# Root merge 3: RS2 confirmation, mechanism searches, and public-fork mining

Keep RS2 for these experiments. Two separate 2,000-seed confirmations agree on a small RS2 gain over RS1 and V10. Widened RS2 searches, side-specific retail state, posterior-prior modifications, and all tested public-fork finalists provide no robust replacement. **None of these 25 inputs supplies the required fresh paired candidate-versus-STACK confirmation, so this root does not establish a top-10 improvement.** [b24-a; b24-b; b25-a–f; b26-a–b; b27-a–b; b28-f1–f7; b28-rest; b28-h1–h4]

Goal context, supplied by the requester rather than measured here: STACK board 519.71 (29th), 10th 524.66, first 528.40; target about +5.0 paired over STACK. The last local-to-board transfer was 0.65. STACK is the code field of https://api.imd.fun/artifacts/c33720874aa1285e9dc766540b8a90410118a8e8b840d949ea0dcc211bdb27c1, SHA-256 `2eaafc7960e520b8a7fd322ec8afaff8fa677c730228a961efb509cc30b25719`; requester reports 0 OOG and peak 214k against the 250,000 afterSwap limit. The prior-goal findings supplied in the brief were hundredths from fee-family tuning, a cap near +1.2, and losses of 18 to 83 for scratch rules, mostly from arbitrage. Those are context, not results independently rechecked by this merge. [requester brief]

## Evidence, attribution, and limits

All25 JSON inputs were present and parsed. Citations identify input/field; the index gives canonical artifact paths (transport digests omitted). This merge ran no simulation, calibration, leaderboard query or source reread. Tables show at most eight significant digits; selected prose figures are exact. Measurements remain worker reports, not independently certified. The brief flags failed source rereads without identifying URLs/passages. These weaken external mechanism/provenance/author/licence/coverage claims; they neither independently verify nor automatically negate embedded code and local measurements. [requester source-warning list]

Index paths: prefix `artifacts/goal/batches/`, suffix `/result.json`; inputs `.imd/reads/artifacts/<ID>`. “None listed” is not independent verification.

| ID | Path between prefix/suffix | Failed sources reported |

| --- | --- | --- | --- |

| b23-tail-b | e05-design-impl/items/tail-b | 1 |

| b24-a | e06-confirm-rs2/items/a | 3 |

| b24-b | e06-confirm-rs2/items/b | 2 |

| b25-a | e06-rs2-wide/items/a | 2 |

| b25-b | e06-rs2-wide/items/b | 2 |

| b25-c | e06-rs2-wide/items/c | 2 |

| b25-d | e06-rs2-wide/items/d | 2 |

| b25-e | e06-rs2-wide/items/e | 2 |

| b25-f | e06-rs2-wide/items/f | 2 |

| b26-a | e06-rs-side/items/a | None listed |

| b26-b | e06-rs-side/items/b | 2 |

| b27-a | e06-rs-post/items/a | 1 |

| b27-b | e06-rs-post/items/b | 2 |

| b28-f1 | e07-fork-mine/items/f1 | 2 |

| b28-f2 | e07-fork-mine/items/f2 | 2 |

| b28-f3 | e07-fork-mine/items/f3 | 2 |

| b28-f4 | e07-fork-mine/items/f4 | 10 |

| b28-f5 | e07-fork-mine/items/f5 | 2 |

| b28-f6 | e07-fork-mine/items/f6 | 3 |

| b28-f7 | e07-fork-mine/items/f7 | 2 |

| b28-h1 | e07-fork-mine/items/h1 | 1 |

| b28-h2 | e07-fork-mine/items/h2 | 1 |

| b28-h3 | e07-fork-mine/items/h3 | None listed |

| b28-h4 | e07-fork-mine/items/h4 | 3 |

| b28-rest | e07-fork-mine/items/rest | 7 |



Raw counters, population/per-file rows, paired arrays and repeated code are consolidated into coverage, findings, finalist metrics and exceptions. Exact RS2 code is retained losslessly; other selected code has hashes/parameters or immutable URLs. Missing measurements and inaccessible repositories remain unknown. [all indexed inputs]

## Common protocol and reference identities

Workers generally report the pinned upstream `6b6921f8d463314aa7c96234e1be25a25245f704`, checksum-verified wheel `aedb68280aca34c110968f00148387c95885207913fb2cf19a1ab4c5cfd6887a`, solc 0.8.24 with optimizer 200/viaIR/paris, and each arm alone against VanillaStrategy with `amm_sim_rs.run_batch(...,8)`. Configs draw size U(19,21), arrival rate U(.6,1), sigma U(.000882,.001008) in that order from default_rng(seed), price seed=seed, retail seed=seed+1, 10,000 steps, price 100 and reserves 100/10000. Paired differences mean candidate minus named reference; SE is sample SD/sqrt(n). The V10 harness reconstructs negative optimal arbitrage profit from preceding reserves/fees and current fair price; retail is total minus arb. Arithmetic reconciliation is not independent verification of trader attribution. Source: https://api.imd.fun/artifacts/33f54e81f3372088e9eced52c0174f310ac348af98ef5727b73785c088679194; pinned repository https://github.com/benedictbrady/amm-challenge/tree/6b6921f8d463314aa7c96234e1be25a25245f704. Exceptions and corrections are retained below. [b23-tail-b.notes.protocol; b24-a.notes; b24-b.notes; b28-f3.protocol]

RS2 SHA-256 `f24b6ca198b744c380842c3a17964a6dc21f311bb04984b26733afa3fee8eaf2` comes from https://api.imd.fun/artifacts/19f2ec857df536e8ab62ab1da89ec6623332521ce46849f15bccff402c63e1e5 and is embedded identically in b24-a/b. RS1 SHA-256 is `df2817ba4ec233a456eb45909faa4a706efe0fabd1734a908fe52545f4d07577`; V10 is `cd9bdb45ab0537ad25c4fd90ca8618898cef6386de3df65f192eaef65fbca245`. CH is `6dd68f77a70900319460215374143dbd5d924198450a46d9f5f91f6b1e3d35c3` at https://raw.githubusercontent.com/ChengHua926/amm-challenge/d34a5199c59a4c44d5336a25f1436b4ae1f21104/contracts/src/Strategy.sol. [b24-a.reference_sha256; b24-b.reference_sha256; b28-f3.baseline]

CH calibration is reported as exactly 525.5258821780834 on seeds 0..999 by b23-tail-b, b24-a/b, b25-a/c/d/e/f, b26-b, b27-a/b, b28-f1–f7, b28-rest, b28-h1 and b28-h4. b25-b explicitly did not run it; b28-h2 explicitly did not run it; b26-a and b28-h3 do not supply that calibration result. Do not infer a successful calibration for those omissions. The calibration mean uses its own block and is not a substitute for paired gains. [respective calibration/checks/notes fields]

Any OOG invalidates a score despite CLI success/low successful-call peaks. Separate init/swap gas. RS2 searches also reject afterSwap>=235000 or init>=245000; mining uses challenge limits, so some observed-safe finalists exceed search limits. Unrun/null gas is unknown. [b25-a.notes; b26-b.search; b28-h1.notes; b28-f6.notes]

## Exact RS2 confirmations

| Source | Seeds inclusive | Reference | n | Paired mean (SE) | Retail delta | Arb delta | RS2 afterSwap max | OOG |

| --- | --- | --- | --- | --- | --- | --- | --- | --- |

| b24-a | 5104000..5105999 | RS1 | 2000 | 0.091529804 (SE 0.011537927) | 1.4442254 | -1.3526956 | 203671 | 0 |

| b24-a | 5104000..5105999 | V10 | 2000 | 0.16386823 (SE 0.012105865) | 1.471244 | -1.3073758 | 203671 | 0 |

| b24-b | 5106000..5107999 | RS1 | 2000 | 0.087808358 (SE 0.011443276) | 1.4584208 | -1.3706124 | 203272 | 0 |

| b24-b | 5106000..5107999 | V10 | 2000 | 0.16274676 (SE 0.012476521) | 1.486081 | -1.3233342 | 203272 | 0 |



[b24-a] Worst five versus RS1: 5105662: -2.424356930677959; 5105783: -2.040894194677435; 5104654: -1.5389455101950489; 5105159: -1.508090239829471; 5105084: -1.412281740853473. Worst five versus V10: 5105662: -2.326220565376957; 5104778: -2.102928235581885; 5105109: -1.5825142067979527; 5104205: -1.5809919588805315; 5104483: -1.532926355229506. Arm mean edges: {"rs2": 521.8534158497197, "rs1": 521.7618860455824, "v10": 521.6895476203318}. RS2 init peak 243099; RS1/V10 afterSwap peaks 202468/194858; all reported OOG/revert/other-halt/tx-error/bad-return counts zero.

[b24-b] Worst five versus RS1: 5107860: -3.7311009975261413; 5107968: -2.1722074247310275; 5107108: -1.9298650706765557; 5107520: -1.9162149838875848; 5106118: -1.6579389821928316. Worst five versus V10: 5107860: -3.5657057114934787; 5107586: -2.5445488561105094; 5107520: -2.3940747560666296; 5107968: -2.200554170558462; 5107108: -1.7365651694613007. Arm mean edges: {"rs2": 525.748038937657, "rs1": 525.660230579923, "v10": 525.5852921821429}. RS2 init peak 243099; RS1/V10 afterSwap peaks 203285/195360; all reported OOG/revert/other-halt/tx-error/bad-return counts zero.

Both report exact-source validation, no tuning/dropped seeds/rerun, and per-chunk counters. RS2 wins versus RS1/V10:1122/1222 of2000 (a),1115/1220 (b). Absolute means vary by block; retail gains mostly offset arb losses. These >=5,000,000-seed confirmations are **not versus STACK**. [b24-a.vs_rs1/vs_v10/notes; b24-b.vs_rs1/vs_v10/notes]

## Size-tail design implementation: identity wins

b23-tail-b implements screen-fitted residual fee tables on V10. Onchain inputs are causal TradeInfo and existing state: pEst=0.7*pHat+0.3*fastHat; fair Y depth=sqrt(reserveX*reserveY*pEst), bilinear lookup in depth and signed spot/pEst. S1 marginalizes a three-node normalizer-price prior; S1-lite sets that deviation to zero; S3 interpolates nonnegative whole-bps premiums in actual fee. Delta is bounded before original compression/caps, with no extra storage. It does not establish a successful new mechanism. The upstream design URL https://api.imd.fun/artifacts/f0c84ebe66608025bb038a242030d6639f81ec87bf35146ee4495f572c7b74e7 is a provenance lead subject to the one failed-source warning. [b23-tail-b.offline_source/design_url/notes]

Fit used V10 trajectories on 790000..790255, every 25th post-step observation; config fitting used 790000..793999 only. Fitted M=20.013303911184643, arrival=.7980330324108584, sigma=.0009448667904304872, opposite fee=.0027878161785832986; three normalizer log-price nodes (-.00295082152799499, .00000036131900447256307, .002970888183358485) with weights (.25,.5,.25). Grid depth 7000..13000 by 1000; spot offset -150..150 bps by 5; fee grid 0,30,60,90,120,150 bps; 141 quadrature points over [-7,7], 1-bps argmax grid 0..150. No recheck/held-out data fitted the tables. [b23-tail-b.offline_source.fitted/input_recipe]

All screen rows below are n=256, seeds 790000..790255, differences versus V10, OOG=0. This retains every named screen arm; zero-overlay F2/F5 controls are identities, not evidence that routing/shuffling never matter. [b23-tail-b.screen]

| Arm | Mean | SE | Retail delta | Arb delta | afterSwap max |

| --- | --- | --- | --- | --- | --- |

| identity | 0.0 | 0.0 | 0.0 | 0.0 | 192874 |

| identity_45 | -0.013828356 | 0.0091437241 | -0.014895109 | 0.001066753 | 192752 |

| s1_extra_k0.5_d2 | -0.65701771 | 0.04445699 | -2.0290774 | 1.3720597 | 202754 |

| s1_extra_k1_d20 | -10.477919 | 0.13093262 | -15.550973 | 5.073054 | 202732 |

| s1_extra_k1.5_d10 | -8.6246759 | 0.12804628 | -13.807105 | 5.1824293 | 202754 |

| s1_l0_k0.5_d5 | -1.6133869 | 0.0666465 | -4.0002743 | 2.3868875 | 206151 |

| s1_l0_k1_d10 | -5.0060496 | 0.098702809 | -9.1759974 | 4.1699478 | 203268 |

| s1_l0.5_k0.5_d5 | -2.8403984 | 0.077163669 | -5.7327182 | 2.8923198 | 202754 |

| s1_l0.5_k1_d10 | -8.1068767 | 0.11649162 | -12.99135 | 4.8844735 | 202754 |

| s1_l1_k0.5_d5 | -5.1768175 | 0.092812406 | -8.4916551 | 3.3148376 | 202715 |

| s1_l1_k1_d10 | -12.741006 | 0.15199231 | -18.374137 | 5.6331307 | 202715 |

| lite_l0 | -0.12912197 | 0.027247384 | -0.8011703 | 0.67204833 | 203347 |

| lite_l0.5 | -0.46775468 | 0.035065729 | -1.3547584 | 0.88700376 | 206179 |

| lite_l1 | -3.1011622 | 0.059752773 | -4.6074519 | 1.5062897 | 202671 |

| s3_k0.25 | -10.572204 | 0.1382288 | -17.180656 | 6.6084522 | 211047 |

| s3_k0.5 | -12.229236 | 0.16063852 | -20.392388 | 8.1631526 | 211085 |

| s3_k1 | -12.229236 | 0.16063852 | -20.392388 | 8.1631526 | 211085 |

| s1_45_l0 | -1.6682311 | 0.067237514 | -4.0606212 | 2.3923901 | 206151 |

| s1_45_l0.5 | -2.8983273 | 0.078869557 | -5.7953913 | 2.897064 | 202754 |

| s1_45_l1 | -5.2033105 | 0.093246808 | -8.5216689 | 3.3183583 | 202715 |

| F2_center | -3.1789718 | 0.074797149 | 2.1980504 | -5.3770222 | 203432 |

| F5_center | -2.6777592 | 0.066119607 | -5.5467296 | 2.8689703 | 202811 |

| frozen_F2 | 0.0 | 0.0 | 0.0 | 0.0 | 192874 |

| frozen_F5 | 0.0 | 0.0 | 0.0 | 0.0 | 192874 |

| probe_F2 | -3.3477909 | 0.070613406 | 1.9480201 | -5.295811 | 203436 |

| probe_F5 | -0.15808699 | 0.024157696 | -0.73333405 | 0.57524706 | 202698 |



Recheck 794000..794255: identity 0 (SE 0), identity_45 -0.0035282948228152478 (SE 0.008421570020916714), active diagnostic lite_l0 -0.18845794091692447 (SE 0.026315392183471772). Identity freezes with kappa=0, DMAX=5 bps, lambda=.5, V10 30/36-bps caps; source hash `c4a9943e43670402fb004e49a5f12c491f55980e99b0ce0b0295ced7329a68c3`. Disabled parity: n=32 on 796000..796031, max absolute edge difference 0. On held-out 795000..795511, identity and its matched unclipped/shuffled zero-strength controls each give n=512, mean=SE=retail=arb=0; all mutual differences zero. Frozen gas aggregated across scored blocks max_ok=196513, init=220972, OOG=0; held-out identity peak=195637. [b23-tail-b.parity/re_check/held_out/gas/params]

Separate prespecified active held-out diagnostics, same 512 seeds, all OOG=0: [b23-tail-b.notes.identity_won_nonzero_probe.held_out]

| Arm | Mean | SE | Retail | Arb | afterSwap max |

| --- | --- | --- | --- | --- | --- |

| lite_l0 | -0.16149063 | 0.020287586 | -0.84679295 | 0.68530232 | 206258 |

| probe_F2 | -3.2198622 | 0.049286058 | 2.0452374 | -5.2650996 | 205534 |

| probe_F5 | -0.20555213 | 0.016303774 | -0.78156563 | 0.5760135 | 206216 |



lite_l0 minus unclipped F2: +3.058371586971366 (SE .057327892442833456), retail -2.892030329059266 / arb +5.950401916030634; minus shuffled F5: +.04406150330252323 (SE .017641526569757274), retail -.06522731652216418 / arb +.10928881982468801. Thus routing clipping matters for this losing active overlay, while the identity’s controls are trivial. F1 active diagnostic upper mean+2SE=-.1209154537955518, below +.2; strict S1 condition was not tested (diagnostic is S1-lite). F3 lambda=0 versus 1 and F4 lite versus strict are the matching screen rows above. F5 uses one fixed seed-790000 permutation of spot rows, common across sides/depth. F6 is a static uniform 854-point grid, not occupancy-weighted trajectories: no reported finite correlation was classified collapsed; correlations include lite_l0 .3300311392952816, lite_l0.5 -.24703026587702126, lite_l1 -.8866054673543716; identity and saturated S3 cases have null correlations. [b23-tail-b.falsifiers_run; notes.identity_won_nonzero_probe.frozen_vs_ablations]

Numerical checks: 18 point-mass cases and mirror test had max error 0; 561-point quadrature checks differ from 141-point residuals by at most 1 bp. Decomposition diagnostic had 6 zero-retail paths among 32 attempted, 100 steps, configured retail rate 0 but effective floor .01, max residual retail 1.8826912251412864e-11. Across 41 reported scored runs / 12584 seed evaluations, max afterSwap 211085, init 220981, OOG=0, reported reconciliation max 0. Limitations: approximate offline normalizer prior, fair value/opposite fee/next-arb approximations, omitted future retail reversals, clipped domain. No STACK confirmation or >=5,000,000 seeds. [b23-tail-b.notes.checks/limitations]

## Widened RS2 family search

RS2 parameter order is `[structure,mix,retention_per_step,rate_scale,size_scale,quote_correction_slope,correction_cap]`. Baseline exact values: `[2,0.3424067991934183,0.999,0.36638342491340714,1.8832961581192982,196.03088039613942,1.5]`. Structure stays 2; correction missing fraction is capped at 1-1/cap. CMA-ES searches six normalized coordinates with population 10/mu 5, warm start RS2 and bounds mix [0,.9], retention [.9,.9999], rate [.02,3], size [.3,4], slope [5,800], cap [1.05,8]. Fitness is paired mean vs V10 on a new 64-seed CRN block per generation. Top three distinct valid hashes are normally rechecked on 256 fresh seeds and frozen before 512 held-out seeds. A zero-mix parity arm matches V10; rebuilding baseline parameters matches RS2. [b25-a–f.params_exact/parameter_names/checks/notes]

| Source | RNG / sigma0 / boundary | Evaluated search | Stop/exception | Recheck |

| --- | --- | --- | --- | --- |

| b25-a | 21 / .15 / clipped | 177 incl. warm start; 800000..801151 | 1504.665 s; partial final generation | 804000..804255 |

| b25-b | 22 / .3 / clipped | 7 of generation 0; 810000..810063 | incomplete; no CMA update or recheck | not run |

| b25-c | 23 / .15 / clipped | 44 incl. warm start; 820000..820319 | 1519.4997 s; gas-parser summaries rebuilt | 824000..824255 |

| b25-d | 24 / .3 / reflected | 69 search; 830000..830447 | 1508 s; 72 log rows include 3 rechecks | 834000..834255 |

| b25-e | 25 / .2 / clipped | 30 / 3 generations; 840000..840191 | 1200 s; fourth generation not started | 844000..844255 |

| b25-f | 26 / .2 / clipped | 61 incl. warm start; 850000..850383 | 1703.478 s; 203.478 s beyond 1500 cap | 854000..854255 |



b25-b throughput forced only 7 of 10 sampled points (c4/c6/c9 unrun), no 814000 recheck, held-out n=32 rather than 512, and no CH calibration. Its c0 zero-mix equals V10; chosen c3 screen +.00029910546910283387 (SE about .031) is not a demonstrated gain. b25-d’s normalized warm-start round trip changed cap 1.5 to 1.5000000000000004, so that warm-start source is not the exact RS2 hash; the held-out RS2 reference was exact. b25-c corrected a lowercase gas-role filter from retained raw counters without changing selection; b25-f’s deadline-overrun is a protocol departure. These limits are part of the evidence. [b25-b.notes/not_evaluated/checks; b25-d.notes; b25-c.notes; b25-f.notes]

Frozen SHA-256 and exact params:

- [b25-a, sha256/params_exact]: `405b3af10af783a4a0e5378e6c7927946c6b9a1d4adf06ddcf1e454b4f60fab9`; `[2,0.5408435104682148,0.9999,0.45495685870999614,2.074880016546598,5.0,3.2748894451493165]`.

- [b25-b, sha256/params_exact]: `524be30b0303cdd51112b4986ffbde8954e623b28e0e62cd01f56f50917b9630`; `[2,0.030907100771999835,0.9881078915160141,0.5389207064566415,0.42575451862327296,390.66735894363785,1.2851429610053908]`.

- [b25-c, sha256/params_exact]: `34b838f2760df9f536f973d67bc7eb9955c56628f46a275ee3987b1341111908`; `[2,0.3010214851295514,0.9999,0.23502976831687478,1.6864329600505317,122.47567345608427,1.5933918067495974]`.

- [b25-d, sha256/params_exact]: `e7ca0c8128743f52e248323b3c1e2d0fd95395aadc4af36569cdcb1daf634932`; `[2,0.3688143491783866,0.993265062498444,0.0242139147572416,3.113400875246335,408.1358805737961,1.3104550727073652]`.

- [b25-e, sha256/params_exact]: `f31f75ce818f10a4402b1a598898fa1acdd498476636336c10b00cf9310b76cc`; `[2,0.313332340948775,0.9999,0.22747405228287973,1.7766714797636054,295.2586309440775,3.095148788646452]`.

- [b25-f, sha256/params_exact]: `56e5b6c8b754d65bea44fbb2e0aebd5dc140eb5b0ae9be58b2788fb300a20f9f`; `[2,0.46358468880646375,0.9978620110336661,0.3412352256177734,1.8847786594195202,292.9874684209921,2.6866067195672363]`.



Recheck means (SE), versus V10, in reported candidate order; winner is the frozen hash above. [each source recheck]

b25-a: 0.2040746123591406 (SE 0.037048796351171816); 0.16932132414446888 (SE 0.02662865979891769); 0.20319127817696558 (SE 0.032006320957242554).

b25-c: 0.20099289982293267 (SE 0.03200287990354934); 0.19562927758176385 (SE 0.03387785048096625); 0.19074558374087625 (SE 0.035387026810613474).

b25-d: 0.24620007177033099 (SE 0.04277569347407655); 0.15228344196189592 (SE 0.03216818437351659); 0.1069879949479906 (SE 0.043860211862755336).

b25-e: 0.12458832190536806 (SE 0.03093713721969953); 0.1292119491356869 (SE 0.03164006084044076); 0.17171694914275415 (SE 0.02801908628926781).

b25-f: 0.1445961352664158 (SE 0.035626954193310315); 0.1381583731020859 (SE 0.022877675422115253); 0.16366874954061728 (SE 0.03689799641527635).

Held-out comparisons below keep references separate; retail/arb are mean deltas. n=512 except b25-b n=32. Gas columns are the frozen **held-out** arm, afterSwap/init; OOG=0 in every listed arm. [each source held_out]

| Source | Seeds | Reference | n | Paired mean (SE) | Retail | Arb | Gas swap/init |

| --- | --- | --- | --- | --- | --- | --- | --- |

| b25-a | 805000..805511 | V10 | 512 | 0.21123276 (SE 0.028825691) | 1.8550066 | -1.6437738 | 200175/243099 |

| b25-a | 805000..805511 | RS2 | 512 | -0.0041938508 (SE 0.014400302) | 0.34382384 | -0.34801769 | 200175/243099 |

| b25-b | 815000..815031 | V10 | 32 | -0.012713812 (SE 0.035763088) | 0.30156209 | -0.3142759 | 197298/243099 |

| b25-b | 815000..815031 | RS2 | 32 | -0.076507221 (SE 0.071481039) | -1.1171883 | 1.040681 | 197298/243099 |

| b25-c | 825000..825511 | V10 | 512 | 0.22209676 (SE 0.024343543) | 1.4370858 | -1.214989 | 202468/243099 |

| b25-c | 825000..825511 | RS2 | 512 | 0.018924149 (SE 0.0093031179) | -0.076841605 | 0.095765754 | 202468/243099 |

| b25-d | 835000..835511 | V10 | 512 | 0.15940198 (SE 0.027117557) | 1.8515583 | -1.6921563 | 202391/243099 |

| b25-d | 835000..835511 | RS2 | 512 | -0.021094578 (SE 0.013215476) | 0.35399252 | -0.37508709 | 202391/243099 |

| b25-e | 845000..845511 | V10 | 512 | 0.18117069 (SE 0.023088116) | 1.1314178 | -0.95024713 | 200257/243099 |

| b25-e | 845000..845511 | RS2 | 512 | 0.010132729 (SE 0.016178456) | -0.37559953 | 0.38573226 | 200257/243099 |

| b25-f | 855000..855511 | V10 | 512 | 0.14585443 (SE 0.025168558) | 1.7094545 | -1.5636 | 200268/243099 |

| b25-f | 855000..855511 | RS2 | 512 | -0.019677955 (SE 0.01038249) | 0.22305734 | -0.2427353 | 200268/243099 |



Worst individual held-out candidate-minus-RS2 seeds (the five-lowest lists in the inputs are compressed here to their minimum):

- [b25-a, held_out.vs_rs2.worst]: 805441: -2.1096102161379804.

- [b25-b, held_out.vs_rs2.worst]: 815004: -0.9030639916093719.

- [b25-c, held_out.vs_rs2.worst]: 825458: -1.807061013603402.

- [b25-d, held_out.vs_rs2.worst]: 835143: -0.9585338441095814.

- [b25-e, held_out.vs_rs2.worst]: 845089: -4.795487757091053.

- [b25-f, held_out.vs_rs2.worst]: 855108: -1.0323457816509745.



a/c/e favor retention .9999, sometimes lower rate_scale; slope/cap vary. c’s small +.018924148805667063 over RS2 is isolated after search selection; a/d/f negative, e within SE, b incomplete. a/d/f retail gains lose to arb; c/e reverse this tradeoff. b25-e absolute528.6692481059199 is only a block mean. Gas-summary scopes differ: a 202100, b 202468, c 202468, d held-out 202391 (search 202425), e held-out 200257 (search 203047), f aggregate 202456; init 243099, no scored OOG. [b25-a–f.notes/gas/held_out/checks]

## Side-specific state and cleaned-rate posterior odds

b26-a uses lazy slot 16 for a decayed posterior-weighted buy/total count, clamps share to [1/(1+cap),cap/(1+cap)], and moves each direction’s lambda toward 2*share*retailRate using side_mix. b26-b packs a share into spare high bits of slot 2, treats trader buys as trade.isBuy=false/ask side, uses an EMA share rather than a count ratio, and targets min(2*share,cap)*already-mixed-lambda. Negative side_mix moves away from the target. Both null side_mix cases return RS2 exact code; their definitions of cap and rate differ, so agreement on lack of gain does not isolate a single side-state design. [b26-a.mechanism/params_note; b26-b.mechanism]

b26-a is severely scaled down: 30 CMA points, RNG31, 3 generations of five-seed blocks 860000..860014, eight-seed recheck 864000..864007, 24-seed held-out 865000..865023. It cannot support the planned 64/256/512-seed protocol; small side_mix .0764474 wins a noisy recheck +.13968589078129412 (SE .16712309101841435). b26-b uses RNG32, 80 points / 8 generations on 870000..870511 in 1446.9 s, recheck 874000..874255: +.17098558474155978 (.030330244245658552), +.16711731107976102 (.028580349508557326), +.13928596568431573 (.02945443343071398). Side dimensions were flat, good signs mixed; cap >2 is inactive; lower ordinary mix explains much of the selected movement. [b26-a.log/recheck/notes; b26-b.search/recheck/notes]

b27-a/b fall back to multiplying first-print arb odds because V10 has no explicit activity-rate prior. a uses updated lambdaHat activity / max(previous cleaned_rate*rate_scale,.001), ratio clamp [.05,20], gamma steps 1/16 via four square roots. b uses lagged slot7 activity / raw lagged cleaned_rate before rate_scale, activity floor .01, ratio clamp [.01,100], gamma quarters via two square roots. Both scale odds by prior_scale*ratio^gamma and also search ordinary mix/rate_scale; gamma=0/prior_scale=1 reproduces RS2 with baseline parameters. Their different denominators, timing and quantization mean these searches do not estimate the same intervention. [b27-a.mechanism; b27-b.mechanism]

b27-a RNG41 scored 36 evaluations on new 64-seed blocks 880064..880319; two preliminary rows on 880000..880063 were discarded for wrong gas-role parsing, other interrupted/unrecorded attempts excluded, old deadline retained. Recheck 884000..884255 means: +.07378480781543661 (.03331698675363763), +.07090049372952589 (.033690338546729925), +.07190823496136955 (.032283060179178855). b27-b RNG42 scored 23 points over 890000..890191; an aborted initial block was replayed unchanged after the same case-sensitive parser repair. Recheck 894000..894255: +.18545525775614968 (.033267437801279776), +.1254297966666451 (.033248223102395585), +.05286564492101875 (.0392431454594157). Discarded attempts with unknown gas are not certified zero-OOG. [b27-a.log/notes.search/repair/recheck; b27-b.log/notes.pre_search/recheck]

[b26-a, params/sha256]: `daf3834688116f4a6bfacc13ee94cd1e927ed5522440c15125c1f958a585e149`; frozen params `{"side_mix":0.0764474,"side_retention":0.953597,"side_cap":1.89725,"mix":0.463539,"size_scale":2.67381}`.

[b26-b, params/sha256]: `01c941bf542ed9e279009e509299295010e4e9e24f78c852db1db50c2284f76a`; frozen params `{"side_mix":-0.2391621520722873,"side_retention":0.9999,"side_cap":2.257346805976611,"mix":0.24185994593719717,"size_scale":1.837966741671827,"other_rs2_params_exact":{"structure":2,"retention_per_step":0.999,"rate_scale":0.36638342491340714,"quote_correction_slope":196.03088039613942,"correction_cap":1.5}}`.

[b27-a, params/sha256]: `e1d5770ed1f046b8eaaceb1957fb7c129b3c0d9ca1d6b0dc8f00ef8ed2f9071e`; frozen params `{"gamma":-0.4375,"prior_scale":1.1959923350130397,"mix":0.16733783642120584,"rate_scale":0.314593414073878,"rs2_params_exact":[2,0.3424067991934183,0.999,0.36638342491340714,1.8832961581192982,196.03088039613942,1.5],"rs2_parameter_order":["kind","mix","decay","rate_scale","size_scale","correction","cap"]}`.

[b27-b, params/sha256]: `39a911c2555cba0e91f1ba83476c41d232ae00933ff688f6e0407e4249ffdd1f`; frozen params `{"gamma":-0.25,"prior_scale":0.7875718867071857,"mix":0.29347467449732445,"rate_scale":0.35907403290845774,"rs2_baseline_params_exact":[2,0.3424067991934183,0.999,0.36638342491340714,1.8832961581192982,196.03088039613942,1.5],"effective_rs2_params_exact":[2,0.29347467449732445,0.999,0.35907403290845774,1.8832961581192982,196.03088039613942,1.5],"gamma_quantum":0.25}`.

Held-out mean differences, with the original worker’s gas summary:

| Source | Seeds | Ref | n | Paired mean (SE) | Retail | Arb | Reported max/init/OOG |

| --- | --- | --- | --- | --- | --- | --- | --- |

| b26-a | 865000..865023 | V10 | 24 | 0.063382395 (SE 0.1139589) | 1.3832292 | -1.3198468 | 221504/243093/0 |

| b26-a | 865000..865023 | RS2 | 24 | -0.04561139 (SE 0.064599813) | -0.066486325 | 0.020874935 | 221504/243093/0 |

| b26-b | 875000..875511 | V10 | 512 | 0.14463839 (SE 0.019502819) | 0.92652127 | -0.78188288 | 205127/243093/0 |

| b26-b | 875000..875511 | RS2 | 512 | -0.025352708 (SE 0.013732127) | -0.58607186 | 0.56071915 | 205127/243093/0 |

| b27-a | 885000..885511 | V10 | 512 | 0.09144632 (SE 0.022462056) | 1.10792 | -1.0164737 | 206156/243099/0 |

| b27-a | 885000..885511 | RS2 | 512 | -0.10186244 (SE 0.022746843) | -0.417138 | 0.31527555 | 206156/243099/0 |

| b27-b | 895000..895511 | V10 | 512 | 0.17969227 (SE 0.025265043) | 1.5270906 | -1.3473983 | 206066/243099/0 |

| b27-b | 895000..895511 | RS2 | 512 | -0.010763191 (SE 0.014982506) | -0.0028066976 | -0.0079564939 | 206066/243099/0 |



No held-out side/prior point estimate beats RS2; b27-a’s loss exceeds its reported SE substantially. b26-a’s extra first-write slot causes a held-out 221504 peak (search max 206541); b26-b aggregate 205127 includes recheck, held-out frozen 202931. b27-a frozen 206156 differs from all-search max 207711; b27-b 206066 summary covers corrected search/rechecks/all held-out arms. These scopes explain, rather than erase, gas differences. All scored arms in these summaries report no halts. Worst versus RS2: b26-a seed865015 -.6306465910088264; b26-b seed875405 -2.235067679955023; b27-a seed885401 -4.034386355553579; b27-b seed895197 -1.3332915155755245. None used >=5,000,000 seeds or ran STACK confirmation. [respective held_out/gas/notes fields]

## Fork mining: coverage and validity

All fork screens share900000..900031 and shortlists901000..901255: duplicates do not increase n. RS2 means542.3651755555325/533.7351592392916 (f3 prints533.7351592392915), gas197184/202386, OOG0. Inventories read branch heads, exclude upstream-identical bytes, deduplicate SHA-256 and apply recency caps with differing ties. Negative whole-strategy tests do not rule out component grafts. No STACK comparison. [b28-f1–f7; b28-h1–h4; b28-rest.checks]

Coverage from supplied files rows: “scored”/ranges include invalid diagnostic scores where retained; valid=true has differing CLI/runtime meanings. Cap exclusions are untested. [respective files arrays]

| Source | Fork | Rows | valid=true | Scored rows | OOG files | Screen d32 range |

| --- | --- | --- | --- | --- | --- | --- |

| b28-f1 | VictorQian03/amm-challenge | 50 | 50 | 50 | 48 | -142.7252 .. -55.2375 |

| b28-f1 | shl0k28/amm-challenge | 50 | 49 | 49 | 0 | -245.326 .. -2.8563 |

| b28-f1 | CleanPegasus/amm-challenge | 45 | 45 | 45 | 0 | -193.1198 .. -99.9697 |

| b28-f1 | crypto-yuvi/amm-challenge | 1 | 1 | 1 | 0 | -24.165 .. -24.165 |

| b28-f2 | jaoz/amm-challenge | 50 | 50 | 50 | 2 | -462.66622 .. -2.8743338 |

| b28-f2 | horacepan/amm-challenge | 4 | 4 | 4 | 0 | -68.470528 .. -6.558325 |

| b28-f2 | Leoeqs/amm-challenge | 2 | 2 | 2 | 0 | -205.22514 .. -158.44052 |

| b28-f2 | alitaslimi/amm-challenge | 1 | 0 | 0 | 0 | not scored |

| b28-f3 | EricForgy/amm-challenge | 1 | 0 | 0 | 0 | not scored |

| b28-f3 | wellimbharath/amm-challenge | 2 | 2 | 2 | 0 | -162.29046 .. -2.8743338 |

| b28-f4 | solosage1/amm-challenge | 50 | 50 | 50 | 0 | -519.93372 .. -2.2378625 |

| b28-f4 | The-CTra1n/amm-challenge | 1 | 1 | 1 | 0 | -4.9938733 .. -4.9938733 |

| b28-f4 | 0xpranav62/amm-challenge | 1 | 1 | 1 | 0 | -148.0876 .. -148.0876 |

| b28-f4 | uniyalabhishek/amm-challenge | 43 | 43 | 43 | 0 | -376.06624 .. -2.8032951 |

| b28-f5 | giorgoschionas/amm-challenge | 1 | 1 | 1 | 0 | -211.30951 .. -211.30951 |

| b28-f5 | piotrostr/amm-challenge | 5 | 5 | 5 | 0 | -184.64722 .. -57.760548 |

| b28-f5 | atvinicius/amm-challenge | 1 | 1 | 1 | 0 | -7.8071925 .. -7.8071925 |

| b28-f5 | alexanderzliu/amm-challenge | 1 | 1 | 1 | 0 | -38.21947 .. -38.21947 |

| b28-f6 | maxnorm/amm-challenge | 47 | 32 | 32 | 15 | -466.2506 .. -2.8725105 |

| b28-f6 | ccashwell/amm-challenge | 1 | 0 | 0 | 0 | not scored |

| b28-f6 | enriquedlh97/amm-challenge | 31 | 31 | 31 | 0 | -386.28077 .. -123.80685 |

| b28-f6 | kmadorin/amm-challenge | 1 | 1 | 1 | 0 | -189.78043 .. -189.78043 |

| b28-f7 | 0zKillua/amm-challenge | 143 | 49 | 49 | 0 | -409.1628 .. -38.7932 |

| b28-f7 | d2p-finance/d2p-fee-strategies | 1 | 1 | 1 | 0 | -189.4645 .. -189.4645 |

| b28-f7 | ChengHua926/amm-challenge | 18 | 18 | 18 | 0 | -388.0496 .. -1.9469 |

| b28-h1 | jaoz | 80 | 80 | 80 | 0 | -252.680217 .. -151.472688 |

| b28-h1 | VictorQian03 | 80 | 7 | 80 | 73 | -292.157628 .. -53.324947 |

| b28-h2 | solosage1/amm-challenge | 80 | 77 | 77 | 0 | -349.6845 .. -2.2379 |

| b28-h3 | giorgoschionas | 1 | 1 | 1 | 0 | -211.3095 .. -211.3095 |

| b28-h3 | maxnorm | 47 | 47 | 47 | 15 | -500.752 .. -2.8725 |

| b28-h4 | 0zKillua/amm-challenge | 80 | 78 | 78 | 0 | -409.16282 .. -38.793216 |

| b28-h4 | ChengHua926/amm-challenge | 18 | 18 | 18 | 0 | -388.04956 .. -1.9468584 |



Coverage exceptions:

- b28-f1: distinct/retained Victor267/50, shl0k28 79/50, CleanPegasus45/45, crypto-yuvi1/1; branch counts3/1/2/6. 145/146 CLI-valid; Strategy_carry_premium fails stack depth. Victor48 OOG files/2,167,165 calls; best diagnostic -55.2375 invalid. shl0k28 best -2.8563, no OOG; CleanPegasus spike/bubble-sort/proportional-spike/vol-regime all lose; crypto FeeStrategy1 -24.165. [b28-f1.summary/per_fork/files]

- b28-f2: jaoz158/newest50, horacepan4, Leoeqs2, alitaslimi1; 56/57 CLI-valid. alitaslimi AdaptiveStrategy fails contract-name rule. Two jaoz OOG files (`b478276c0e819c2e88c1bbbd4b43471ee882be846efd7aa59d612f51a2a67cf1`: 378184 calls; `f87fb49ab6ceef279ab64d075e1a48992a802fab4e5065740bbbe37a64664512`: 247156) are excluded. [b28-f2.notes.coverage/execution_failures]

- b28-f3: sisyphusSmiling4 heads/muttoni2 have no changed afterSwap; pipeline code is Merkle-drop. EricForgy V2 starter fails name/inheritance/imports, unmodified/unrun. wellimbharath has two valid originals, hence only two finalists. [b28-f3.forks/files/notes]

- b28-f4: solosage155/newest50, uniyalabhishek43, The-CTra1n1,0xpranav62 1; all95 validate, OOG0. Ten failed sources weaken coverage claims. Finalists rerun after gas-role/initial-fee fixes; total unchanged. [b28-f4.fork_scan/files/notes]

- b28-f5: giorgoschionas 1, piotrostr 5, atvinicius 1, alexanderzliu 1; all 8 pass, no cap exclusion or OOG. Actual initial fees used for finalist attribution: fixed25, BoundedExp15, DecayVol3, ExpVol5, FastVol3, SqVol8, Rookie Edge2, DirContrarian24 bps, RS2 4. Nine-arm 100-step zero-retail diagnostic on 11000..11031 found 10 zero-retail paths per arm, max reconstruction errors up to 1.7601253787802307e-11. [b28-f5.fork_inventory/protocol/decomposition_check]

- b28-f6: maxnorm47/ccashwell1/enriquedlh97 31/kmadorin1, 80 distinct;79 CLI-valid,15 maxnorm OOG. ccashwell base failed validation, unrepaired; V34/V40/YQ tie uses inventory order. [b28-f6.forks/files/notes]

- b28-f7: 0zKillua143 eligible, 50 kept using descending version as recency proxy because one commit adds all files; 93 unscored. acemasterjb has no changed afterSwap; d2p-finance one; ChengHua92618. 69 retained, 68 validate; V30C stack-depth failure, zero OOG among scored arms. Text summary says every Cheng scratch/d2p loss is at least 140 and every 0zKillua loss at least 38; the full screen ranges above preserve worse tails. Comment claims such as 0zKillua 502.xx were not validated. [b28-f7.summary/per_fork/files/notes]

- b28-h1: jaoz158/Victor267 each skip50/retain80. All80 jaoz validate, best d32 -151.472688. Victor7/80 runtime-valid;73 invalid/4,535,966 OOG calls. Three valid finalists lose badly. Different slices explain f1/f2 rankings. [b28-h1.files/notes.inventory]

- b28-h2: sisyphusSmiling empty; solosage159 paths/155 unique, skip50/keep80/leave25. 77 validate; TmpWildcard_v100_candidate/v130/v160_candidate initialization-OOG, unscored. 131-file commit tie uses path order. No CH calibration. [b28-h2.forks/notes/files]

- b28-h3 re-evaluates all47 maxnorm plus giorgoschionas1, not a genuinely later >50 slice. All48 CLI-validate; 15 maxnorm files OOG. V10 has130101 OOG calls; V25–29/V31–33 each533460; V8/V9 each11. Only V43/V39 got 256 seeds; V34 stayed at32 for lack of time. Mechanisms were explicitly not analysed in depth. [b28-h3.files/top3/notes]

- b28-h4:0zKillua path order skip50/keep80/leave13 plus Cheng18=98. V30C stack-depth/V30 initialization250000 failures;96 scored without errors. f7 version-order/h4 path-order slices overlap, not complementary exhaustive coverage. [b28-h4.notes/summary]



### All reported fork finalists

Rows cite top3 entries; n256 on901000..901255 unless32, reported metrics unpooled, OOG0. Negative means lose to RS2. Gas uses each worker’s scope (h3 caveat below). Duplicate hashes share the identity ledger; all finalist observations remain. [respective top3 arrays]

| Source/entry | Code ID | n | Mean | SE | Retail | Arb | Reported gas |

| --- | --- | --- | --- | --- | --- | --- | --- |

| b28-f1/1 | F1 | 256 | -2.165215 | 0.079078106 | -2.552807 | 0.38759193 | 150732 |

| b28-f1/2 | F2 | 256 | -2.4002477 | 0.072032422 | -3.3834363 | 0.98318862 | 150727 |

| b28-f1/3 | F3 | 256 | -4.1128702 | 0.089635791 | -4.6489086 | 0.53603842 | 150822 |

| b28-f2/1 | F4 | 256 | -2.7309055 | 0.077579197 | -1.8374946 | -0.89341087 | 147039 |

| b28-f2/2 | F5 | 256 | -2.7844602 | 0.079453798 | -1.9309773 | -0.85348287 | 129422 |

| b28-f2/3 | F6 | 256 | -3.7299507 | 0.099959493 | -4.5726344 | 0.84268371 | 152739 |

| b28-f3/1 | F7 | 256 | -2.7309055 | 0.077579197 | -1.8372431 | -0.89366233 | 147039 |

| b28-f3/2 | F8 | 256 | -158.15673 | 2.598951 | -27.644722 | -130.51201 | 77064 |

| b28-f4/1 | F9 | 256 | -2.0280744 | 0.06779645 | -2.5141766 | 0.48610216 | 150905 |

| b28-f4/2 | F10 | 256 | -2.7794585 | 0.68213395 | -2.8434026 | 0.063944114 | 153327 |

| b28-f4/3 | F11 | 256 | -2.13634 | 0.066602905 | -1.7174686 | -0.41887132 | 154121 |

| b28-f5/1 | F12 | 256 | -7.5645453 | 0.10280381 | 1.0408719 | -8.6054172 | 150946 |

| b28-f5/2 | F13 | 256 | -37.805198 | 0.48798083 | -17.869437 | -19.935761 | 33390 |

| b28-f5/3 | F14 | 256 | -57.762205 | 0.66690075 | -35.749689 | -22.012515 | 83290 |

| b28-f6/1 | F15 | 256 | -2.7302719 | 0.077574889 | -1.8372338 | -0.89303816 | 152687 |

| b28-f6/2 | F16 | 256 | -2.7309055 | 0.077579197 | -1.8372431 | -0.89366233 | 147039 |

| b28-f6/3 | F17 | 256 | -2.7309055 | 0.077579197 | -1.8372431 | -0.89366233 | 147613 |

| b28-f7/1 | F18 | 256 | -1.8404066 | 0.069791009 | -0.0085150708 | -1.8318915 | 151576 |

| b28-f7/2 | F19 | 256 | -38.621897 | 0.39634455 | -22.984913 | -15.636984 | 92071 |

| b28-f7/3 | F20 | 256 | -38.652861 | 0.3951314 | -23.492822 | -15.160039 | 92083 |

| b28-h1/1 | F21 | 256 | -65.796568 | 0.62286741 | -26.58057 | -39.215998 | 248611 |

| b28-h1/2 | F22 | 256 | -72.88381 | 0.61575333 | -39.252692 | -33.631118 | 222065 |

| b28-h1/3 | F23 | 256 | -87.565533 | 1.139955 | -82.064557 | -5.5009764 | 220984 |

| b28-h2/1 | F9 | 256 | -2.0281 | 0.0678 | -2.5144 | 0.4864 | 150905 |

| b28-h2/2 | F24 | 256 | -2.194 | 0.0723 | -2.3099 | 0.1159 | 150737 |

| b28-h2/3 | F25 | 256 | -2.2409 | 0.0648 | -1.4262 | -0.8147 | 150915 |

| b28-h3/1 | F15 | 256 | -2.7302719 | 0.077574889 | -1.8374853 | -0.8927867 | 207629 |

| b28-h3/2 | F26 | 256 | -2.7266193 | 0.076597337 | -1.8374638 | -0.88915551 | 185532 |

| b28-h3/3 | F16 | 32 | -2.8743338 | 0.24043747 | -2.1086068 | -0.76572695 | 185532 |

| b28-h4/1 | F18 | 256 | -1.8404066 | 0.069791009 | -0.0085150708 | -1.8318915 | 151576 |

| b28-h4/2 | F19 | 256 | -38.621897 | 0.39634455 | -22.985313 | -15.636584 | 92071 |

| b28-h4/3 | F20 | 256 | -38.652861 | 0.3951314 | -23.493222 | -15.159639 | 92083 |



Identity/mechanism ledger: reported source-inspection findings carry the failed-source qualifications. Hash identity is not authorship; equal scores need not mean equal bytes. “Ancestor” is not a causal ablation.

**F1** — `3896e5cf013459844d9bd294484ec0d01fb2a27c3cc7d504a556e3a01cb7bf33`; https://github.com/shl0k28/amm-challenge/blob/9c4be0b64479ebb41c5cb5dd7082007130a155a2/Strategy.sol . [b28-f1/top3/0] BandShield_ghost: EWMA fair estimate with likely-arb/retail alphas, sigma/flow/toxicity/activity, directional skew and tail compression. Shared RS2 family; retail loss outweighs arb saving.

**F2** — `1a55f24618c69c6db659174708ba9e929a3571c844b424283e89b559397d45ce`; https://github.com/shl0k28/amm-challenge/blob/9c4be0b64479ebb41c5cb5dd7082007130a155a2/Strategy_ghost_arbclass.sol . [b28-f1/top3/1] Same ghost family, changed signal/decay/flow/activity constants and 5% tail knee; no reported new mechanism.

**F3** — `967d7dec2264eee29232d908c279ee32e5f0e7db11ae2cd825c1aebdd0aaf503`; https://github.com/shl0k28/amm-challenge/blob/9c4be0b64479ebb41c5cb5dd7082007130a155a2/Strategy_ghost_quick.sol . [b28-f1/top3/2] ghost_quick simplifies price updates to one alpha and first-print sigma; worse retail with small arb saving.

**F4** — `04bd879abe5937e7ad75f3c59ef3dcf58c22a9ba3a9b04db650b573c7ac01353`; https://github.com/jaoz/amm-challenge/blob/7daa7c8cfaa5125fe1a659685066125f5e39f312/Strat/baseline_sourced.sol . [b28-f2/top3/0] Cubic toxicity, uncapped later-print aligned surcharge and a two-speed gated EMA; omits censored/silent-price inference.

**F5** — `017b40f7645b8f981c0183c6f950cf35da91a800ab42a867268f2860db3307f6`; https://github.com/jaoz/amm-challenge/blob/7daa7c8cfaa5125fe1a659685066125f5e39f312/Strat/baseline_sourced_enriched_simple.sol . [b28-f2/top3/1] Cubic toxicity, 1-bp opening/3-bp recurring base, aligned boost only first two prints, zero-fee fallback; simpler EMA.

**F6** — `33659309a9cbe17929389b4e46b7427460d04a741299894b99f5e45d4549b626`; https://github.com/jaoz/amm-challenge/blob/7daa7c8cfaa5125fe1a659685066125f5e39f312/Cursor/strategies/champions/theo1_islandga10h_expanded_worker0_best_20260217.sol . [b28-f2/top3/2] Hard first-print arb classification gates quadratic/cubic raw-size tails; later-print 1.5x boost; endangered-side no-arb floor plus buffer. New elements lose overall.

**F7** — `bcbccc7544a89d3e3c4c1eb13925fbca1df60c13f71ef2f9852988591a2e438f`; https://github.com/wellimbharath/amm-challenge/blob/99c509062af3ed06e5e0c7945dd1b9565014b8e9/contracts/src/Strategy.sol . [b28-f3/top3/0] godknows: cubic toxicity, smoothed-toxicity stale sign around one slow EMA, no later-print16-bp or cheap-side30/36-bp caps; same broad fee family.

**F8** — `d148381bd99e2d2ab06537bfbe576ce06b0e0e9ac88b0a35980bd4d942294b10`; https://github.com/wellimbharath/amm-challenge/blob/19912e7e374fd5eaa993605de97fcee32b5541f4/contracts/src/AdaptiveFeeStrategy.sol . [b28-f3/top3/1] AdaptiveFeeStrategy: hard50-bp ceiling/49-bp discount; size>2.6x updated EMA or >=3 same-direction trades classify arb, else close67% of fee gap. Mostly arb loss.

**F9** — `1235d6641ab3fbfd0ef54f8326ee8cfae24894ecbbb39ce8e5bd926810f5c5e6`; https://github.com/solosage1/amm-challenge/blob/dee8cbf998a81fcfc4e3d4d64be2a11f0f011461/contracts/src/TmpDirectionalSkewMod_v16.sol . [b28-f4/top3/0] 3-bp stale deadband,12-bp quadratic-toxicity knee, hard small-first-print classifier, retail sigma decay, asymmetric tail slopes.

**F10** — `ee70f2cbb58e5d78f1f967b55f4426bd48b93e2129043b19dac4c205e3f75f32`; https://github.com/solosage1/amm-challenge/blob/817a6add437d2c41c52d90c7dd4a52469d85976d/contracts/src/TmpToxAct_v280_candidate.sol . [b28-f4/top3/1] Favorable/adverse toxicity/activity split, classifier multiplier, arrival/size/toxicity gates, capped quadratic term and callback-change limiter, sigma-dependent tox cap.

**F11** — `f922d025153a5d26de29b2da7ac6dec7210b6c3d7b5207e52481c2b440d51f56`; https://github.com/solosage1/amm-challenge/blob/817a6add437d2c41c52d90c7dd4a52469d85976d/contracts/src/StrategyCandidateV274.sol . [b28-f4/top3/2] Sigma softens capped quadratic tox down to .76; side-specific gated activity/toxicity with callback limiter and retail sigma decay.

**F12** — `c6a8136443126776fb5a32ecc16be78539d40671481ae8b87a16bb84738bdfd6`; https://raw.githubusercontent.com/atvinicius/amm-challenge/78752c17e06b3f4b5d304291e29642f13693c2d7/strategy_submission.sol . [b28-f5/top3/0] Idle re-entry discount1bp/empty step after first, max6bps; first-print-only price EMA and toxicity-derived stale shift. Retail gain is overwhelmed by arb loss.

**F13** — `72e777f283b5a9f4cb64670902d47a339907dbd810fb3fdaaf364fbb6c441fb1`; https://raw.githubusercontent.com/alexanderzliu/amm-challenge/e15b10d2193387e14b863342adacf9508821b471/my_strategy.sol . [b28-f5/top3/1] DirContrarian: opposite-direction linear/quadratic size spike above24-bp floor, same-side decay8/9, max of decayed/new spike: explicit directional hysteresis.

**F14** — `1a9e8ba24931986c398505555b576c185129cd750ed5c9c40673db5eb61a07c2`; https://raw.githubusercontent.com/piotrostr/amm-challenge/35b9625f7e3aa57a07627f530240b69e1761e702/contracts/src/SqVolStrategy.sol . [b28-f5/top3/2] SqVol: RMS of EMA squared relative spot changes; symmetric8bps +100bps*RMS/80bps; replaces directional/flow/tox construction.

**F15** — `508c9a449ea20119590ec0e058b4d635251a826796c9125a5384eed198a070ab`; https://github.com/maxnorm/amm-challenge/blob/ee285c088f214145a743bee69628e2a780267492/contracts/src/SapientStrategyV43.sol . [b28-f6/top3/0] V43: slow EMA update .05; symmetric premium when fast/slow gap>1%, cap10%; retains cubic tox. Incremental shared-family trigger.

**F16** — `7cd055c850ae327ab5d3509ef579d3d71d514f5925e1d1785b4fee3d514d37a4`; https://github.com/maxnorm/amm-challenge/blob/ee285c088f214145a743bee69628e2a780267492/contracts/src/SapientStrategyV34.sol . [b28-f6/top3/1] V34: cubic toxicity, first/retail price weights, 3-bp opening; same broad family.

**F17** — `88e280c3b2b4eda33ff528e3814d7b5810cf74fc8d3c936e4f89ef2886a3c3bc`; https://github.com/maxnorm/amm-challenge/blob/ee285c088f214145a743bee69628e2a780267492/contracts/src/SapientStrategyV40.sol . [b28-f6/top3/2] V40: elapsed-step silence surcharge .002 per step, cap8, plus cubic tox; exact same total-edge rows as V34 on requested blocks, no measured benefit.

**F18** — `6dd68f77a70900319460215374143dbd5d924198450a46d9f5f91f6b1e3d35c3`; https://raw.githubusercontent.com/ChengHua926/amm-challenge/d34a5199c59a4c44d5336a25f1436b4ae1f21104/contracts/src/Strategy.sol . [b28-f7/top3/0] CH/no_cubic shared-core variant: one EMA/toxicity stale shift; omits censored price, retail-route, separate cleaned-flow and fee caps. Whole-strategy deficit does not isolate posterior value.

**F19** — `20a47bf7b5fbdf6b6631fe7ffafbb17a32bf6120d0d56041b62521a5978be3ac`; https://raw.githubusercontent.com/0zKillua/amm-challenge/5ae0b44ad59cbb359ebbc0de4cf462d3ceec35e4/contracts/src/V75Strategy.sol . [b28-f7/top3/1] V75: virtual price with .5% spot reversion and signed-size nudge; 9-bp total spread, capped deviation lean, half opposite discount, impact momentum and volume imbalance.

**F20** — `e9dcefcc03554b7b65cd25f1c74d1db731845a8b847eed36c1145b35a5347fae`; https://raw.githubusercontent.com/0zKillua/amm-challenge/5ae0b44ad59cbb359ebbc0de4cf462d3ceec35e4/contracts/src/V73Strategy.sol . [b28-f7/top3/2] V73 uses same virtual-anchor/momentum/volume scheme with10-bp total spread. Initialization is6bps for V73/V75, not their recurring total spread.

**F21** — `2be952154185b4528907794a2a28525bce5d43453b6dc4cfbc9ab792e10f51e0`; https://raw.githubusercontent.com/VictorQian03/amm-challenge/80fd97db0a6420d0b1562b986bb7efd3067776f8/artifacts/scratch_probes/apr21-screen490-1431/round32/two_account_spread_assembly.sol . [b28-h1/top3/0] Separate benign service memory and adverse hazard/directional protection; common and side credits; side credit capped18% of spread above base.

**F22** — `f5d8f3c0f1098441c76224c99508370e2401cc87e56ea1ebf7d2cfb59ca99519`; https://raw.githubusercontent.com/VictorQian03/amm-challenge/80fd97db0a6420d0b1562b986bb7efd3067776f8/artifacts/scratch_probes/apr21-screen490-1431/round34/rank_copula_risk_splitter.sol . [b28-h1/top3/1] Fixed ordinal size/movement/imbalance ranks select one of five hazard states; threshold interactions, not an estimated statistical copula; reconvergence decay.

**F23** — `a36457499faf7997ea301b6f9b2efebc83198676719e67ad8aa0c00c1b91a983`; https://raw.githubusercontent.com/VictorQian03/amm-challenge/80fd97db0a6420d0b1562b986bb7efd3067776f8/artifacts/scratch_probes/apr21-screen490-1431/round34/meet_semilattice_quote_compiler.sol . [b28-h1/top3/2] Adverse floor and calm-service ceiling assembled with max; ceiling=floor+nonnegative slack, so calm slack raises executed fee despite benign-reduction terminology.

**F24** — `4fa0f2768ca32de600a4f358ed21412165bf79ecf1297865f8b3111db6bfcb19`; https://github.com/solosage1/amm-challenge/blob/99dc1cfd0963c274df5f2145e8a282a78226fe9e/contracts/src/TmpEval.sol . [b28-h2/top3/1] TmpEval: ungated stale overlay/full quadratic tox, cap1/6, symmetric .90 compression above700bps; no posterior or sigma-tox cross/trade-size boost.

**F25** — `99c8fde58314973c50a38aa32dff44bcb521553203e0bf25b8e59af6e270b70a`; https://github.com/solosage1/amm-challenge/blob/99dc1cfd0963c274df5f2145e8a282a78226fe9e/contracts/src/TmpToxAct_v206_candidate.sol . [b28-h2/top3/2] v206: activity only if lambda>=.95 and size>=.3%; fades as tox grows from12bps; three-case tail slopes and ungated stale shift.

**F26** — `5567ee16c871c753010cfba3c6e1e25173b0b209289735a58ed6d752e15a526e`; https://github.com/maxnorm/amm-challenge/blob/ee285c088f214145a743bee69628e2a780267492/contracts/src/SapientStrategyV39.sol . [b28-h3/top3/1] V39: YQ-derived Sapient family; no in-depth mechanism analysis supplied. Its small numerical difference from V43 is not a demonstrated advance over RS2.

Worst individual seed for each distinct finalist, from its first occurrence (six significant digits). The other four entries of each reported five-lowest array are consolidated into this minimum; repeated identical blocks add no independent tail samples. h3 V34 has only32 seeds. [respective top3.paired.worst]

F1 [b28-f1]: 901180:-9.42133.

F2 [b28-f1]: 901033:-7.01918.

F3 [b28-f1]: 901033:-11.0111.

F4 [b28-f2]: 901156:-9.29795.

F5 [b28-f2]: 901156:-9.70382.

F6 [b28-f2]: 901180:-11.7745.

F7 [b28-f3]: 901156:-9.29795.

F8 [b28-f3]: 901169:-333.99.

F9 [b28-f4]: 901034:-7.11632.

F10 [b28-f4]: 901030:-175.598.

F11 [b28-f4]: 901180:-7.00598.

F12 [b28-f5]: 901068:-13.5861.

F13 [b28-f5]: 901007:-89.9751.

F14 [b28-f5]: 901133:-88.2724.

F15 [b28-f6]: 901156:-9.2975.

F16 [b28-f6]: 901156:-9.29795.

F17 [b28-f6]: 901156:-9.29795.

F18 [b28-f7]: 901156:-8.5179.

F19 [b28-f7]: 901181:-56.8201.

F20 [b28-f7]: 901181:-56.7024.

F21 [b28-h1]: 901169:-93.2677.

F22 [b28-h1]: 901133:-98.567.

F23 [b28-h1]: 901196:-178.512.

F24 [b28-h2]: 901180:-7.5985.

F25 [b28-h2]: 901180:-7.3191.

F26 [b28-h3]: 901156:-9.34468.

### Reconciling duplicated and conflicting fork evidence

b28-f2 baseline_sourced, b28-f3 godknows and b28-f6 V34 have different source hashes but the same total mean -2.730905464885787 and SE .07757919730518127 on256 seeds. The retail/arb split in f2 is -1.8374945991763987 / -.8934108657093915; f3/f6 correct actual3-bp initialization and report about -1.83724313757450 / -.893662327311287. Keep hashes/correction distinct; equal scores prove neither independent trials nor byte identity. f3’s zero-retail diagnostic with actual initial fees gives max reconstruction error9.070300066582604e-12 across its tested arms, including50-bp AdaptiveFeeStrategy. [b28-f2.top3/notes.statistics; b28-f3.top3/decomposition_check; b28-f6.top3/notes]

solosage TmpDirectionalSkewMod_v16 is the identical hash in f4 and h2, with equal total mean at h2 rounding. f4 actual-initial-fee decomposition -2.514176578293492 / +.4861021550901968 differs slightly from h2 unchanged-harness rounded -2.5144 / +.4864. f4 reports no root licence and no source SPDX; h2 additionally reports inherited pyproject MIT metadata. No separate original licence grant is established. Same-hash overlap despite intended recency slices reflects tied commit/path-selection order; slices are not assured disjoint. [b28-f4.top3/license/notes; b28-h2.top3/code/notes]

V43 is the same source hash and identical total mean/SE in f6/h3. f6 reports corrected initial-fee retail/arb -1.8372337884720742 / -.8930381566214871 and specifically afterSwap152687. h3’s unchanged-harness split is -1.8374852500739687 / -.8927866950195917 and gas207629 includes initialization. h3 wording says max_ok “summed over all rows”; values185532 for V34/V39 and207629 for V43 are not certified afterSwap maxima. Do not compare them directly with f6’s afterSwap values. h3 explicitly lacks deep mechanism analysis and a256-seed V34 run; f6 supplies V34’s256-seed evidence, without increasing independent n for repeated blocks. f6 reports source SPDX MIT; h3 says no fork-specific licence. Embedded V43 source contains SPDX MIT, while neither proves a separate root LICENSE exists. [b28-f6.code/top3/notes; b28-h3.top3/notes/code]

CH, V75 and V73 repeat exactly between f7/h4 on the same256 seeds. CH split agrees. V75 f7 retail/arb -22.98491344951256 / -15.636983522717886 becomes h4 corrected -22.985313099884568 / -15.636583872345883; V73 f7 -23.492822269742483 / -15.160038960041561 becomes h4 -23.493221920114486 / -15.159639309669556. Actual initialization6bps versus harness4bps explains the first-step attribution adjustment (max absolute correction .0031375040894321415); totals/ranking unchanged. h4 checked first-step fair-price invariance on an eight-seed prefix. f7’s claim the entire CH deficit “is what that posterior adds” overstates causal isolation: h4/source inspection identifies other omitted caps, routing and flow mechanisms. Retain only the whole-strategy deficit, not that attribution. [b28-f7.top3; b28-h4.top3/notes]

Author/licence reports remain qualified: shl0k28 selected source attributed to Shlok, SPDX MIT/no root licence; jaoz last source-changing author jaoz, source SPDX/pyproject MIT; wellimbharath selected author recorded as Claude (owner wellimbharath), SPDX/pyproject MIT; solosage1 as above; atvinicius selected author Vinicius, SPDX MIT/no root file; maxnorm selected author maxnorm, source SPDX MIT/no root file; CH author ChengHua926, SPDX MIT/no root file; Victor selected author Victor Qian, SPDX MIT/no root file. Metadata proves neither sole authorship nor clearance. [b28-f1.code; b28-f2.author/licence; b28-f3.author/licence_evidence; b28-f4.author/license; b28-f5.author/licence; b28-f6.author/licence; b28-f7.code; b28-h1.author/licence; b28-h2.code; b28-h3.code; b28-h4.author/licence]

Unproven mechanism leads: cubic tox; classifier/floor/tail surcharge; deadband/knee; side gates; idle discounts; RMS volatility; virtual anchor/momentum; benign-service budgets; ordinal hazard. Tested whole strategies lost; isolated components were not thereby falsified. Repeated ancestor-family tests do not satisfy the goal’s whole-family exploration intent. [fork mechanism ledger and top3 results above; requester prior-goal guidance]

## Remaining-fork audit and missing coverage

b28-rest reports one GitHub forks API page https://api.github.com/repos/benedictbrady/amm-challenge/forks?per_page=100 returning74 forks:46 pushed before the exclusive2026-02-08T00:00:00Z cutoff,28 date-excluded. Of46,43 had only upstream branch heads,2 were inaccessible, and1 accessible eligible fork (adietrichs) had no changed Solidity afterSwap. n=0; files/top3 empty; code/hash/author/licence null. This is a qualified audit report (7 failed sources), not proof that every possible fork or historical file was searched. [b28-rest.discovery/inspection/blocked]

adietrichs has extra head `3dd21541389b92d594268023b024a99a8ba810b0` versus upstream `09d54fb66a0dd0f74aaf961af28478400e121f05`; only Rust engine/result trade-count fields change and both branch heads retain upstream Solidity blobs. Comparison source: https://api.github.com/repos/adietrichs/amm-challenge/compare/09d54fb66a0dd0f74aaf961af28478400e121f05...3dd21541389b92d594268023b024a99a8ba810b0 . `0xAxiom/amm-challenge` and `mawdbot/amm-challenge` failed anonymous git access with “could not read Username”; their eligibility/content is unknown, not empty. No requester input or fabricated candidate is supplied for those runtime access failures. [b28-rest.inspection/blocked/notes]

43 upstream-only exclusions: Edward-Alkimie/amm-challenge, hackingdecentralized/amm-challenge, dmnkf/amm-challenge, mmsaki/amm-challenge, Oxkai/amm-challenge, rohanphw/amm-challenge, darrylyeo/optimization-arena-amm-challenge, b0ggs/amm-challenge, chadury2021/amm-challenge, detectivekim/amm-challenge, orepapas/amm-challenge, Rajesh171196/amm-challenge-paradigm, Keeeeeeeks/amm-challenge, Sub-MesonBrain/amm-challenge, jalbrekt85/amm-challenge, consentsam/amm-challenge, kubaplata/amm-challenge, yandhii/amm-challenge, sammed-21/amm-challenge, ericli/amm-challenge, thogiti/amm-challenge, iamvon/amm-challenge, orperelman123/amm-challenge, babyhome/amm-challenge, demirelo/amm-challenge, Aditya-dom/amm-challenge, anishbuilds/amm-challenge, artyfikalny/amm-challenge, Define101/amm-challenge, devblixt/amm-challenge, wenakita/amm-challenge, MuhtasimTanmoy/amm-challenge, BintuParis/amm-challenge, nomorelonelylife/amm-challenge, nabaruns/amm-challenge, srinathhariharan/amm-challenge, davidrhodus/amm-challenge, euler-mab/amm-challenge, joeabbey/amm-challenge, sneg55/amm-challenge, RohanShrothrium/amm-challenge, ChiShengChen/amm-challenge, igoryuzo/amm-challenge. [b28-rest.discovery.fork_audit]

The28 date exclusions are the f1–f7 groups; rest does not supersede them. Untested older/non-head/cap-excluded files, inaccessible forks, partial searches and missing checks remain gaps. Next useful evidence is a frozen mechanism versus exact STACK on >=2,000 fresh >=5,000,000 seeds with OOG/retail/arb deltas. **This root has none**, without proving the goal impossible. [b28-rest.discovery.excluded_dates; all coverage exceptions above; requester method]

## Confirmed incumbent code (lossless encoding)

Exact b24-a/b UTF-8 code, bzip2/Base64 encoded. Decoded SHA-256: `f24b6ca198b744c380842c3a17964a6dc21f311bb04984b26733afa3fee8eaf2`. Recover with standard-library `bz2.decompress(base64.b64decode(payload))`. Imports challenge AMMStrategyBase.sol/IAMMStrategy.sol; recovery does not certify metrics. [b24-a.code/sha256; b24-b.code/sha256]

```base64
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
```