# Root merge 5: fee-policy training and T03 confirmation

All 25 supplied results are present. The strongest added confirmation concerns existing T03 (code cb6d9455…): +0.8004589765907124 (SE 0.025905856501574405, n=1024) and +0.772455789197696 (SE 0.024109904134075353, n=1024) versus IB on adjacent disjoint blocks. The first uses a native engine build rather than the required wheel. These confirm an existing policy, not new training gains. [S10, S11]

Several monitor-selected changes improve the replica slightly, but their reported real frozen-minus-START results do not decisively confirm additional gains; jc03 is slightly negative. Most later searches keep START unchanged. Different START families, shared validation sets and different callback protocols prevent treating these records as one uninterrupted training chain. [S02, S05–S09, S13–S16, S18–S25]

## Evidence, conflicts and limits

Each S-label identifies the supplied JSON and byte SHA-256; source paths below are relative to artifacts/goal/batches/. The input filename is the section title under .imd/reads/artifacts/. Every number in a source section comes from that record's named fields. Worker measurements are reported claims, not independently rerun experiments. No external page was revalidated for this merge.

The requester flags failed/unreadable/unsupported upstream citations for every source except S01: two each for S02–S07, S09, S13, S14, S17, S21, S22; three for S08; one each for S10, S11, S12, S15, S16, S18, S19, S20, S23, S24, S25. The warning does not identify the individual URLs/quotations that failed. Thus upstream performance references, artifact provenance and workers' hash-verification assertions remain attributed, uncorroborated claims. Input hashes do not certify results; see S17's inherited-data contradiction.

Requester context: IB=522.00, 20th, T plus learned 19-8-2 correction, +0.59 over HA. Sister goal cc45e1a7 reportedly gained +0.33 in 210 CMA generations and +0.59 in 910. These figures are context, not measurements produced here. Requester's IB code SHA-256 is 301941829a44bb60977bb28181a5e4bfa6544f3696763f987054789b2611d686. The [IB artifact](https://api.imd.fun/artifacts/afd5183e9ff3235334e041c00b0cd19079599a2c6464db95f8c1586712105557) and [KIT-B artifact](https://api.imd.fun/artifacts/48cd6a10f39fc22a79c5f12398474bd65282803d59314819acd586f187015808) are requester-provided identifiers, not newly fetched evidence. No result establishes a new board score/rank or independently verifies the context's IAD132k-versus238k gas claim.

Required gates: A is 50×2000 callbacks, error <0.01 bps; B reproduction tolerance 0.15; frozen C error <0.1 bps. Worker passes remain conditional on their harnesses.

**Gate-reference mismatch:** the goal requests reproduction versus HA. Most records instead calibrate START minus IB on replica 8588000..8589023 against supplied real +0.819 (some also mention+0.765 after outlier exclusion); jc records calibrate on 8078848..8079359 against +0.595. Their B passes refer to those alternative checks, not a new HA comparison. Repeated+0.7692214604247253/SE0.026912958923414727 comes from the same fixed block, not independent fresh evidence. Common holdouts 41030000..41038191 and 41050000..41058191 are also reused across workers and must not be pooled as independent samples. [S02–S09, S12–S25]

**Protocol differences:** k17/k18 fail unbounded 15%-trade random walks while k19/k20 pass canonical callbacks; later bounded 2%-trade/spot gates generally pass alongside stress failures. A pass on a changed harness does not erase the prior failure. k20,m03,m13 and k16's initial segment search below this goal's 300 markets/s minimum, despite claimed local item overrides. m12 searches replica-only without A/C. Validity scopes differ: m04 overall false but real.valid true; m10 true on retained calls despite missing real block; some skipped-search records true without C. Preserve those distinctions. Replayed real attempts and below-floor probes are identified individually. [S02–S07, S12–S25]

**Metric meaning:** mean_diff is paired edge difference; se is its reported standard error, not a leaderboard uncertainty. Workers describe real tests as each strategy alone against 30 bps Vanilla via amm_sim_rs.run_batch(X,VanillaStrategy,configs,8), MatchRunner config draw order, usually 64-seed segments. Y-volume deltas are not edge deltas. Some workers reconstruct retail/arb edge components from traces; initial-fee and volume-check caveats remain attached. The 245000 gas limit concerns afterSwap, not afterInitialize. Measured zero OOG does not cover missing observations. [S02–S25]

Most searches use 219 coordinates: 178 head weights + 41 constants, frozen feature scaling, common fresh markets among proposals each generation, ES-mean selection on monitor 30000000..30002047. Earlier k/jc maps generally use .03×block RMS for weights, constants 2^clip(.02z,-1,1), decays via 1-d; later m maps generally .05×RMS and2^(.03z), with decoder clamps noted below. Weight freezing rounds 1e-6, constants to WAD/integers. Common port vector: [1,19 feature multipliers,19 offsets,W1 row-major,b1,W2 row-major,b2,3 anchors,TH_VALUES]; NW=217, anchors W[217:220], TH from W[220:], constants substituted by name. IB must use its own exact port/head. Identical parameterized-port hashes do not imply identical policy weights/code. [S02–S09, S12–S25]

This ledger keeps statistical summaries, identities and caveats; repeated speed timing probes are condensed. For the size limit, raw per-seed arrays, full training curves, redundant sums/sumsq, per-callback gas counters, base64/source payloads, full weights and optimizer vectors are intentionally not reproduced. This merge is a write-up, not a runnable continuation package; omitted payloads cannot be recovered from a code hash alone. No missing value is filled as zero.

## Source ledger

Comparison F/S/I means frozen/START/IB. Tables use n/mean_diff/se, with R/A for retail/arb. Missing entries are unavailable, not zero. Core fields that are identical by code identity are reported once.

Hash aliases (exact SHA-256; code and port role appears per item):

- H1: cb6d945511e1e10af78baa64c2a6cb25806f0998403ce6440654bfc3c70092d9
- H2: b29cd2a078b4c3fc57945a79e927624dd346c44d8cc149dae43f4929aa6d736b
- H3: 41cd45e156aa173f614607de55add46a7677bbc5f9cbee09f6e62e83060ef45b
- H4: 732c0fb615480020795039a111e4a49a46f851e6f29720b092665b247d79006b
- H5: 2d064ed29e272f0f5b2c926361b510346f11c53e41517642d73a504ab1269f09
- H6: 1be16eb82ba3170aef776d5f4061fb8c882ca52210628d97fd17f2dfe1096b9c
- H7: 572bb0e6b7eb78ce84d95a48730ba05e699914643c0a062892a5a73970e1f372
- H8: ea4047d3a6deedeaeeb155797b58cee3517c2004f8c5961c68485598f133ff4e
- H9: e87231ca0fcdc5a009d143e479b739065b0ec498970ff773970cd5b5db806155
- H10: 5bd6fbe55ad12166c148bc0273da4997eafe86af196f8dc5d13db426a6d01f55
- H11: daa73a4cea0b8faf3578672b355da9adf04f0dae6178c9fde375d7bbdea03627
- H12: 802d1401c1f602bfe19d9ec50caa9dd2f1ee816509d77cc5327efc9f1bff6dc1
- H13: 3f157d3e7c6f8e4c82a4c3fb2332f4503da729c72232dfe72f0e7c9980ae0949

### S01 — b25-k15-s04-h90-p24

Source: t3-c/items/k15-s04-h90-p24/result.json; SHA-256 b07d40d15b0d95232331aa1bbc5017810476e97e107dd26e89f65b7cc96b37ef.

Setup stopped before dependency installation because the supplied script split an unquoted workspace path containing spaces. No gates, speed, search, holdout, real run or gas measurement occurred. START returned unchanged; valid=false means unvalidated, not measured OOG. Hashes were reportedly checked, but fee equivalence was not.

Code H1. Port H2. valid=false.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|not_run|—|
|B|not_run|—/—/—; target —|
|C|not_run|—|

Speed (markets/s, workers named): {}.

Search: {"gens":0,"min":0,"M":null}.

Replica holdout: unmeasured. 

Real evaluation: unmeasured. 

Gas: {"afterSwap_peak":{"frozen":null,"START":null,"IB":null},"oog":null}.

### S02 — b26-k16-s05-h45-p16

Source: t3-d/items/k16-s05-h45-p16/result.json; SHA-256 b1eaf76246f77efbc91cbbf7d431343b8738904813c57b0f7528e1037db1a77c.

Changed monitor-selected candidate. Discarded 46 generations/17.269318279922786 minutes omitted NumPy RNG on resume; retained 250 generations restarted at z=0 with full RNG checkpointing and unused seeds. Total search 75.63225869133991 minutes. Initial throughput 259.0135352502567 with M=256 for generations 1–54 conflicts with this goal's >=300 rule; later segments used M=512 and faster throughput. Two checkpoint restores yielded identical next proposals. Only numeric literals changed; baseline freeze reproduced START bytes. Small real gain over START is inconclusive. No real edge split; volume/PnL deltas measured. First IB chunk ran during training for budgeting, not selection.

Code H3. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.00067878087|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|true|0.00126937975|

Speed (markets/s, workers named): {"markets_per_s":615.4411135639375,"initial_markets_per_s":259.0135352502567,"cpus":8}.

Search: {"gens":250,"seed":20261316,"pop":16,"dim":219,"min":58.3629404,"M":512,"bestMonitorGain":0.0242561637,"seed_start":1276011776,"seed_end":1276125951,"monitor0":30000000,"monitorEnd":30002047}.

Replica seeds: 41030000..41038191 (from source notes):

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0.0254837673/0.00418933224|—/—|—/—|
|frozen_vs_IB|8192/0.793357281/0.00965071482|—/—|—/—|
|start_vs_IB|8192/0.767873513/0.00947268305|—/—|—/—|

Real seeds: [8112640,8113151].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|512/0.0221887515/0.0164820491|—/—|72.5566359 ± 2.70979155/144.763128 ± 2.7497175|8113015:-1.96714538; 8113034:-1.49198382; 8112756:-1.24007731; 8113134:-1.06670028; 8112653:-1.03127391|
|frozen_vs_IB|512/0.76278106/0.0384088803|—/—|369.645855 ± 14.852061/108.583966 ± 10.0190376|8112750:-2.55021665; 8112896:-1.63949491; 8112776:-1.61935872; 8113034:-1.50413984; 8112701:-1.46796017|
|start_vs_IB|512/0.740592309/0.0372942588|—/—|297.089219 ± 13.9922164/-36.1791616 ± 9.17470908|8112750:-2.67313544; 8112776:-1.91696748; 8112896:-1.61936886; 8112701:-1.35294374; 8113133:-1.30801483|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_start/1:128/0.0149729985/0.0216506789; frozen_vs_start/2:128/-0.0169557103/0.0366178932; frozen_vs_start/3:128/0.100174332/0.0305901911; frozen_vs_start/4:128/-0.00943661396/0.0394787971; frozen_vs_IB/1:128/0.729205624/0.0555036939; frozen_vs_IB/2:128/0.784448807/0.0746808449; frozen_vs_IB/3:128/0.822138925/0.0840155319; frozen_vs_IB/4:128/0.715330885/0.0892677061; start_vs_IB/1:128/0.714232625/0.0520256061; start_vs_IB/2:128/0.801404518/0.068911371; start_vs_IB/3:128/0.721964594/0.084944458; start_vs_IB/4:128/0.724767498/0.0876065742.

frozen_vs_start PnL delta: 0.00898386611 ± 0.0200483731.

frozen_vs_IB PnL delta: 0.758103931 ± 0.0479043795.

start_vs_IB PnL delta: 0.749120065 ± 0.0480850272.

Gas: {"afterSwap_peak":{"frozen":237493,"ib":236928,"start":237493},"oog":0}.

### S03 — b26-k17-s03-h60-p16

Source: t3-d/items/k17-s03-h60-p16/result.json; SHA-256 3c2d5b4eee05a49124d711ce038827bba8e8cd51f070b15e25af15d4b921a032.

No search after failed A: absence of evaluation, not evaluated candidates losing. Unbounded 15%-X-reserve harness completed 49/50 sequences and 99,424 callbacks; seed 310037 callback 1424 panic 0x12; max discrepancy 0.4057340000000742 bps. Seed 310019 independently failed at 0.014010709184150705 bps. B passed independently. No holdout, real, C, peak gas or OOG measurement. Returned argument-driven port was not established exact by this gate.

Code H1. Port H4. valid=false.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|false|0.405734|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|—|—|

Speed (markets/s, workers named): {"markets_per_s":122.07163041944544,"cpus":8}.

Search: {"gens":0,"min":0,"M":256,"status":"not_started_failed_gate_A","es_seed":20261317,"pop":16,"sigma0":0.3}.

Replica holdout: unmeasured. not_run_after_failed_gate_A

Real evaluation: unmeasured. not_run_after_failed_gate_A

Gas: {"afterSwap_peak":{"frozen":null,"START":null,"IB":null},"oog":null,"status":"amm_sim_rs.gas_stats not measured because real-engine evaluation was not reached","gate_A_reverts":1}.

### S04 — b26-k18-s05-h30-p24

Source: t3-d/items/k18-s05-h30-p24/result.json; SHA-256 104dc044eaef14e07999ee8550021fbf275a5022a73b64535473052d160a457a.

Same failed 15%-reserve random walk as k17: 0.4057340000000742 bps, seed 310037 callback 1424 panic 0x12, 99,424 successful callbacks. Seed 310019 also failed at 0.014010709184150705 bps. B and speed measured, no search/C/holdout/real. START unchanged. Extreme synthetic excursions do not prove failure on real-engine market paths.

Code H1. Port H4. valid=false.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|false|0.405734|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|—|—|

Speed (markets/s, workers named): {"markets_per_s":349.72278018308043,"cpus":6}.

Search: {"gens":0,"min":0,"M":512,"status":"not run: Gate A failed"}.

Replica holdout: unmeasured. not run: Gate A failed

Real evaluation: unmeasured. not run: Gate A failed

Gas: {"afterSwap_peak":{"frozen":null,"START":null,"IB":null},"oog":null,"status":"Real-engine gas statistics not measured; callback test had a division-by-zero revert, not an OOG."}.

### S05 — b26-k19-s06-h60-p16

Source: t3-d/items/k19-s06-h60-p16/result.json; SHA-256 f6c839b542619ee07ab013e9ccd54f171e5eaf8ba51e9d31289855afeb291afe.

Changed monitor-selected policy. Search pool measured 323.5607175873776 markets/s; slower concurrent B timing is not search speed. Thirteen sub-eight-minute commands with confirmation reserve reported. Separate unanchored 15%-trade diagnostic had 0.01401 bps discrepancy and division-by-zero; replaced by KIT-B canonical generator for required A, which passed. This only validates the latter harness. No real edge split; measured volumes. First IB real chunk during training for budgeting only.

Code H5. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.000198228122|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|true|4.25730075e-05|

Speed (markets/s, workers named): {"markets_per_s":323.5607175873776,"cpus":4}.

Search: {"gens":215,"min":78.5856186,"M":512,"bestMonitorGain":0.0201303356,"best_monitor":{"gain_over_START":0.0201303356},"settings":{"es_seed":20261319,"popsize":16,"sigma0":0.6,"sigma_cap":".6 * 0.5^(search_minutes/60)","weight_transform":"start + 0.03*rms(block)*z; blocks W1,b1,W2,b2","constant_transform":"start*2^clip(.02*z,-1,1); DECAY via 1-d; four integer constants rounded"}}.

Replica seeds: [41030000,41038191]:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0.0254118297/0.00425338494|0.0358963803/-0.0104845506|-10.1671265/—|
|frozen_vs_IB|8192/0.793285343/0.00953907233|-0.201150321/0.994435664|276.010278/—|
|start_vs_IB|8192/0.767873513/0.00947268305|-0.237046701/1.00492021|286.177405/—|

Real seeds: [8115712,8116223].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|512/0.0164049171/0.0162553594|—/—|-10.3764865/25.3996208|8115759:-1.35569957; 8116093:-1.09682817; 8116162:-1.04944881; 8116141:-0.942067271; 8116202:-0.909723021|
|frozen_vs_IB|512/0.826588782/0.035736142|—/—|286.124999/-21.8297742|8115892:-2.05003669; 8115963:-1.51876079; 8115723:-1.44356254; 8115991:-1.07577104; 8115981:-1.00075148|
|start_vs_IB|512/0.810183865/0.0367559037|—/—|296.501486/-47.229395|8116024:-1.62984337; 8115892:-1.49791746; 8115963:-1.46611203; 8115991:-1.336865; 8115779:-1.30163973|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_start/1:128/0.0311658919/0.0280428856; frozen_vs_start/2:128/0.0386975957/0.0294006921; frozen_vs_start/3:128/-0.0150314875/0.0312064102; frozen_vs_start/4:128/0.0107876683/0.0401802411; frozen_vs_IB/1:128/0.764066029/0.0600219313; frozen_vs_IB/2:128/0.812169463/0.0612523091; frozen_vs_IB/3:128/0.915974121/0.0782910619; frozen_vs_IB/4:128/0.814145518/0.0835062981; start_vs_IB/1:128/0.732900137/0.0614284412; start_vs_IB/2:128/0.773471867/0.0585968944; start_vs_IB/3:128/0.931005608/0.0834934589; start_vs_IB/4:128/0.803357849/0.0859792788.

Gas: {"afterSwap_peak":{"frozen":235388,"start":235406,"ib":232035},"oog":{"frozen":0,"start":0,"ib":0}}.

Training range: 1288000000..1288110079. [S05, seeds.train]

### S06 — b26-k20-s04-h60-p16

Source: t3-d/items/k20-s04-h60-p16/result.json; SHA-256 b4669de39651a7498c2a914e1d59ff8c4e16fe6cff54fb1a1ef8d288b9e202d9.

Changed monitor-selected policy. Search at 215.5854595598111/M=256 used a claimed item override, inconsistent with this goal's >=300 rule. Decays through 1-d; optimizer and NumPy RNG checkpointed. Numeric-only freeze. Real edge deltas null; replica components and real volumes measured. Real gain over START has SE nearly equal to gain. No leaderboard gain measured.

Code H6. Port H7. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.000198228122|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|true|8.6802904e-05|

Speed (markets/s, workers named): {"markets_per_s":215.5854595598111,"cpus":8}.

Search: {"gens":196,"min":62.1943488,"M":256,"bestMonitorGain":0.023872532,"es_seed":20261320,"pop":16,"dim":219,"sigma0":0.4,"seeds":[1292000000,1292050175],"best_monitor":{"gain_over_START":0.023872532,"seeds":[30000000,30002047],"selected_curve_row":[53.5015497,0.023872532,0.00712751824,0.215823074]},"settings":{"es_seed":20261320,"sigma0":0.4,"weight_transform":"start + 0.03*rms(block)*z","constant_transform":"start*2**clip(0.02*z,-1,1); decay via 1-d; integers rounded","sigma_cap":"0.4*0.5**(minutes/60)"}}.

Replica seeds: [41030000,41038191]:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0.0214116238/0.00361625001|0.00273696401/0.0186746598|-28.4812607/—|
|frozen_vs_IB|8192/0.789285137/0.00956412957|-0.234309737/1.02359487|257.696144/—|
|start_vs_IB|8192/0.767873513/0.00947268305|-0.237046701/1.00492021|286.177405/—|

Real seeds: [8116736,8116991].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|256/0.0217607629/0.0212601462|—/—|-25.2851723 ± 3.66250242/2.19959158 ± 3.89831742|8116829:-0.970046315; 8116868:-0.922718983; 8116741:-0.867135487; 8116845:-0.823978805; 8116984:-0.802843777|
|frozen_vs_IB|256/0.796781342/0.0520657417|—/—|301.235871 ± 22.032062/-13.5610814 ± 14.8927241|8116974:-1.42501822; 8116766:-1.34740428; 8116932:-1.34172644; 8116978:-1.09379876; 8116827:-0.866348658|
|start_vs_IB|256/0.775020579/0.0522572947|—/—|326.521044 ± 21.5181649/-15.760673 ± 13.8857684|8116978:-1.56093838; 8116974:-1.49232049; 8116932:-1.43422875; 8116890:-1.31487697; 8116766:-1.11544093|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_start/1:64/0.0188813331/0.0264583339; frozen_vs_start/2:64/0.045665757/0.0497028981; frozen_vs_start/3:64/0.00392159081/0.0456032899; frozen_vs_start/4:64/0.0185743708/0.0453155469; frozen_vs_IB/1:64/0.632177015/0.0867959615; frozen_vs_IB/2:64/0.822549612/0.114070032; frozen_vs_IB/3:64/0.91313284/0.0826140593; frozen_vs_IB/4:64/0.819265902/0.125919416; start_vs_IB/1:64/0.613295682/0.0882867593; start_vs_IB/2:64/0.776883855/0.107466104; start_vs_IB/3:64/0.909211249/0.0828189159; start_vs_IB/4:64/0.800691532/0.131527053.

Gas: {"oog":0,"afterSwap_peak":{"frozen":235406,"start":235417,"ib":234957}}.

### S07 — b27-jc01-s03-h45-p16

Source: j11-a/items/jc01-s03-h45-p16/result.json; SHA-256 64097db26c03be81e8de1f06a9e4fe76ff6268bf0e87b0d113b5ff88079257d9.

Different START family (ea4047d3 rather than cb6d9455). No search at S=44.3; inherited restart sigma 0.119/next seed 1152251904 is not new training. Initial A had five ~1.49 bps callbacks in sequence 320025 at degenerate spot; fresh bounded [77,130] seeds 330000..330049 passed. C not run. Real 256/1024 planned. IB probe 7999100..7999103 below goal seed floor. Concurrent real attempts timed out and were rerun singly on same seeds: 'once' refers to retained outputs, not all execution.

Code H8. Port H9. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.000470261736|
|B|true|512/0.544040078/0.0361890368; target 0.595|
|C|—|—|

Speed (markets/s, workers named): {"markets_per_s":44.3,"cpus":8,"processes":8}.

Search: {"gens":0,"min":0,"M":null}.

Replica holdout: unmeasured. 

Real seeds: [8117760,8118015].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|start_vs_IB|256/0.541735841/0.0554473048|—/—|108.018799/-132.154434|8117882:-3.20176435; 8117930:-2.60093852; 8117933:-1.82188883; 8117989:-1.41831079; 8117829:-1.36546028|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): start_vs_IB/1:64/0.493227021/0.0685602573; start_vs_IB/2:64/0.583441144/0.0906680049; start_vs_IB/3:64/0.642865998/0.130973917; start_vs_IB/4:64/0.447409201/0.139040165.

Gas: {"afterSwap_peak":{"START":235369,"IB":234837},"oog":{"START":0,"IB":0}}.

### S08 — b27-jc02-s02-h90-p16

Source: j11-a/items/jc02-s02-h90-p16/result.json; SHA-256 1916ceb7a993c61bd8f58658fc1b16d526bb49e2b437305218b80630f83ad7b4.

ea4047d3 START family. Selected ES mean at generation 417, minute 60.96, monitor +0.023396653888904523 SE 0.007072795465346065; port unchanged. Baseline freeze byte-identical. Holdout and real improvement each only about 1.6 SE, not a strong gain. Final sigma 0.10114461788720204; restart mean is relative to this leg's START, not returned frozen point. Real edge split unavailable; volumes differ from edge.

Code H10. Port H9. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.000686542046|
|B|true|512/0.544040078/0.0361890368; target 0.595|
|C|true|0.00111338743|

Speed (markets/s, workers named): {"markets_per_s":939,"processes":30,"cpus":32}.

Search: {"es_seed":20261322,"pop":16,"dim":219,"sigma0":0.2,"cap":"0.2*0.5^(search_minutes/90) after every generation","gens":595,"min":69.24,"M":512,"train":[1300000000,1300304639],"selected":{"generation":417,"minutes":60.96,"monitor_gain_over_START":0.0233966539,"monitor_se":0.00707279547},"final_sigma":0.101144618}.

Replica seeds: [41030000,41038191]:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0.00633103672/0.00397923062|—/—|—/—|
|frozen_vs_IB|8192/0.616902381/0.00983710678|—/—|—/—|
|start_vs_IB|8192/0.610571345/0.00977042302|—/—|—/—|

Real seeds: [8118784,8119295].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|512/0.0221922316/0.0140636563|—/—|-93.5231628/-124.611272|8118940:-1.76271145; 8119257:-1.69396036; 8118999:-1.25637661; 8119235:-1.05672659; 8119288:-0.913882434|
|frozen_vs_IB|512/0.589859227/0.0381171642|—/—|31.3307698/-242.671245|8118820:-2.31125246; 8119185:-2.22371661; 8119063:-2.18265955; 8118993:-2.08928165; 8119284:-1.85578147|
|start_vs_IB|512/0.567666996/0.0363469465|—/—|124.853933/-118.059973|8119063:-2.16674677; 8118820:-2.06231816; 8118993:-2.01719907; 8119284:-1.7426246; 8119185:-1.73424704|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_start/1:128/0.00662955733/0.0254570427; frozen_vs_start/2:128/0.013487662/0.0252282471; frozen_vs_start/3:128/0.043629173/0.0291900911; frozen_vs_start/4:128/0.025022534/0.0322567707; frozen_vs_IB/1:128/0.615456031/0.0610237478; frozen_vs_IB/2:128/0.481686837/0.0765769131; frozen_vs_IB/3:128/0.602990179/0.0809117969; frozen_vs_IB/4:128/0.659303863/0.0843368788; start_vs_IB/1:128/0.608826474/0.0604493874; start_vs_IB/2:128/0.468199175/0.0740021668; start_vs_IB/3:128/0.559361006/0.0746128497; start_vs_IB/4:128/0.634281328/0.0802382864.

Gas: {"oog":0}.

### S09 — b27-jc03-s03-h60-p24

Source: j11-a/items/jc03-s03-h60-p24/result.json; SHA-256 4065949695a0dcff528f2f54144ae6bcef8efd6ae75834fd45da880f9912499d.

ea4047d3 START family. Initial mispacked harness corrected; feature scaling recovered from Solidity because params only said unchanged from IA. Selected generation 280/minute 40.10253425819974 with monitor +0.033086066694529986 SE 0.00732559194423055. Holdout positive but real frozen–START -0.0044877363978245155 SE 0.015150724432319582 does not confirm improvement. Monitor-selected point returned without holdout/real reselection. First IB chunk during training for diagnostics.

Code H11. Port H9. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.000686542046|
|B|true|512/0.544040078/0.0361890368; target 0.595|
|C|true|0.000648197185|

Speed (markets/s, workers named): {"markets_per_s":1441.0813615846969,"cpus":8,"processes":8,"timed_markets":6254592}.

Search: {"es_seed":20261323,"dim":219,"pop":24,"sigma0":0.3,"cap":".3 * .5 ** (search_minutes / 60)","gens":558,"min":79.8396109,"M":512,"train":[1304000000,1304285695],"selected":{"generation":280,"gain_over_START":0.0330860667,"minutes":40.1025343,"se":0.00732559194}}.

Replica seeds: [41030000,41038191]:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0.017328628/0.00417934404|0.0217792841/-0.00445065612|—/—|
|frozen_vs_IB|8192/0.627899973/0.00965257318|-0.370315923/0.998215895|—/—|
|start_vs_IB|8192/0.610571345/0.00977042302|-0.392095207/1.00266655|—/—|

Real seeds: [8119808,8120319].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|512/-0.0044877364/0.0151507244|—/—|-6.85991763/27.0235555|8119858:-1.52021181; 8120053:-1.50789609; 8119888:-1.20993959; 8120307:-1.20154792; 8120025:-1.18763502|
|frozen_vs_IB|512/0.686148612/0.0450124665|—/—|126.344253/-92.6351963|8120020:-2.06351694; 8120004:-1.9779421; 8119940:-1.70047895; 8120099:-1.64362431; 8119878:-1.42588327|
|start_vs_IB|512/0.690636349/0.0467243324|—/—|133.204171/-119.658752|8120020:-2.17738672; 8120004:-2.13756011; 8120099:-1.63395465; 8119811:-1.48303598; 8120273:-1.28708621|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_IB/1:128/0.672093142/0.0603449333; frozen_vs_IB/2:128/0.813426778/0.12208354; frozen_vs_IB/3:128/0.628778616/0.0862234332; frozen_vs_IB/4:128/0.630295912/0.0803083129; start_vs_IB/1:128/0.65258522/0.0618088831; start_vs_IB/2:128/0.810887607/0.131001178; start_vs_IB/3:128/0.619388207/0.0867521718; start_vs_IB/4:128/0.67968436/0.0803936053; frozen_vs_start/1:128/0.019507922/0.0226356395; frozen_vs_start/2:128/0.00253917103/0.0324473843; frozen_vs_start/3:128/0.00939040887/0.0313201522; frozen_vs_start/4:128/-0.0493884475/0.0335548781.

Gas: {}.

### S10 — b28-t03-block-a

Source: confirm-t03/items/t03-block-a/result.json; SHA-256 04c270ed6ae2614b77e1037b547516c94bff4b836d0ec928509d0f0217a6385a.

Confirmation only, T03 cb6d9455 vs IB. No port/replica/search/gates. Native arm64 engine build substituted for specified Linux wheel after executable-format failure 126; not byte-identical. Cargo.lock differed from recipe, so source similarity is not dependency equivalence. External check seeds 8588481/8588793 matched published differences within 1e-10, outside confirmation block and excluded. This supports reported results but does not certify wheel equivalence. No retail/arb edge split, only volumes. All block seeds reportedly once per arm.

Code {"IB":"301941829a44bb60977bb28181a5e4bfa6544f3696763f987054789b2611d686","CAND":"cb6d945511e1e10af78baa64c2a6cb25806f0998403ce6440654bfc3c70092d9"}. valid=true.

Real seeds [8700000,8701023]; n/mean_diff/se=1024/0.800458977/0.0259058565. Worst5 seed:diff: 8700653:-3.48317583; 8700380:-1.92709734; 8700460:-1.56735779; 8700002:-1.44802909; 8700435:-1.35168642.

Retail/arb Y-volume deltas: 280.907008 ± 10.1291101/-53.6149856 ± 6.5652969; PnL delta 0.79344196 ± 0.0327109505. Mean edges {"IB":525.9683261521303,"CAND":526.7687851287205,"vanilla_in_IB_arm":350.5938738158806,"vanilla_in_CAND_arm":349.57814317418894}.

Other paired findings: {"median":0.761940647276873,"share_positive":0.8681640625,"t_stat":30.898765170804726,"excluding_largest_gain_seed":{"seed":8700658,"n":1023,"mean_diff":0.7937971826116726,"se":0.025059137798260932},"best5":[[8700658,7.615474217148403],[8700388,5.130114092103099],[8700804,5.105919550321289],[8700661,4.427500837684647],[8700902,4.184480892642114]]}.

IB-retail-volume quartiles (label:n/mean/se): q1:256/0.886670351/0.0402500931; q2:256/0.783093826/0.0496601129; q3:256/0.753149659/0.0603781196; q4:256/0.778922071/0.0547426413.

Vanilla edge delta -1.01573064 ± 0.0437479109; mean Y volumes {"IB_retail":78142.4784282976,"CAND_retail":78423.38543659353,"IB_arb":24361.067336457458,"CAND_arb":24307.452350873304}.

Gas summary: {"IB":{"afterSwap_peak":237141,"afterInitialize_peak":243227,"oog":0,"other_failures":{"revert":0,"other_halt":0,"tx_error":0,"bad_return":0},"afterSwap_mean_gas":135235.09844268917},"CAND":{"afterSwap_peak":237789,"afterInitialize_peak":243234,"oog":0,"other_failures":{"revert":0,"other_halt":0,"tx_error":0,"bad_return":0},"afterSwap_mean_gas":135574.9387330649}}.

### S11 — b28-t03-block-b

Source: confirm-t03/items/t03-block-b/result.json; SHA-256 63b220cb48c1e04047f283f6d82134ad258d0aeca2cab155587e40053f9f7c5b.

Confirmation only, T03 cb6d9455 vs IB. Required Linux setup reportedly succeeded; no search/port/replica/parity gates. Sixteen 64-seed chunks; after chunk 0, three concurrent sequences, each policy still alone vs Vanilla. Counters reset per arm/chunk; validity requires no OOG in either role and strategy afterSwap <=245000, no afterInitialize limit. No edge split or leaderboard claim.

Code {"IB":"301941829a44bb60977bb28181a5e4bfa6544f3696763f987054789b2611d686","CAND":"cb6d945511e1e10af78baa64c2a6cb25806f0998403ce6440654bfc3c70092d9"}. valid=true.

Real seeds [8701024,8702047]; n/mean_diff/se=1024/0.772455789/0.0241099041. Worst5 seed:diff: 8701430:-2.15808795; 8701112:-2.05054008; 8701921:-1.96465622; 8701380:-1.67742688; 8701090:-1.51959212.

Retail/arb Y-volume deltas: 289.303175 ± 10.2108008/-44.1425277 ± 6.64485353; PnL delta 0.763304265 ± 0.0297647363. Mean edges {"IB":527.4335285796226,"CAND":528.2059843688204,"vanilla_vs_IB":351.9144923055467,"vanilla_vs_CAND":350.8782295829468}.

Other paired findings: {"median":0.769267957310376,"share_positive":0.85546875,"best5":[[8701496,4.143894590339073],[8701269,3.9195166709873774],[8701650,3.9178605416669825],[8701512,3.7903346798196367],[8701779,3.543727497752684]]}.

IB-retail-volume quartiles (label:n/mean/se): q1:256/0.760136809/0.0429837086; q2:256/0.781188007/0.043350709; q3:256/0.768758337/0.049229447; q4:256/0.779740005/0.0563593548.

Gas summary: {"IB":{"afterSwap_peak":237130,"afterInitialize_peak":243227,"oog":0,"other_failures":0},"CAND":{"afterSwap_peak":237689,"afterInitialize_peak":243234,"oog":0,"other_failures":0}}.

### S12 — b29-m01-s03-h60-p16

Source: t4-a/items/m01-s03-h60-p16/result.json; SHA-256 eed9b1128a9bf7981f9b63b8f44955441a4472b62802ef3bc157b69997f5ab61.

No search S=65.6; C/frozen comparisons null though code equals START. Bounded A and ten stress sequences reportedly pass. Real 224/1024: 8120848..8120879 and 8120944..8121135. Consumed/unreported 8120832..8120847 (decomposition failure), 8120880..8120943 (timeout). Seed 8120982 loses -21.210995215143328 vs IB. KIT edge decomposition relaxed 0.01 arb-volume assertion to recording, max error 0.02654745658946922; affects components, not engine total edge.

Code H1. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.0012218133|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|—|—|

Speed (markets/s, workers named): {"markets_per_s":65.6,"single_process_markets_per_s":88.4,"procs":16}.

Search: {"gens":0,"min":0,"M":null,"status":"skipped: S<100"}.

Replica holdout: unmeasured. not run: S<100 branch is step 6 only

Real seeds: —.

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|start_vs_IB|224/0.59203009/0.118466465|-0.417621804/1.00965189|242.198472/—|8120982:-21.2109952; 8120946:-1.99217881; 8120856:-1.58954455; 8120869:-1.15557242; 8120998:-1.10977978|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): start_vs_IB/1:56/0.685329413/—; start_vs_IB/2:56/0.251311106/—; start_vs_IB/3:56/0.84268157/—; start_vs_IB/4:56/0.588798271/—.

start_vs_IB retained seed ranges: [8120848..8120879,8120944..8121135].

Gas: {"afterSwap_peak":{"start":232588,"IB":232014,"frozen":null},"oog":0,"reverts":0}.

Stress diagnostic: max fee error 0.00020589728 bps; reverts 0.

### S13 — b29-m02-s03-h90-p16

Source: t4-a/items/m02-s03-h90-p16/result.json; SHA-256 933c404e0c689ed01ccba5ba30f07afb4cc476041dd85ff9b0c0316d1da151fc.

359 generations, selected gen 40 monitor +0.007854987222231102, within about one SE. Note 'mean never improved' conflicts literally with positive monitor; 'nothing beat START' means no convincing improvement. Returned changed params/code follow monitor rule, but worker recommends unchanged START for continuation. Holdout negative; real tied. IB probe 8000000..8000007. START chunk 8122048 killed before persistence and rerun on same 64 seeds, violating literal never-reused execution despite unique retained outputs.

Code H12. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.000404270911|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|true|0.000321787121|

Speed (markets/s, workers named): {"gateB_markets_per_s":473.71195135048754,"search_pool_markets_per_s_first_gen":637.7560580465632,"search_pool_markets_per_s_median":2664.7383385400567,"search_pool_markets_per_s_max":5370.895242204053}.

Search: {"es_seed":20261402,"pop":16,"sigma0":0.3,"cap":"0.3*0.5^(minutes/90) applied each generation","gens":359,"min":33.23,"M":512,"train":[1310000000,1310183807],"best":{"gen":40,"src":"es_mean","monitor_gain_vs_start":0.00785498722}}.

Replica seeds: [41050000,41058191]:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/-0.0140268335/0.00399041437|—/—|—/—|
|frozen_vs_IB|8192/0.757088323/0.00978963238|—/—|—/—|
|start_vs_IB|8192/0.771115156/0.00979618408|—/—|—/—|

Real seeds: [8121856,8122367].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|512/0.00637997047/0.01442547|—/—|-145.845565/-136.91769|8121899:-1.45227049; 8122098:-1.30132437; 8122196:-0.970543222; 8122342:-0.968264914; 8122266:-0.900756376|
|frozen_vs_IB|512/0.819068287/0.037760906|—/—|148.808503/-176.741401|8121899:-2.27936993; 8122353:-1.55666011; 8122281:-1.23669992; 8121920:-1.16049518; 8122262:-0.886513603|
|start_vs_IB|512/0.812688317/0.0375673333|—/—|294.654068/-39.8237107|8122353:-1.86027573; 8121968:-1.74374375; 8122281:-1.61520126; 8122183:-1.41359317; 8122037:-1.06474645|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_start/1:128/0.00261921635/0.0238871315; frozen_vs_start/2:128/-0.0183762512/0.0267213417; frozen_vs_start/3:128/0.046149523/0.0311248654; frozen_vs_start/4:128/-0.00487260626/0.0328237599; frozen_vs_IB/1:128/0.748337595/0.078385579; frozen_vs_IB/2:128/0.894793269/0.0772985842; frozen_vs_IB/3:128/0.763723714/0.0671702159; frozen_vs_IB/4:128/0.86941857/0.0786706647; start_vs_IB/1:128/0.745718378/0.0749673791; start_vs_IB/2:128/0.913169521/0.0759459864; start_vs_IB/3:128/0.717574191/0.0708427994; start_vs_IB/4:128/0.874291176/0.078049289.

Gas: {"afterSwap_peak":{"frozen":238345,"start":238336,"ib":237772},"oog":{"frozen":0,"start":0,"ib":0},"reverts":{"frozen":0,"start":0,"ib":0}}.

Stress diagnostic: max fee error 3.52121448 bps; reverts 105:1359.

### S14 — b29-m03-s05-h60-p16

Source: t4-a/items/m03-s05-h60-p16/result.json; SHA-256 d3f7feef95ba5076e82bcce8ec7feee2eba50b30223d18e346361b2aaf23019c.

137 generations at 221.7/M=256 fail goal's >=300 rule despite claimed item override. Every monitored mean lost; START unchanged. Search minutes include failed-launch overhead (~ten minutes lost). Unbounded stress panicked 0x12/0x11 at X reserves ~1e-5..1e-8; no OOG/revert in 768 real markets. Real edge components reconstructed, total edge/volumes from engine.

Code H1. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.0012218133|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|true|0.0012218133|

Speed (markets/s, workers named): {"markets_per_s":221.7,"cpus":4}.

Search: {"gens":137,"min":59.34,"M":256,"pop":16,"es_seed":20261403,"sigma0":0.5,"train":[1312000000,1312035071],"best":{"gen":0,"monitor_diff":0.0,"is_START":true}}.

Replica seeds: [41050000,41058191]:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0/0|0/0|—/—|
|frozen_vs_IB|8192/0.771115156/0.00979618408|-0.228235903/0.99935106|—/—|
|start_vs_IB|same as frozen_vs_IB|same|same|

Real seeds: [8122880,8123647].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|768/0/0|—/—|—/—|unavailable|
|frozen_vs_IB|768/0.799544182/0.0316322574|-0.208570372/1.00811455|286.799797/-43.0600936|8123487:-3.6502694; 8123585:-2.00308111; 8123305:-1.60095928; 8122955:-1.58391415; 8122978:-1.34244158|
|start_vs_IB|same as frozen_vs_IB|same|same|same|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_IB/1:192/0.710343874/0.0411343834; frozen_vs_IB/2:192/0.898827107/0.0555420812; frozen_vs_IB/3:192/0.780314683/0.0793162082; frozen_vs_IB/4:192/0.808691064/0.0700628227.

Gas: {"afterSwap_peak":238235,"oog":0}.

Stress diagnostic: max fee error 0.000350262837 bps; reverts 1008:1182,1009:1463.

### S15 — b29-m04-s04-h90-p16

Source: t4-a/items/m04-s04-h90-p16/result.json; SHA-256 79e612b9bd8d81c67f7944e07d4073f9f3131a7abcb30561bc88c735857c51fe.

No monitor beat START, unchanged. First restart omitted NumPy RNG; gens 26–54 not reproducible from seed alone, later restores preserve RNG. Both-token-capped A/C pass; stricter A ran early during search after X-only A. Ungated stress nine completed, sequence 1 stopped before callback 1081 for invalid next trade, spot ~3.58 million. Separate 15%-X stress panic 0x12 at sequence 1/callback 1758, gas 62671. Overall valid=false because preliminary stress reverted, while real.valid=true: validity scope differs from other workers. Real split used actual initial fee and arb-volume check. B arrays recovered after serialization error without replay.

Code H1. Port H4. valid=false.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.00103038478|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|true|0.00103038478|

Speed (markets/s, workers named): {"markets_per_second":468.43567260744487,"workers":8}.

Search: {"gens":349,"min":80.2006347,"M":512}.

Replica seeds: [41050000,41058191]:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0/0|0/0|0/—|
|frozen_vs_IB|8192/0.771115156/0.00979618408|-0.228235903/0.99935106|290.76647/—|
|start_vs_IB|same as frozen_vs_IB|same|same|

Real seeds: [8123904,8124415].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|512/0/0|0/0|0/0|8123904:0; 8123905:0; 8123906:0; 8123907:0; 8123908:0|
|frozen_vs_IB|512/0.80534769/0.033254984|-0.199067915/1.00441561|277.658359/-39.4749238|8123905:-1.64279929; 8124203:-1.30450821; 8124197:-1.11913298; 8124050:-0.965162046; 8123930:-0.955652079|
|start_vs_IB|same as frozen_vs_IB|same|same|same|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_start/1:128/0/0; frozen_vs_start/2:128/0/0; frozen_vs_start/3:128/0/0; frozen_vs_start/4:128/0/0; frozen_vs_IB/1:128/0.755266253/0.057873916; frozen_vs_IB/2:128/0.839936572/0.0595587613; frozen_vs_IB/3:128/0.809879676/0.0716373915; frozen_vs_IB/4:128/0.81630826/0.0757625763.

Gas: {"afterSwap_peak":237453,"oog":0}.

Stress diagnostic: max fee error 0.00053571285 bps; reverts —.

### S16 — b29-m05-s02-h90-p16

Source: t4-a/items/m05-s02-h90-p16/result.json; SHA-256 aaa3195086991e7a82ec10115a4ba7bd1d7905c2e605edd9257008365e412c41.

Changed gen 162/177 point; monitor +0.022665627199444316 then ~+0.0136 at gen177. Holdout positive, real +0.01470904157897679 SE 0.014246205571251808 within noise. Literal-only freeze; anchor cap 715 to720, baseline byte-identical. No real edge decomposition because KIT assumes 4 bps initial fee unlike START; volumes direct engine. B slightly differs from common repeated figure; retain its own value. Ten no-revert stress sequences use different RNG and do not refute prior panic.

Code H13. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.0012218133|
|B|true|1024/0.769212938/0.0269134747; target 0.819|
|C|true|0.000485364537|

Speed (markets/s, workers named): {"markets_per_s":371.4784059149651,"ib_markets_per_s":476.26537100384195,"procs":4,"cpus":4}.

Search: {"gens":177,"min":61.56,"M":512,"pop":16,"es_seed":20261405,"sigma0":0.2,"cap":"0.2*0.5^(minutes/90)","final_sigma":0.124635719,"train":"1316000000..1316090623 (gen g uses 1316000000+512g..+511, each once)","best":{"generation":162,"monitor_mean_edge":529.039686,"monitor_gain_vs_start":0.0226656272}}.

Replica seeds: 41050000..41058191:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0.0155116719/0.00404354496|—/—|—/—|
|frozen_vs_IB|8192/0.786629212/0.00971083055|—/—|—/—|
|start_vs_IB|8192/0.771117541/0.00979614438|—/—|—/—|

Real seeds: 8124928..8125439.

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|512/0.0147090416/0.0142462056|—/—|-58.760769/-23.2109722|8125377:-1.24633574; 8125154:-1.17840861; 8125196:-0.859048667; 8125341:-0.8448878; 8124971:-0.82448523|
|frozen_vs_IB|512/0.813341521/0.039421066|—/—|240.013857/-68.4827922|8125302:-2.33353158; 8125104:-1.97998948; 8124946:-1.72212142; 8125335:-1.54117459; 8124956:-1.53959322|
|start_vs_IB|512/0.79863248/0.0401485286|—/—|298.774626/-45.27182|8124946:-1.80732409; 8125302:-1.63888084; 8125128:-1.57946431; 8125104:-1.5627355; 8125049:-1.54263334|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_start/1:128/0.00100152583/—; frozen_vs_start/2:128/0.0348666321/—; frozen_vs_start/3:128/0.0553579958/—; frozen_vs_start/4:128/-0.0323899875/—; frozen_vs_IB/1:128/0.792496727/—; frozen_vs_IB/2:128/0.937426177/—; frozen_vs_IB/3:128/0.717864114/—; frozen_vs_IB/4:128/0.805579067/—; start_vs_IB/1:128/0.791495201/—; start_vs_IB/2:128/0.902559545/—; start_vs_IB/3:128/0.662506118/—; start_vs_IB/4:128/0.837969054/—.

Gas: {"afterSwap_peak":{"frozen":237695,"start":237701,"IB":237142},"oog":0}.

Stress diagnostic: max fee error 0.000184640324 bps; reverts 0.

### S17 — b30-m06-s03-h60-p24

Source: t4-b/items/m06-s03-h60-p24/result.json; SHA-256 2a0bc0323223bae4b074e3ec286837a8e43b4bc54be74879945e82d9aeb33eb9.

Blocked A, no training/confirmation. 42 completed sequences max 0.0013129291984159686 bps; seq42/RNG440145/callback1223 panic0x12, X0.06663318548727094/Y15007537.050604152, gas62659. Harness15%-X cap, no bounded spot; seq43–49 unattempted. Provenance conflict: failure.reproduced=true and blocked text say reproduced, but final note explicitly says failure data retained from fetched START artifact. Treat as inherited/reported, not established independent local replication. B reportedly concurrent before failure known; no C/holdout/real/gas/OOG measurement.

Code H1. Port H4. valid=false.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|false|0.0013129292|
|B|true|1024/0.76922146/0.0269129589; target —|
|C|—|—|

Speed (markets/s, workers named): {"markets_per_s":601.2506805918943,"cpus":5}.

Search: {"gens":0,"min":0,"M":512,"status":"not run: Gate A failed"}.

Replica holdout: unmeasured. not run: Gate A failed

Real evaluation: unmeasured. not run: Gate A failed

Gas: {"afterSwap_peak":{"frozen":null,"start":null,"IB":null},"oog":null,"status":"real engine not run; Gate A observed a non-OOG revert"}.

### S18 — b30-m07-s05-h45-p16

Source: t4-b/items/m07-s05-h45-p16/result.json; SHA-256 d0eaf0af43501e805e2a8a432749190c824fc3993d69555c5a5088d2a6be43ae.

322 generations; all monitored means below START (-.009 to-.123, final ~-.069 SE.011), retain z=0. Decoder added <=1-1e-6 clamps to decay/PHAT-alpha constants because Solidity subtracts from WAD; frozen START unaffected. Bounded A pass; stress seq1008 callback1449 panic0x12, discrepancy1.6460860000000466 bps at extreme spot. IB chunk8127232 timed out then reran same seeds. Real arb-volume null because not saved; do not infer from reconstructed arb edge. START/frozen identity substitutes separate arm.

Code H1. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.0012218133|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|true|0.0012218133|

Speed (markets/s, workers named): {"markets_per_s":603.5966229194164,"procs":8}.

Search: {"gens":322,"min":71.28,"M":512,"es_seed":20261407,"pop":16,"sigma0":0.5,"cap":"0.5 * 0.5^(minutes/45)","final_sigma":0.167032559,"train":[1320000000,1320164863]}.

Replica seeds: [41050000,41058191]:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0/0|—/—|—/—|
|frozen_vs_IB|8192/0.771115156/0.00979618408|-0.228235903/0.99935106|—/—|
|start_vs_IB|same as frozen_vs_IB|same|same|

Real seeds: [8126976,8127487].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|512/0/0|—/—|—/—|unavailable|
|frozen_vs_IB|512/0.846817057/0.0590158|-0.135234048/0.982051105|307.46873/—|8127100:-1.97036869; 8127419:-1.73834455; 8127385:-0.988330647; 8127401:-0.929842509; 8127072:-0.928143236|
|start_vs_IB|512/0.846817057/0.0590158|—/—|—/—|8127100:-1.97036869; 8127419:-1.73834455; 8127385:-0.988330647; 8127401:-0.929842509; 8127072:-0.928143236|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_IB/Q1_low:128/0.72327074/0.0558206059; frozen_vs_IB/Q2:128/0.809501037/0.0732465871; frozen_vs_IB/Q3:128/0.824355948/0.0963498696; frozen_vs_IB/Q4_high:128/1.0301405/0.194655785.

Gas: {"afterSwap_peak":237441,"oog":0}.

### S19 — b30-m08-s03-h90-p24

Source: t4-b/items/m08-s03-h90-p24/result.json; SHA-256 0291a847d1d09d1d45f22f1458ca7988a6dcf3d42776c357067fb4c01a333cdb.

Retained85 generations/45.15957590341568 minutes; discarded57/31.28743494749069 omitted proposal RNG on early resumes. Corrected run starts first unused seed1322029184. Total142 generations/76.44701085090637 minutes; only corrected eligible, no mean beat START. Bounded A/C pass, stress fails. First frozen real64 chunk diagnostics failed on KIT initial fee .0004 vs START .000394715277459523; rows/gas alias separately measured identical START, not independent retained frozen measurements.448 overlapping retained rows equal. Next IB persisted core fields recovered without replay. Via-IR built five files in9.01s after missing forge-std and stack-depth failures. Example tests not run, Slither absent, two Aderyn attempts failed before analysis due read-only filesystem; no static-analysis assurance.

Code H1. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.00106604284|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|true|0.00106604284|
|checkpoint_resume|true|—|

Speed (markets/s, workers named): {"initial_warmed_markets_per_second":336.82303708312634,"initial_workers":5,"worker_benchmarks":[{"workers":5,"markets_per_second":383.9722844102132},{"workers":8,"markets_per_second":393.8914623123247},{"workers":16,"markets_per_second":403.0329796093788},{"workers":24,"markets_per_second":366.1902998477677}]}.

Search: {"gens":85,"min":45.1595759,"M":512,"ES_seed":20261408,"pop":24,"sigma0":0.3,"cap":".3 * 0.5 ** (search_minutes/90)","dim":219,"first_seed":1322029184,"last_seed":1322072703,"unique_training_markets":43520,"proposal_market_evaluations":1044480,"bestMonitorGain":0.0}.

Replica seeds: [41050000,41058191]:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0/0|0/0|0/—|
|frozen_vs_IB|8192/0.771115156/0.00979618408|-0.228235903/0.99935106|290.76647/—|
|start_vs_IB|same as frozen_vs_IB|same|same|

Real seeds: [8128000,8128511].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|512/0/0|—/—|0/0|8128000:0; 8128001:0; 8128002:0; 8128003:0; 8128004:0|
|frozen_vs_IB|512/0.731796628/0.0407310762|—/—|282.9766/-55.0911618|8128487:-8.90574346; 8128125:-3.58608627; 8128431:-1.55069535; 8128421:-1.43468761; 8128231:-1.35077197|
|start_vs_IB|same as frozen_vs_IB|same|same|same|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_start/1:128/0/0; frozen_vs_start/2:128/0/0; frozen_vs_start/3:128/0/0; frozen_vs_start/4:128/0/0; frozen_vs_IB/1:128/0.683367562/0.0582009584; frozen_vs_IB/2:128/0.62780304/0.0994213901; frozen_vs_IB/3:128/0.816121124/0.0699081392; frozen_vs_IB/4:128/0.799894784/0.0913556157.

Gas: {"afterSwap_peak":237473,"oog":0}.

Stress diagnostic: max fee error 2.152891 bps; reverts 53:1251,55:890,58:1461.

### S20 — b30-m09-s04-h60-p16

Source: t4-b/items/m09-s04-h60-p16/result.json; SHA-256 8ad9016a81ffdb1056e14db231835bd75a69ee1e50d3a09c3f12246d75d9e378.

No search S85.4, START unchanged. Bounded A passes; stress reproduces reported panic RNG440145, ungated. C not run. Real384/1024, seeds8129024..8129407;8129408..8130047 unused. Real deltas direct Y volumes, no edge split.

Code H1. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.0012218133|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|—|—|

Speed (markets/s, workers named): {"markets_per_s":85.4,"single_process_markets_per_s":91.6}.

Search: {"gens":0,"min":0,"M":null,"status":"not run: S = ~85 markets/s < 100 (item override rule)"}.

Replica holdout: unmeasured. not run: S < 100 branch runs only step 6

Real seeds: [8129024,8129407].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|start_vs_IB|384/0.753846851/0.04180254|—/—|286.566668/-45.8741273|8129226:-1.12324095; 8129393:-1.05751304; 8129189:-1.05542483; 8129285:-0.976417397; 8129102:-0.742151873|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): start_vs_IB/1:96/0.638373799/—; start_vs_IB/2:96/0.845424091/—; start_vs_IB/3:96/0.70303879/—; start_vs_IB/4:96/0.828550723/—.

Gas: {"afterSwap_peak":{"start":238235,"IB":234885,"frozen":null},"oog":0}.

Stress diagnostic: 10 sequences; max fee error 4.12001203 bps; failure seq:callback 42:1223.

### S21 — b30-m10-s06-h60-p16

Source: t4-b/items/m10-s06-h60-p16/result.json; SHA-256 7a340c8dad05e8a0df121348a7defc9cd9c77c6d88172e27ae479d0f151cc490.

260 generations/4160 proposals; no mean beat START, best mean-.05090118003201302. Unchanged START/frozen, differences zero by identity. First19-generation segment RNG reconstructed from Gaussian draws, subsequent checkpoints persist RNG. Corrected net-of-fee stress model all ten pass; exploratory gross-update stress failures discarded, not evidence on corrected model. Initial IB8130048..8130111 lost before persistence: BASE_FEE=4*BPS parsed as4wei, no replay. Retained IB comparisons8130112..8130559 only448; identity/START comparisons may have larger n. blocked records missing block despite valid=true on retained calls/gates. Gas only persisted records. Early real diagnostics overlapped training, not selection. Edge split reconstructed from traces with arb-volume check.

Code H1. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.000728968402|
|B|true|1024/0.76922146/0.0269129589; target —|
|C|true|0.000728968402|

Speed (markets/s, workers named): {"replica_markets_per_second":516.0846593801841,"processes":8}.

Search: {"best_gain":0.0,"gens":260,"min":77.2706906,"M":512,"pop":16,"dim":219,"seed":20261410,"train0":1326000000,"trainEndExclusive":1326133120,"sigma0":0.6,"sigma_cap_halflife_minutes":60}.

Replica seeds: 41050000..41058191 (from source notes):

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0/0|0/0|0/—|
|frozen_vs_IB|8192/0.771115156/0.00979618408|-0.228235903/0.99935106|290.76647/—|
|start_vs_IB|same as frozen_vs_IB|same|same|

Real seeds: —.

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|512/0/0|0/0|0/0|0; 0; 0; 0; 0|
|frozen_vs_IB|448/0.792273133/0.0410444558|-0.213788056/1.00606119|289.858401/-45.9303812|-2.62260877; -2.23477345; -1.97252709; -1.59432677; -1.43491141|
|start_vs_IB|same as frozen_vs_IB|same|same|same|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_start/1:112/0/0; frozen_vs_start/2:112/0/0; frozen_vs_start/3:112/0/0; frozen_vs_start/4:112/0/0; frozen_vs_IB/1:112/0.81498694/0.0786533727; frozen_vs_IB/2:112/0.594532658/0.0713997291; frozen_vs_IB/3:112/0.85376859/0.0801939862; frozen_vs_IB/4:112/0.905804347/0.094414144.

Gas: {"afterSwap_peak":235575,"oog":0,"reverts":0}.

Stress diagnostic: 10 sequences; max fee error 0.00186086325 bps; failure seq:callback none.

### S22 — b31-m11-s03-h60-p16

Source: t4-c/items/m11-s03-h60-p16/result.json; SHA-256 2912f3af8eb72d817ac6b319604c628170c50b9cde0cc3c1adccfc6aa424544c.

241 generations; no mean beat START. From gen32 decoder caps *_DECAY at.9999: LAMBDA_DECAY>1 wasted ~one fifth first31 generations (~-97 edge/market); START unaffected. Final mean monitor-.01472 SE.00865 at gen230, not better frozen point. Search stopped early for slower evaluation/confirmation budget. Stress metadata says amount cap null but note says6%: unresolved description discrepancy. A/C pass, real zero difference by code identity. IB anchor noted (674,.7127,1.2952). Full restart vector only in input.

Code H1. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|6.18716973e-05|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|true|6.18716973e-05|

Speed (markets/s, workers named): {"markets_per_s":673.091187962698,"cpus":4}.

Search: {"es_seed":20261411,"pop":16,"sigma0":0.3,"cap":"0.3 * 0.5^(search_minutes/60)","gens":241,"min":55.73,"M":512,"train":[1328000000,1328123391],"best":{"gain_vs_start_monitor":0.0,"se":0.0,"gen":0,"minutes":0.0},"sigma_final":0.158290374}.

Replica seeds: [41050000,41058191]:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0/0|0/0|0/—|
|frozen_vs_IB|8192/0.771115156/0.00979618408|-0.228235903/0.99935106|290.76647/—|
|start_vs_IB|same as frozen_vs_IB|same|same|

Real seeds: [8131072,8131583].

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|512/0/0|—/—|0/0|8131072:0; 8131073:0; 8131074:0; 8131075:0; 8131076:0|
|frozen_vs_IB|512/0.71900798/0.0451632851|—/—|282.6333/-51.8286787|8131480:-8.62530972; 8131393:-2.70534067; 8131238:-2.30343669; 8131174:-1.89698722; 8131215:-1.65405365|
|start_vs_IB|same as frozen_vs_IB|same|same|same|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_start/1:128/0/0; frozen_vs_start/2:128/0/0; frozen_vs_start/3:128/0/0; frozen_vs_start/4:128/0/0; frozen_vs_IB/1:128/0.709291987/0.0554813153; frozen_vs_IB/2:128/0.687194394/0.065128375; frozen_vs_IB/3:128/0.680048905/0.135320468; frozen_vs_IB/4:128/0.799496632/0.0844103137.

Gas: {"afterSwap_peak":{"frozen":237611,"start":237611,"IB":237052},"oog":{"frozen":0,"start":0,"IB":0},"frozen_submission_failures":{"afterSwap_peak":237611,"oog":0,"revert":0,"other_halt":0,"tx_error":0,"bad_return":0,"afterSwap_calls":5547061}}.

Stress diagnostic: max fee error 4.18952417e-05 bps; reverts none.

### S23 — b31-m12-s03-h90-p16

Source: t4-c/items/m12-s03-h90-p16/result.json; SHA-256 2381b4874e2ac6b125beece70503218ee4cbf429bfe443b8db11863be57284f4.

Replica-only after Linux setup executable-format exit126 on arm64. A/C,frozen Solidity,real,gas,validity null, not passed/zero.529 generations ran without EVM gates, outside goal's gates-first requirement; cannot promote as validated training. No monitor beat START (range-.0526..-.0088); final mean-.0224 SE.0089. Params unchanged, recommended over final mean. Small platform replica differences retained. Restart sigma.14960245429832852,next seed1330270848; full vector only in input.

Code —. Port H4. valid=—.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|—|—|
|B|true|1024/0.769114176/0.0269119347; target 0.819|
|C|—|—|

Speed (markets/s, workers named): {"markets_per_s":1265.9280593690885,"ib_markets_per_s":1701.8391874537638,"procs":10}.

Search: {"gens":529,"min":70.42,"M":512,"es_seed":20261412,"pop":16,"sigma0":0.3,"cap":"0.3*0.5^(search_minutes/90) applied before each ask","train":[1330000000,1330270847],"best":{"source":"START (no ES-mean monitor point beat START)","gain_vs_start":0.0,"gen":0}}.

Replica seeds: [41050000,41058191]:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0/0|0/0|—/—|
|frozen_vs_IB|8192/0.771094075/0.00979565912|-0.228234674/0.999328749|—/—|
|start_vs_IB|same as frozen_vs_IB|same|same|

Real evaluation: unmeasured. not run: replica-only branch (amm_sim_rs wheel is manylinux x86_64)

Gas: {"afterSwap_peak":null,"oog":null,"status":"not measured: replica-only branch"}.

### S24 — b31-m13-s05-h60-p16

Source: t4-c/items/m13-s05-h60-p16/result.json; SHA-256 f0f205f689be87c9c6299b9b3e400875180d6c0933cca02ba37ec0a72afe89f7.

48 generations S105.3/M256 fail >=300 goal despite claimed override. No monitor beat START, unchanged. Bounded A/ten stress pass; different generators/seeds cannot refute prior panic. C reuses A by identical bytes, not independent frozen rerun. Real192/512, START/frozen identity not separately simulated. Saved optimizer state mentioned but explicitly not delivered.

Code H1. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|5.3558574e-05|
|B|true|1024/0.76922146/0.0269129589; target 0.819|
|C|true|—|

Speed (markets/s, workers named): {"replica_markets_per_s":105.3,"real_engine_markets_per_s":0.42}.

Search: {"gens":48,"min":37.36,"M":256,"pop":16,"es_seed":20261413,"sigma0":0.5,"train":"1332000000..1332012287","final_sigma":0.327460689}.

Replica seeds: 41050000..41058191:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0/0|—/—|—/—|
|frozen_vs_IB|8192/0.771115156/0.00979618408|-0.228235903/0.99935106|—/—|
|start_vs_IB|same as frozen_vs_IB|same|same|

Real seeds: —.

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|192/0/0|—/—|—/—|unavailable|
|frozen_vs_IB|192/0.773563376/0.0494059286|—/—|265.11979/-53.1080315|-1.06031713; -0.65315775; -0.630111249; -0.578668553; -0.455374062|
|start_vs_IB|same as frozen_vs_IB|same|same|same|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_IB/1:48/0.690370298/—; frozen_vs_IB/2:48/0.734425299/—; frozen_vs_IB/3:48/0.667070549/—; frozen_vs_IB/4:48/1.00238736/—.

Gas: {"afterSwap_peak":{"frozen":235259,"start":235259,"IB":234712},"oog":0,"reverts":0}.

Stress diagnostic: max fee error 4.44740684e-05 bps; reverts 0.

### S25 — b31-m14-s04-h90-p16

Source: t4-c/items/m14-s04-h90-p16/result.json; SHA-256 81fb34129503a52fb24a3a5d691ab061caac8dd834bfc5ae01a6b8102b1f74cc.

205 generations; no mean beat START (monitor529.0170205766398; means~528.91..528.998), unchanged. Stress different seeds/generator does not refute panic. C fresh540000..540049. First frozen ran8134144..8134591 then postprocessing failed on KIT4bps assumption; same seeds replayed using BASE_FEE~3.947bps. Not used for selection, but violates literal never-reused real seeds. START comparison by identity; split uses corrected initial fee.

Code H1. Port H4. valid=true.

| Gate | Reported pass | Max fee error (bps) / calibration |
|---|---|---|
|A|true|0.000109866084|
|B|true|—/0.76922146/0.0269129589; target 0.819|
|C|true|0.00013170124|

Speed (markets/s, workers named): {"markets_per_s":445.23568875777283,"cpus":4}.

Search: {"gens":205,"min":62.87,"M":512,"pop":16,"es_seed":20261414,"sigma0":0.4,"cap":"0.4*0.5^(minutes/90)","final_sigma":0.246899826,"seeds":"1334000000..1334104959","dim":219,"best":{"gen":0,"monitor":529.017021,"is_start":true}}.

Replica seeds: 41050000..41058191:

| Pair | n/mean/se | Edge R/A | Volume R/A |
|---|---|---|---|
|frozen_vs_start|8192/0/0|—/—|—/—|
|frozen_vs_IB|8192/0.771115156/0.00979618408|—/—|—/—|
|start_vs_IB|same as frozen_vs_IB|same|same|

Real seeds: 8134144..8134655.

| Pair | n/mean/se | Edge R/A | Volume R/A | Worst5 seed:diff |
|---|---|---|---|---|
|frozen_vs_start|512/0/0|—/—|—/—|unavailable|
|frozen_vs_IB|512/0.80723927/0.034532381|-0.190156594/0.997395864|298.146181/-40.0148178|-1.6253849; -1.46173761; -1.36265596; -1.20148141; -1.15363902|
|start_vs_IB|same as frozen_vs_IB|same|same|same|

Quartiles by ascending IB retail volume (comparison/quartile:n/mean/se): frozen_vs_IB/1:128/0.784438864/0.066265679; frozen_vs_IB/2:128/0.748489284/0.0647429371; frozen_vs_IB/3:128/0.805355913/0.0646997891; frozen_vs_IB/4:128/0.890673019/0.0796082842.

Gas: {"afterSwap_peak":{"frozen":235500,"start":235500,"IB":234873},"oog":0,"non_ok_submission_calls":{"frozen":0,"IB":0}}.

Stress diagnostic: max fee error 4.14585628e-05 bps; reverts 0.

B outlier diagnostic: excluding seed 8588481 gives +0.769598621, SE 0.0269366342, n=1023; common replica retail/arb edge deltas are -0.209157449/+0.978378909. [S03–S06, S17, S22]

## Continuation and selection implications

The two START families and their supplied +0.595/+0.819 references cannot be added. Distinct real seed blocks prevent ranking changed policies by raw IB-relative means. [S02–S09,S13–S16]

S13 recommends START despite returning a changed point; S09's real incremental mean is negative. S23 lacks frozen Solidity; S24 lacks optimizer state. S22/S23 restart vectors are not validated improvements. Unchanged selection after training differs from the no-search outcomes. [S01,S03,S04,S07,S09,S12,S13,S17,S20,S22–S24]

Continuation requires the chosen input's params and matching code/port; this write-up cannot execute them. The actual artifacts still need this goal's HA calibration, speed, fee parity and gas/OOG checks, not uniformly established by these records.