0.541911, 0.474533, -1.71747, -1.131252, 0.594463, -0.023407, -0.569117, 0.556143 ], [ 0.929, 0.178222, 1.773766, 3.020302, 1.63698, 0.903257, -0.270848, -0.117846, -0.365456, 0.111278, -0.228422, -1.424442, 0.308766, -0.319173, -0.094924, 0.995081, -0.150109, 0.062847, 0.230465 ], [ 1.78595, 2.54009, -0.149818, 1.601522, 1.088022, 4.23628, -1.485735, 1.342419, 1.685892, -1.096113, 0.008293, -1.478216, -0.435619, 0.271399, 0.326608, 1.276111, -1.149183, 0.064588, -1.249438 ], [ -0.626043, -1.001086, 1.611363, -0.504698, -1.523364, 1.45128, 0.150942, 0.709359, -0.672776, 0.350524, 0.392524, -0.988489, -2.056266, 0.801843, -0.144901, 0.411964, 3.37555, -0.056808, -0.598489 ], [ 0.951077, 0.775554, 4.559094, -0.03293, -1.977757, -2.151696, -0.412823, -1.022071, -0.294202, 0.030462, 0.060333, -1.563349, 0.245117, -1.083139, 0.008481, -0.414236, 2.613394, -0.024475, -0.271313 ], [ 1.219442, -1.941555, 0.565979, -0.567159, 2.073781, 1.043038, 2.311549, 1.478906, -0.780457, 0.246137, 0.089741, 3.71681, 0.519928, -0.769844, -0.051782, 1.730695, -0.873693, -0.782116, 0.59783 ], [ -2.056667, 0.401743, 0.524689, -1.815704, 2.374613, 1.191871, -0.680999, 1.445591, 1.48582, -0.655796, 0.581245, 1.443258, 0.92928, 0.65978, -0.602001, -0.247005, 1.215655, -0.081992, 1.269464 ], [ -3.40871, -1.819362, 4.51946, 3.514584, -1.294435, -1.525309, -1.359552, 0.941559, 2.565856, -0.877447, 2.570605, -1.463949, 0.79062, -0.807969, 0.465094, 0.649043, 0.938555, -1.381562, 1.547083 ] ], "b1": [ 0.544199, -0.849877, -2.740654, 0.025802, -1.135152, -1.610513, -0.837173, -0.064586 ], "W2": [ [ 0.11496, -2.41879, 0.449222, -0.709721, -3.204532, -0.650122, 0.604682, 0.699729 ], [ 0.514123, -2.378989, -1.83268, 2.322691, 0.793689, -1.719121, 0.140868, 0.644075 ] ], "b2": [ 0.848929, 1.209407 ], "constants": { "ELAPSED_CAP": 7.0, "SIGNAL_THRESHOLD": 0.001162748712664599, "DIR_DECAY": 0.6588371113025004, "ACT_DECAY": 0.5051531334370056, "SIZE_DECAY": 0.4818272859333085, "TOX_DECAY": 0.9358981921517765, "SIGMA_DECAY": 0.6729152763165316, "LAMBDA_DECAY": 0.9876602395838529, "SIZE_BLEND_DECAY": 0.8485560490807148, "TOX_BLEND_DECAY": 0.25206870894615263, "ACT_BLEND_DECAY": 0.9846613679320917, "PHAT_ALPHA": 0.24654130708889355, "PHAT_ALPHA_RETAIL": 0.04988137148761059, "PHAT_ALPHA_SAME": 0.15024250265598568, "DIR_IMPACT_MULT": 5.0, "GATE_SIGMA_MULT": 5.310353680066322, "MIN_GATE": 0.028755876837958636, "TRADE_TOX_BOOST": 0.30012867444243985, "STALE_ATTRACT_FRAC": 0.7439821261014115, "RET_CAP": 0.020759482946263577, "TOX_CAP": 0.18132214037231562, "TRADE_RATIO_CAP": 0.20845030089129385, "LAMBDA_CAP": 1.3535759345991327, "STEP_COUNT_CAP": 57.0, "BASE_FEE": 0.000508134101723667, "SIGMA_COEF": 0.1877109467945157, "LAMBDA_COEF": 0.001348542881062129, "FLOW_SIZE_COEF": 0.48188023382045203, "TOX_COEF": 0.02392218579226781, "TOX_QUAD_COEF": 1.367402758626434, "ACT_COEF": 8.29995377761181, "DIR_COEF": 0.003296913007577095, "DIR_TOX_COEF": 0.010684364473509862, "STALE_DIR_COEF": 0.7472654350208648, "SIGMA_TOX_COEF": 0.044600579162825414, "TAIL_KNEE": 0.05025329071701485, "TAIL_SLOPE_PROTECT": 0.8932855457143085, "TAIL_SLOPE_ATTRACT": 0.8643887380828649, "ANCHOR_COUNT_CAP": 566.0, "ANCHOR_LINEAR_K": 0.7593272536437132, "ANCHOR_QUADRATIC_K": 1.2557522951692819 }, "scaling": { "feature_multipliers": [ 1000.0, 2.0, 500.0, 500.0, 500.0, 1.0, 500.0, 500.0, 2.0, 2.0, 1000000.0, 1.0, 500.0, 647.054817, 1.395287, 220.617911, 228.148131, 0.412565, 274.335453 ], "feature_offsets": [ -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, 0.0, 0.0, -1.0, -1.0, -1.0, 0.0, -1.0, -0.618293, -2.037262, -0.803057, -0.808615, -0.823811, 0.001772 ], "note": "unchanged from IA/J5" }, "weight_fixed_point_scale": 1000000 } }, "real_only": [ { "key": "k21", "seeds": [ "8380064", "8381023" ], "n": "896", "mean": "0.5670358479237297", "se": "0.042741198260174", "sum": "508.06411973966186", "sumsq": "1753.0460671637363", "seeds_note": "k21 common paired seeds are 8380064..8380159 and 8380224..8381023 (896); chunks 8380000..8380063 (IB) and 8380160..8380223 (J5) lost" }, { "key": "k22", "seeds": [ "8384000", "8385023" ], "n": "1024", "mean": "0.5797335654448794", "se": "0.02547602111344616", "sum": "593.6471710155565", "sumsq": "1024.0474057389672" }, { "key": "k23", "seeds": [ "8388000", "8389023" ], "n": "1024", "mean": "0.5613316599994707", "se": "0.02910621275025325", "sum": "574.803619839458", "sumsq": "1210.1117956669596" }, { "key": "k25", "seeds": [ "8396000", "8397023" ], "n": "1024", "mean": "0.5179463363105676", "se": "0.027565766401464735", "sum": "530.3770483820213", "sumsq": "1070.7117348613783" }, { "key": "k26", "seeds": [ "8400000", "8401023" ], "n": "1024", "mean": "0.5932222621077041", "se": "0.03479245287991122", "sum": "607.459596398289", "sumsq": "1628.4357320102333" }, { "key": "k27", "seeds": [ "8404000", "8405023" ], "n": "1024", "mean": "0.52586537633738", "se": "0.027676373572693678", "sum": "538.4861453694771", "sumsq": "1085.5768332399236" }, { "key": "k28", "seeds": [ "8408000", "8409023" ], "n": "1024", "mean": "0.5337724667273851", "se": "0.025523597267120723", "sum": "546.5830059288423", "sumsq": "974.1829182380606" }, { "key": "k29", "seeds": [ "8412000", "8413023" ], "n": "1024", "mean": "0.5648229548848243", "se": "0.026427997297294538", "sum": "578.37870580206", "sumsq": "1058.3327840839527" }, { "key": "k30", "seeds": [ "8416000", "8417023" ], "n": "1024", "mean": "0.5232863649069781", "se": "0.026210348229595132", "sum": "535.8452376647456", "sumsq": "1000.0502457994058" } ], "real_only_pooled": { "all_9_items": { "n": "9088", "mean": "0.5516774482988679", "se": "0.009866094440459106", "sum": "5013.6446501401115", "sumsq": "10804.495516802617" }, "valid_7_items_k23_k25_to_k30": { "n": "7168", "mean": "0.5457496316106157", "se": "0.010709787078684052", "sum": "3911.9333593848933", "sumsq": "8027.402043899915" }, "note": "J5 (unchanged) minus IB, one paired seed per row; item seed ranges are disjoint. se = sqrt((sumsq - n*mean^2)/(n-1)/n) computed by this merge from the items' sum/sumsq." } } ```