# Merge 01 of batch r3-deepen-current: evidence table for ten candidates against incumbent B

## Summary

Ten items each developed one change to the incumbent B (code SHA-256 `bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae`, JSON.code at https://api.imd.fun/artifacts/db22aff7167a4798251ed3d20432509f30e511e4958799aba5b17b41bb39b4b6). All ten results are complete: each ran 200 paired tuning seeds (310000..310199) and 500 paired holdout seeds (410000..410499) in one serial process on the pinned simulator (https://github.com/danrobinson/prediction-market-challenge at b244530ee2a6247b5f930830eeca9dd950da0e7e), reported zero failed simulations and zero missing seeds, and named the same incumbent hash the brief gives. I verified that the embedded code of every item hashes to its stated `code_sha256`, and that every input file hashes to the artifact digest supplied with this merge.

No item is promoted. Every holdout is screening evidence on 500 seeds; the goal's rule requires a paired gain over twice its standard error on the untouched finalist seeds 900000..903999, and no item ran those (each says so in its reproduction notes).

Ranking by holdout paired difference (C minus B, mean edge per simulation, 500 seeds):

1. **retreat-crossing** (spread 3): +0.2410 (SE 0.0756), about 3.2 SE; s3 alone +0.8733 (SE 0.2670). Tuning was only +0.0534 (SE 0.0966). The only candidate whose screening clears 2 SE. It is the one to extend on thousands of independent seeds first.
2. **side-jump-budget** (spreads 2 and 3): +0.4990 (SE 0.3496), about 1.4 SE, with the largest per-seed variance in the batch. It shifts edge from retail (-0.726) to arbitrage (+1.225), which is the lever the goal names, but the selected multiplier had a negative tuning mean (-0.0195, SE 0.4248). Worth a replicate, not yet trustworthy.
3. **stock-entry-gate** (all spreads): +0.3000 (SE 0.4026), under 1 SE, driven by 22 of 500 pairs. s4 is negative (-0.2946, SE 0.3624).
4. **horizon-drift** (spread 3): +0.0660 (SE 0.0578) on holdout but -0.0431 (SE 0.0216) on tuning; signs disagree.
5. **grid-resolution**, **variance-floor**, **probe-attribution**: null within noise (holdout +0.0106, +0.0051 and -0.0173 with SEs 0.0323, 0.0261 and 0.0135).
6. **probe-extremes**: -0.0000697 (SE 0.0000178), a statistically clear but economically irrelevant loss.
7. **tail-quadrature**: -0.1185 (SE 0.1880), after +0.1937 (SE 0.1791) on tuning; sign flip.
8. **posterior-payoff**: -0.1075 (SE 0.0650), negative on both tuning and holdout.

Three items ("tail-quadrature", "probe-extremes", "probe-attribution") were flagged by the merge brief as citing one source each that did not hold when re-read. None of the ten inputs contains a sources list (two mention a `sources.json` that was not supplied), so I cannot say which claim each flagged source supported. Every paired number in those three results is stated to come from local simulation, so the flag bears on background claims, not on the measured differences. The three are null or negative anyway.

Two machines produced a B baseline that is not identical to the other eight: horizon-drift and probe-attribution both report B's tuning mean as 85.162644703073 against 85.16265316504199 elsewhere, and B's holdout mean as 104.39217582931519 against 104.4137531881468. Within each item the comparison is still paired on the same machine, so its own difference is valid, but B's edges are not byte-reproducible across environments and cross-item comparisons of raw edge levels should not be made at the sixth decimal. The section "Comparator mismatches" lists the figures.

## Hash verification

| item | code_sha256 (candidate) | code bytes verified | incumbent matches brief | input artifact sha256 | complete |
|---|---|---|---|---|---|
| retreat-crossing | b5158a963a479a0919146dc0d56a347b68ddc2d66dd7c467b48473baef8997cf | yes | yes | 459d3ec15ee1323593c23f1933500cb42ac662c51e5e2aa89f9b1befee3640d1 | yes |
| side-jump-budget | 8c249a7d2ea89675599b3605807a195ee3c960f7db34650935c1f158ab210f14 | yes | yes | a53f876973198c8c1f235f17d201084f6ba7a97c6ad758b0b1229192ff731420 | yes |
| stock-entry-gate | 682cdc9b5eb7cb7833d95b59f03abb5b68d18e7eb6e37761b0dc2f0403ec708d | yes | yes | 6846a33e73fa0c4685a2a7e5072ec59ba300f454ab95612e9ab7a0eade8274fe | yes |
| horizon-drift | 951a669194e7b87eb54ae2796676f7db52707ee7ecbfe2d3c7b1fb284334a25e | yes | yes | 4838d029fa2e454634bfdb47f0e5cb17bed8c690f227d165df619da3a95f56e4 | yes |
| variance-floor | af654263059b776567f552d371113020e5c9a37f922fdb8fa528de9fac62aa3a | yes | yes | c762056de359e467be3b18327480640b00074a3846dbaa27644be28cd11e1019 | yes |
| grid-resolution | 22806a5844bd249e9a7b89b2b34b49292f7553e56a37c5781861bc9d30f4f36b | yes | yes | 91390b3932027df0bfc5f1130c0221055bcab2aebb6da36794b69909209e9a01 | yes |
| probe-attribution | ff72a56eaa8f77734cabf4d902a2b12fdbe8a9dd71bfee5bff147842cccf1902 | yes | yes | 23dc453b7a19305f3bf8fdd996ccb1254aec80bf1758b00c60d9b4071474ffc1 | yes |
| tail-quadrature | 851f2f2d5b9a5bad8f140c635ac97a901ea99b083d95291bef447c89678b0849 | yes | yes | 27431b07fb1bf1dbc667c1271dfda472459232cb96b864bee872488278f6e1c4 | yes |
| probe-extremes | c2c076dd69826effc18840314068a1155792c93416d565f7e6a82641deca02c9 | yes | yes | 5e00c9698814138ab644d73c7f3697317ca50a60a15208b7083d4c8d0e5fa813 | yes |
| posterior-payoff | 94d8fd887742b27444643264bc7309a241c377abfc7b9d0fa3d8a11e42d85f68 | yes | yes | 0fbc810ac55efec9ab4437c06c1a5256754aaf95e126a2302d95b424ae269529 | yes |


Code URL for each row is `https://api.imd.fun/artifacts/<input artifact sha256>`; the candidate's full code is the `code` field of that JSON, so no code is copied here. The hash of each input file was recomputed locally and matched the digest supplied with this merge before anything was read from it.

## Holdout screening, paired difference per spread (seeds 410000..410499)

Values are mean C minus B per simulation with the standard error in parentheses and z = mean/SE in brackets. Displayed values are rounded to four decimals for reading only; the JSON table below carries the exact strings. "untouched" means the item's guard left that spread byte-identical to B (difference exactly zero on every seed). No pooling across items: each row is a different policy.

| item | targets | overall mean (SE) [z] | s1 | s2 | s3 | s4 |
|---|---|---|---|---|---|---|
| retreat-crossing | 3 | +0.2410 (SE 0.0756) [+3.19] n=500 | 0 (untouched, n=115) | 0 (untouched, n=116) | +0.8733 (SE 0.2670) [+3.27] n=138 | 0 (untouched, n=131) |
| side-jump-budget | 2,3 | +0.4990 (SE 0.3496) [+1.43] n=500 | 0 (untouched, n=115) | +0.9798 (SE 0.9756) [+1.00] n=116 | +0.9842 (SE 0.9671) [+1.02] n=138 | 0 (untouched, n=131) |
| stock-entry-gate | 1,2,3,4 | +0.3000 (SE 0.4026) [+0.75] n=500 | +1.0758 (SE 1.1028) [+0.98] n=115 | +0.5255 (SE 1.2840) [+0.41] n=116 | +0.0284 (SE 0.0845) [+0.34] n=138 | -0.2946 (SE 0.3624) [-0.81] n=131 |
| horizon-drift | 3 | +0.0660 (SE 0.0578) [+1.14] n=500 | 0 (untouched, n=115) | 0 (untouched, n=116) | +0.2392 (SE 0.2092) [+1.14] n=138 | 0 (untouched, n=131) |
| variance-floor | 2 | +0.0051 (SE 0.0261) [+0.19] n=500 | 0 (untouched, n=115) | +0.0218 (SE 0.1127) [+0.19] n=116 | 0 (untouched, n=138) | 0 (untouched, n=131) |
| grid-resolution | 2 | +0.0106 (SE 0.0323) [+0.33] n=500 | 0 (untouched, n=115) | +0.0458 (SE 0.1396) [+0.33] n=116 | 0 (untouched, n=138) | 0 (untouched, n=131) |
| probe-attribution | 2 | -0.0173 (SE 0.0135) [-1.28] n=500 | 0 (untouched, n=115) | -0.0745 (SE 0.0580) [-1.28] n=116 | 0 (untouched, n=138) | 0 (untouched, n=131) |
| tail-quadrature | 2 | -0.1185 (SE 0.1880) [-0.63] n=500 | 0 (untouched, n=115) | -0.5110 (SE 0.8120) [-0.63] n=116 | 0 (untouched, n=138) | 0 (untouched, n=131) |
| probe-extremes | 3 | -6.97e-05 (SE 1.78e-05) [-3.91] n=500 | 0 (untouched, n=115) | 0 (untouched, n=116) | -0.000253 (SE 6.21e-05) [-4.07] n=138 | 0 (untouched, n=131) |
| posterior-payoff | 3 | -0.1075 (SE 0.0650) [-1.65] n=500 | 0 (untouched, n=115) | 0 (untouched, n=116) | -0.3895 (SE 0.2344) [-1.66] n=138 | 0 (untouched, n=131) |


## Tuning, paired difference per spread (seeds 310000..310199, selected variant)

| item | targets | overall mean (SE) [z] | s1 | s2 | s3 | s4 |
|---|---|---|---|---|---|---|
| retreat-crossing | 3 | +0.0534 (SE 0.0966) [+0.55] n=200 | 0 (untouched, n=47) | 0 (untouched, n=57) | +0.2017 (SE 0.3664) [+0.55] n=53 | 0 (untouched, n=43) |
| side-jump-budget | 2,3 | -0.0195 (SE 0.4248) [-0.05] n=200 | 0 (untouched, n=47) | -0.9269 (SE 0.4686) [-1.98] n=57 | +0.9232 (SE 1.5222) [+0.61] n=53 | 0 (untouched, n=43) |
| stock-entry-gate | 1,2,3,4 | +0.0135 (SE 0.2444) [+0.06] n=200 | -0.4114 (SE 0.3206) [-1.28] n=47 | +0.3650 (SE 0.8185) [+0.45] n=57 | +0.0233 (SE 0.0224) [+1.04] n=53 | 0 (untouched, n=43) |
| horizon-drift | 3 | -0.0431 (SE 0.0216) [-2.00] n=200 | 0 (untouched, n=47) | 0 (untouched, n=57) | -0.1627 (SE 0.0797) [-2.04] n=53 | 0 (untouched, n=43) |
| variance-floor | 2 | +0.0382 (SE 0.0349) [+1.09] n=200 | 0 (untouched, n=47) | +0.1341 (SE 0.1224) [+1.10] n=57 | 0 (untouched, n=53) | 0 (untouched, n=43) |
| grid-resolution | 2 | +6.56e-05 (SE 0.0178) [+0.00] n=200 | 0 (untouched, n=47) | +0.00023 (SE 0.0629) [+0.00] n=57 | 0 (untouched, n=53) | 0 (untouched, n=43) |
| probe-attribution | 2 | +0.0079 (SE 0.0068) [+1.16] n=200 | 0 (untouched, n=47) | +0.0279 (SE 0.0240) [+1.16] n=57 | 0 (untouched, n=53) | 0 (untouched, n=43) |
| tail-quadrature | 2 | +0.1937 (SE 0.1791) [+1.08] n=200 | 0 (untouched, n=47) | +0.6796 (SE 0.6277) [+1.08] n=57 | 0 (untouched, n=53) | 0 (untouched, n=43) |
| probe-extremes | 3 | -0.0004 (SE 0.000358) [-1.12] n=200 | 0 (untouched, n=47) | 0 (untouched, n=57) | -0.0015 (SE 0.0013) [-1.12] n=53 | 0 (untouched, n=43) |
| posterior-payoff | 3 | -0.0775 (SE 0.0642) [-1.21] n=200 | 0 (untouched, n=47) | 0 (untouched, n=57) | -0.2926 (SE 0.2415) [-1.21] n=53 | 0 (untouched, n=43) |


## Where the holdout difference comes from: retail versus arbitrage

Signed mean C minus B over all 500 holdout pairs, except posterior-payoff, which reports signed sums over pairs (they add to its sum_d) and is marked "sum_over_pairs".

| item | H retail_diff | H arb_diff | unit |
|---|---|---|---|
| retreat-crossing | +0.0855 | +0.1555 | mean_per_simulation |
| side-jump-budget | -0.7260 | +1.2249 | mean_per_simulation |
| stock-entry-gate | -0.1089 | +0.4089 | mean_per_simulation |
| horizon-drift | +0.0199 | +0.0461 | mean_per_simulation |
| variance-floor | -0.1834 | +0.1885 | mean_per_simulation |
| grid-resolution | +0.0217 | -0.0111 | mean_per_simulation |
| probe-attribution | -0.0273 | +0.0100 | mean_per_simulation |
| tail-quadrature | +0.2714 | -0.3899 | mean_per_simulation |
| probe-extremes | -7.95e-07 | -6.89e-05 | mean_per_simulation |
| posterior-payoff | -202.0576 | +148.3018 | sum_over_pairs |


Reading: side-jump-budget and stock-entry-gate buy arbitrage improvement with retail loss; retreat-crossing improves both; tail-quadrature and posterior-payoff gain retail and lose more arbitrage; the others are within rounding of zero.

## Comparator mismatches

- Majority baseline (eight items: retreat-crossing, side-jump-budget, stock-entry-gate, variance-floor, grid-resolution, tail-quadrature, probe-extremes, posterior-payoff): B tuning overall 85.16265316504199, s4 112.9910514314628; B holdout overall 104.4137531881468, s1 28.495154459731307, s2 97.76626247947758, s3 128.4054514027384, s4 151.67248160310686.
- Minority baseline (two items: horizon-drift, probe-attribution; one reports Python 3.10.12, the other does not state its version): B tuning overall 85.162644703073, s4 112.99101207346745; B holdout overall 104.39217582931519, s1 28.495164940043477, s2 97.76628686140518, s3 128.40565590283623, s4 151.5898790537267.
- The s4 holdout gap (about 0.083 per simulation over 131 seeds) is larger than floating-point rounding alone would suggest and is consistent with one or a few seeds diverging after a rounding-sensitive branch. Neither item claims a modified simulator; both report the pinned commit and verified incumbent bytes. The flag is recorded in both rows. It does not invalidate their paired differences but should be resolved before any cross-machine finalist run is accepted as evidence.
- Every row's mean_diff equals sum_d/n and every SE recomputes from (n, sum_d, sum_d2) to within floating tolerance; the per-row `arithmetic_check` field records this.

## Protocol notes and source flags

- Seed hygiene: all ten used 310000..310199 for tuning and 410000..410499 for holdout, with no overlap and no finalist seeds. variance-floor additionally ran a smoke test on seeds 1000..1007 and a post-holdout diagnostic on spread-2 seeds in 1000..1099, both outside every protected range and excluded from statistics. Nothing here is a violation.
- Interrupted holdouts: stock-entry-gate, tail-quadrature and probe-extremes each had the holdout process killed by a tool timeout partway and resumed the exact remaining suffix in a new process, reporting no seed repeated. Their SEs assume one run per seed, which the reports support but which this merge cannot independently verify.
- Variant selection: every multi-variant item selected on tuning only and froze before holdout. side-jump-budget selected the larger of two negative tuning means; posterior-payoff selected the less negative of two negative means. Those selections are legitimate but mean the holdout is the only positive evidence for side-jump-budget.
- Spread labels: nine items verified the strategy's own first-observed spread against the sampled regime. probe-extremes grouped by regime only and did not log the strategy's label; its nonzero differences appear only at s3, consistent with its guard.
- Unreported variant hashes: grid-resolution C16, tail-quadrature GH-16, probe-extremes C2 and posterior-payoff v1 are rejected variants whose hashes the items did not include; no digest is invented for them.
- Source flags from the merge brief: tail-quadrature, probe-extremes and probe-attribution each cite one source that did not hold on re-read. The inputs contain no source lists, so the affected passage is unknown here. Weigh any non-numerical claim in those three (for example tail-quadrature's statement about B's node second moment of 1.35, or probe-attribution's description of CancelOrder semantics) as unverified; the paired statistics are local measurements.
- posterior-payoff ran through the simulator CLI and paired rows by seed rather than calling the engine directly; the other nine called SimulationEngine with sample_config per seed. Both are the pinned simulator with defaults.

## What each mechanism changed (short)

- **retreat-crossing** (s3): two guarded insertions after the posterior transition; when the rival ask or bid moves exactly one tick outward, grid weight outside the crossed interval is multiplied by 0.5 and renormalized, each side independently. Probe placement, transition variance and sizing untouched.
- **side-jump-budget** (s2, s3): two three-line additions before qsize that charge the bid for upward jump mass and the ask for downward jump mass, scaled by 2/(1+s) and sqrt(min(1, 500/H)); existing qsize then applies.
- **stock-entry-gate** (s1..s4): in the penny-bank stock branch the upper gate on the probe tick is replaced by "competitor best bid at one tick".
- **horizon-drift** (s3): posterior support mapped to Gaussian scores, diffused one step with 3-point Gauss-Hermite, moment-matched for the AB valuation with H-1; jump component unchanged.
- **variance-floor** (s2): effective variance floor rises from 1e-08 to (0.5*sp)^2 where sp is the one-step diffusion sd; feeds AB sizing and probe range only.
- **grid-resolution** (s2): posterior bins per tick 8 to 32 with cap 48 to 192; about 15% slower per simulation.
- **probe-attribution** (s2): probe orders carry identities and a cancel ledger; fill-derived bounds use genuinely missing probes.
- **tail-quadrature** (s2): six ad-hoc jump nodes replaced by ten normalized Gauss-Hermite nodes.
- **probe-extremes** (s3): same probe counts relocated to the 0.1 and 0.9 posterior quantile ticks, clamped inside the rival spread.
- **posterior-payoff** (s3): diffusion AB computed as a posterior-cell-weighted average of expected losses instead of a Gaussian moment collapse.

## Recommendation for the next round (no promotion here)

- Extend retreat-crossing (hash `b5158a963a479a0919146dc0d56a347b68ddc2d66dd7c467b48473baef8997cf`) on thousands of fresh paired seeds; if it holds near +0.24 overall it is the first promotion candidate for the finalist seeds. Its gain is concentrated at s3, where it adds both retail and arbitrage edge.
- Replicate side-jump-budget (hash `8c249a7d2ea89675599b3605807a195ee3c960f7db34650935c1f158ab210f14`) on a larger independent sample before combining; its variance is too high for 500 seeds to settle, and its tuning sign was negative.
- Drop grid-resolution, probe-extremes, probe-attribution, tail-quadrature and posterior-payoff; their own authors reach the same conclusion. Posterior resolution, probe placement, probe attribution and quadrature accuracy are not levers at this stage.
- stock-entry-gate and horizon-drift are not worth seeds on their own; a restricted stock-entry-gate (s1 and s2 only) is an untested idea the item itself proposes.
- Resolve the cross-machine B baseline difference before accepting finalist runs from more than one machine.

## JSON evidence table

One row per item. Numbers are strings copied from the item results without rounding; `retail_arb_unit` says whether the retail and arbitrage differences are per-simulation means or sums over pairs. `tuning.selected` and `paired` carry overall and s1..s4 groups with n, sum_d, sum_d2, mean_diff, se, retail_diff, arb_diff and both strategies' mean edges. `tuning.variants_overall` lists every variant the item tried with its overall tuning mean, SE and n, each kept separate (no pooling). `code_url` is the input artifact that holds the exact code in its `code` field.

```json
{"merge":"r3-deepen-current / merges/01","incumbent":{"label":"B","code_sha256":"bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae","code_url":"https://api.imd.fun/artifacts/db22aff7167a4798251ed3d20432509f30e511e4958799aba5b17b41bb39b4b6"},"simulator":"https://github.com/danrobinson/prediction-market-challenge at b244530ee2a6247b5f930830eeca9dd950da0e7e, default ChallengeConfig/ParameterVariance, 2000 steps, one serial process","convention":"d = C.total_edge - B.total_edge per paired seed; mean_diff = sum_d/n; se = sample SD / sqrt(n); spreads s1..s4 are the competitor spread regime; all numbers are strings copied from the item results without rounding","promotion":"none; every row is screening evidence on 500 holdout seeds and the brief requires paired gain over twice the SE on finalist seeds 900000..903999, which no item ran","rows":[{"key":"retreat-crossing","idea":"Spread 3: a one-tick outward rival retreat softly reweights the tick posterior toward the crossed interval (strength 0.5 selected; 0.9 rejected on tuning).","mechanism_changed":"Two guarded insertions after the unchanged posterior transition: weights outside the crossed interval multiplied by (1-strength), then normalized; both sides independent.","target_spreads":["3"],"code_sha256":"b5158a963a479a0919146dc0d56a347b68ddc2d66dd7c467b48473baef8997cf","code_bytes_verified_against_code_sha256":true,"code_utf8_length":"47380","incumbent_code_sha256":"bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae","incumbent_matches_brief":true,"simulator_commit":"b244530ee2a6247b5f930830eeca9dd950da0e7e","complete":true,"failures":{"count":"0","detail":"failures []; failed_flags all false (T: B 200, weak 200, strong 200; H: B 500, weak 500); no missing seeds"},"input_result_locator":"artifacts/goal/batches/r3-deepen-current/items/retreat-crossing/result.json (also .imd/reads/artifacts/item-retreat-crossing)","input_artifact_sha256":"459d3ec15ee1323593c23f1933500cb42ac662c51e5e2aa89f9b1befee3640d1","code_url":"https://api.imd.fun/artifacts/459d3ec15ee1323593c23f1933500cb42ac662c51e5e2aa89f9b1befee3640d1","tuning":{"seeds":"310000..310199","selected_variant":"weak (strength 0.5)","variants_overall":{"weak (0.5), sha256 b5158a96...":{"mean_diff":"0.05344717880249896","se":"0.09663575554969678","n":"200"},"strong (0.9), sha256 e4e4788e...":{"mean_diff":"-0.059619851348500334","se":"0.24029790820920865","n":"200"}},"selected":{"overall":{"n":"200","sum_d":"10.689435760499792","sum_d2":"372.2423963606872","mean_diff":"0.05344717880249896","se":"0.09663575554969678","retail_diff":"0.05852414197399965","arb_diff":"-0.005076963166499737","retail_arb_unit":"mean_per_simulation","B_mean_edge":"85.16265316504199"},"s1":{"n":"47","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"57","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s3":{"n":"53","sum_d":"10.689435760499792","sum_d2":"372.2423963606872","mean_diff":"0.20168746717924138","se":"0.36644769818725903","retail_diff":"0.22084581876980997","arb_diff":"-0.01915835157169712"},"s4":{"n":"43","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}}},"paired":{"seeds":"410000..410499","screening_only":true,"overall":{"n":"500","sum_d":"120.50877344089994","sum_d2":"1453.4680382275553","mean_diff":"0.24101754688179988","se":"0.0755586623758019","retail_diff":"0.08553578669799969","arb_diff":"0.1554817601828","retail_arb_unit":"mean_per_simulation","B_mean_edge":"104.4137531881468"},"s1":{"n":"115","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"116","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s3":{"n":"138","sum_d":"120.50877344089994","sum_d2":"1453.4680382275553","mean_diff":"0.8732519814557967","se":"0.26704394325898234","retail_diff":"0.3099122706449264","arb_diff":"0.5633397108072464"},"s4":{"n":"131","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}},"seconds_per_simulation":{"value":"0.6195127642076114","basis":"1600 simulations, 991.22 s"},"flags":["strongest screening signal in the batch: H overall +0.241 (SE 0.0756), s3 +0.873 (SE 0.267); tuning was +0.053 (SE 0.097), so most of the signal is in the holdout","screening only (500 seeds); not run on finalist seeds; no promotion","retreat counts differ slightly between B and C (T ask 5041 vs 5038), expected because C's quotes change"],"promotion":false,"comparator_check":{"B_mean_edge_tuning_overall":"85.16265316504199","B_mean_edge_holdout_overall":"104.4137531881468","matches_majority_baseline":true},"arithmetic_recomputed_ok":true},{"key":"side-jump-budget","idea":"Spreads 2 and 3: add a side-specific jump charge to sizing loss (bid charged by upward jump mass, ask by downward), multiplier 1.0 selected over 0.5.","mechanism_changed":"Two guarded three-line additions immediately before qsize: bid loss += lj*sum(w*max(0,x-p))*phat*2/(1+s)*sqrt(min(1,500/H)); ask loss symmetric with (1-phat).","target_spreads":["2","3"],"code_sha256":"8c249a7d2ea89675599b3605807a195ee3c960f7db34650935c1f158ab210f14","code_bytes_verified_against_code_sha256":true,"code_utf8_length":"47013","incumbent_code_sha256":"bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae","incumbent_matches_brief":true,"simulator_commit":"b244530ee2a6247b5f930830eeca9dd950da0e7e","complete":true,"failures":{"count":"0","detail":"records []; failed_flags 0; exceptions []; spread_mismatches 0; no missing suffixes"},"input_result_locator":"artifacts/goal/batches/r3-deepen-current/items/side-jump-budget/result.json (also .imd/reads/artifacts/item-side-jump-budget)","input_artifact_sha256":"a53f876973198c8c1f235f17d201084f6ba7a97c6ad758b0b1229192ff731420","code_url":"https://api.imd.fun/artifacts/a53f876973198c8c1f235f17d201084f6ba7a97c6ad758b0b1229192ff731420","tuning":{"seeds":"310000..310199","selected_variant":"multiplier 1.0","variants_overall":{"multiplier 0.5, sha256 9b2dd6c5...":{"mean_diff":"-0.16597273997700074","se":"0.2631352761677776","n":"200"},"multiplier 1.0, sha256 8c249a7d...":{"mean_diff":"-0.019499517180500137","se":"0.4247550286728196","n":"200"}},"selected":{"overall":{"n":"200","sum_d":"-3.8999034361000273","sum_d2":"7180.666054671397","mean_diff":"-0.019499517180500137","se":"0.4247550286728196","retail_diff":"-0.9455321801715003","arb_diff":"0.9260326630065004","retail_arb_unit":"mean_per_simulation","B_mean_edge":"85.16265316504199"},"s1":{"n":"47","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"57","sum_d":"-52.831454163399926","sum_d2":"749.8669234214161","mean_diff":"-0.9268676169017531","se":"0.4685935735330723","retail_diff":"-1.673202724749123","arb_diff":"0.7463351078684216"},"s3":{"n":"53","sum_d":"48.93155072729989","sum_d2":"6430.799131249981","mean_diff":"0.9232368061754697","se":"1.522166134069477","retail_diff":"-1.7685637872377367","arb_diff":"2.6918005934490576"},"s4":{"n":"43","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}}},"paired":{"seeds":"410000..410499","screening_only":true,"overall":{"n":"500","sum_d":"249.48447364920014","sum_d2":"30624.774888752123","mean_diff":"0.4989689472984003","se":"0.3496364552445071","retail_diff":"-0.7259764881544005","arb_diff":"1.2249454354416","retail_arb_unit":"mean_per_simulation","B_mean_edge":"104.4137531881468"},"s1":{"n":"115","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"116","sum_d":"113.6588088609001","sum_d2":"12807.128551337022","mean_diff":"0.9798173177663801","se":"0.9755544182391407","retail_diff":"-2.132425071843966","arb_diff":"3.1122423895732765"},"s3":{"n":"138","sum_d":"135.82566478830003","sum_d2":"17817.646337415103","mean_diff":"0.9842439477413045","se":"0.9671413332353783","retail_diff":"-0.8378763459659435","arb_diff":"1.8221202936978258"},"s4":{"n":"131","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}},"seconds_per_simulation":{"value":"0.6494659236661391","basis":"1600 simulations, 1039.15 s"},"flags":["selected variant had a negative tuning mean (-0.0195, SE 0.425); both variants were negative on tuning","H overall +0.499 (SE 0.350) is under 2 SE; per-seed differences are large (sum_d2 30624.77), so the estimate is noisy","trades retail edge for arbitrage edge: H retail_diff -0.726, arb_diff +1.225","screening only; no promotion"],"promotion":false,"comparator_check":{"B_mean_edge_tuning_overall":"85.16265316504199","B_mean_edge_holdout_overall":"104.4137531881468","matches_majority_baseline":true},"arithmetic_recomputed_ok":true},{"key":"stock-entry-gate","idea":"All spreads: in the penny-bank stock branch, replace the upper price gate with a competitor-best-bid-at-one-tick gate.","mechanism_changed":"In the q1 penny-bank stock branch the gate ptick < P['g1']+1 becomes cb <= 1 (competitor best bid at one tick); lower bound, ydf1>0, amt!=1 and sizing unchanged.","target_spreads":["1","2","3","4"],"code_sha256":"682cdc9b5eb7cb7833d95b59f03abb5b68d18e7eb6e37761b0dc2f0403ec708d","code_bytes_verified_against_code_sha256":true,"code_utf8_length":"46674","incumbent_code_sha256":"bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae","incumbent_matches_brief":true,"simulator_commit":"b244530ee2a6247b5f930830eeca9dd950da0e7e","complete":true,"failures":{"count":"0","detail":"no failed seeds in T or H; exceptions []; failed_flags 0"},"input_result_locator":"artifacts/goal/batches/r3-deepen-current/items/stock-entry-gate/result.json (also .imd/reads/artifacts/item-stock-entry-gate)","input_artifact_sha256":"6846a33e73fa0c4685a2a7e5072ec59ba300f454ab95612e9ab7a0eade8274fe","code_url":"https://api.imd.fun/artifacts/6846a33e73fa0c4685a2a7e5072ec59ba300f454ab95612e9ab7a0eade8274fe","tuning":{"seeds":"310000..310199","selected_variant":"single fixed variant (no knobs)","variants_overall":{"single variant, sha256 682cdc9b...":{"mean_diff":"0.013507516474500023","se":"0.24441394611431186","n":"200"}},"selected":{"overall":{"n":"200","sum_d":"2.7015032949000046","sum_d2":"2377.615937396017","mean_diff":"0.013507516474500023","se":"0.24441394611431186","retail_diff":"-0.2936520264955078","arb_diff":"0.30715954296899994","retail_arb_unit":"mean_per_simulation","B_mean_edge":"85.16265316504199"},"s1":{"n":"47","sum_d":"-19.336894515","sum_d2":"230.19583998996052","mean_diff":"-0.4114232875531915","se":"0.3206147189395828","retail_diff":"-1.0393294129978656","arb_diff":"0.6279061254425518"},"s2":{"n":"57","sum_d":"20.80485140480002","sum_d2":"2146.0034757751796","mean_diff":"0.364997393066667","se":"0.8184912152485206","retail_diff":"-0.17336706821403425","arb_diff":"0.5383644612789489"},"s3":{"n":"53","sum_d":"1.233546405099986","sum_d2":"1.416621630877301","mean_diff":"0.02327446047358464","se":"0.022440952222622098","retail_diff":"0.0","arb_diff":"0.023274460473583503"},"s4":{"n":"43","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}}},"paired":{"seeds":"410000..410499","screening_only":true,"overall":{"n":"500","sum_d":"149.9963371537001","sum_d2":"40483.37776511172","mean_diff":"0.2999926743074002","se":"0.40258871718116224","retail_diff":"-0.10893709662403239","arb_diff":"0.4089297709306017","retail_arb_unit":"mean_per_simulation","B_mean_edge":"104.4137531881468"},"s1":{"n":"115","sum_d":"123.71741965370002","sum_d2":"16076.331352848158","mean_diff":"1.0758036491626088","se":"1.1027749376140752","retail_diff":"-0.23659401695564952","arb_diff":"1.312397666117393"},"s2":{"n":"116","sum_d":"60.96190247970006","sum_d2":"22023.967057376663","mean_diff":"0.5255336420663798","se":"1.2839666776403489","retail_diff":"-0.31815369868795074","arb_diff":"0.8436873407551673"},"s3":{"n":"138","sum_d":"3.912728822899993","sum_d2":"135.09588755372147","mean_diff":"0.02835310741231879","se":"0.08449730303695445","retail_diff":"-0.0553188399768203","arb_diff":"0.08367194738913142"},"s4":{"n":"131","sum_d":"-38.59571380259996","sum_d2":"2247.9834673331775","mean_diff":"-0.2946237694854959","se":"0.36239986388180995","retail_diff":"0.1319052870419739","arb_diff":"-0.4265290565305335"}},"seconds_per_simulation":{"value":"0.85635755443573","basis":"holdout mean; tuning 0.8314468866586685"},"flags":["sparse effect: only 10 of 200 tuning pairs and 22 of 500 holdout pairs differ from B","H overall +0.300 (SE 0.403) is under 1 SE; s1 +1.076 (SE 1.103), s4 -0.295 (SE 0.362)","only item targeting all four spreads","holdout process hit a 10-minute tool timeout after seed 410348 and resumed at 410349 in a new process; no seed rerun (reported by the item)","screening only; no promotion"],"promotion":false,"comparator_check":{"B_mean_edge_tuning_overall":"85.16265316504199","B_mean_edge_holdout_overall":"104.4137531881468","matches_majority_baseline":true},"arithmetic_recomputed_ok":true},{"key":"horizon-drift","idea":"Spread 3: push the posterior support one latent diffusion step forward (3-point Gauss-Hermite) and shorten the horizon by one before the AB valuation.","mechanism_changed":"Single insertion guarded by s==3 and H>1: map support probabilities to Gaussian scores, diffuse with 3-point Gauss-Hermite, moment-match for the AB valuation with H-1; jump component unchanged.","target_spreads":["3"],"code_sha256":"951a669194e7b87eb54ae2796676f7db52707ee7ecbfe2d3c7b1fb284334a25e","code_bytes_verified_against_code_sha256":true,"code_utf8_length":"47628","incumbent_code_sha256":"bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae","incumbent_matches_brief":true,"simulator_commit":"b244530ee2a6247b5f930830eeca9dd950da0e7e","complete":true,"failures":{"count":"0","detail":"failures []; missing_seeds []; failed_seeds [] in T and H"},"input_result_locator":"artifacts/goal/batches/r3-deepen-current/items/horizon-drift/result.json (also .imd/reads/artifacts/item-horizon-drift)","input_artifact_sha256":"4838d029fa2e454634bfdb47f0e5cb17bed8c690f227d165df619da3a95f56e4","code_url":"https://api.imd.fun/artifacts/4838d029fa2e454634bfdb47f0e5cb17bed8c690f227d165df619da3a95f56e4","tuning":{"seeds":"310000..310199","selected_variant":"posterior-diffusion-horizon-moments (only variant, fixed before tuning)","variants_overall":{"only variant, sha256 951a6691...":{"mean_diff":"-0.04310277276749977","se":"0.02158838135701504","n":"200"}},"selected":{"overall":{"n":"200","sum_d":"-8.620554553499954","sum_d2":"18.920686546762745","mean_diff":"-0.04310277276749977","se":"0.02158838135701504","retail_diff":"-0.029518481137499712","arb_diff":"-0.013584291618499871","retail_arb_unit":"mean_per_simulation","B_mean_edge":"85.162644703073"},"s1":{"n":"47","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"57","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s3":{"n":"53","sum_d":"-8.620554553499954","sum_d2":"18.920686546762745","mean_diff":"-0.1626519727075463","se":"0.07972771228981608","retail_diff":"-0.11139049485848948","arb_diff":"-0.05126147780565989"},"s4":{"n":"43","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}}},"paired":{"seeds":"410000..410499","screening_only":true,"overall":{"n":"500","sum_d":"33.00332127110009","sum_d2":"835.5974023027458","mean_diff":"0.06600664254220018","se":"0.0577958179164276","retail_diff":"0.019873491177199915","arb_diff":"0.046133151368800016","retail_arb_unit":"mean_per_simulation","B_mean_edge":"104.39217582931519"},"s1":{"n":"115","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"116","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s3":{"n":"138","sum_d":"33.00332127110009","sum_d2":"835.5974023027458","mean_diff":"0.23915450196449342","se":"0.20923668650257599","retail_diff":"0.07200540281594171","arb_diff":"0.1671490991623189"},"s4":{"n":"131","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}},"seconds_per_simulation":{"value":"1.3155260955086643","basis":"1400 simulations, 1841.74 s"},"flags":["tuning and holdout disagree in sign: T -0.0431 (SE 0.0216, about -2 SE); H +0.0660 (SE 0.0578, about +1.1 SE)","comparator mismatch: B baseline differs from the eight-item majority (see 'Comparator mismatches'); within-item pairing unaffected","approximation acknowledged by the item: uses the incumbent Gaussian score map, not the simulator's compound-Poisson map","screening only; no promotion"],"promotion":false,"comparator_check":{"B_mean_edge_tuning_overall":"85.162644703073","B_mean_edge_holdout_overall":"104.39217582931519","matches_majority_baseline":false},"arithmetic_recomputed_ok":true},{"key":"variance-floor","idea":"Spread 2: floor the effective variance at (c*sp)^2 scaled to one-step diffusion sd instead of 1e-08 (c=0.5 selected over 0.25).","mechanism_changed":"After B computes s at spread 2, s = sqrt(sp^2 + max(var_t/1e4 + 1e-08, (c*sp)^2)); feeds AB sizing and the probe range; filter untouched.","target_spreads":["2"],"code_sha256":"af654263059b776567f552d371113020e5c9a37f922fdb8fa528de9fac62aa3a","code_bytes_verified_against_code_sha256":true,"code_utf8_length":"46730","incumbent_code_sha256":"bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae","incumbent_matches_brief":true,"simulator_commit":"b244530ee2a6247b5f930830eeca9dd950da0e7e","complete":true,"failures":{"count":"0","detail":"missing_seeds []; failed_flags []; exceptions []; label mismatches 0"},"input_result_locator":"artifacts/goal/batches/r3-deepen-current/items/variance-floor/result.json (also .imd/reads/artifacts/item-variance-floor)","input_artifact_sha256":"c762056de359e467be3b18327480640b00074a3846dbaa27644be28cd11e1019","code_url":"https://api.imd.fun/artifacts/c762056de359e467be3b18327480640b00074a3846dbaa27644be28cd11e1019","tuning":{"seeds":"310000..310199","selected_variant":"c=0.5 (C050)","variants_overall":{"c=0.25, sha256 52051d7a...":{"mean_diff":"0.004161767001500036","se":"0.0015356134301066048","n":"200"},"c=0.5, sha256 af654263...":{"mean_diff":"0.038226703622","se":"0.03493212985277214","n":"200"}},"selected":{"overall":{"n":"200","sum_d":"7.6453407244","sum_d2":"48.85835327678804","mean_diff":"0.038226703622","se":"0.03493212985277214","retail_diff":"-0.18473503258599067","arb_diff":"0.22296173621199955","retail_arb_unit":"mean_per_simulation","B_mean_edge":"85.16265316504199"},"s1":{"n":"47","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"57","sum_d":"7.6453407244","sum_d2":"48.85835327678804","mean_diff":"0.13412878463859648","se":"0.12241422373781641","retail_diff":"-0.6481930967929799","arb_diff":"0.7823218814456148"},"s3":{"n":"53","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s4":{"n":"43","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}}},"paired":{"seeds":"410000..410499","screening_only":true,"overall":{"n":"500","sum_d":"2.5322396334000206","sum_d2":"169.34663027278427","mean_diff":"0.0050644792668000415","se":"0.02605172947056422","retail_diff":"-0.18343166808142541","arb_diff":"0.18849614734159914","retail_arb_unit":"mean_per_simulation","B_mean_edge":"104.4137531881468"},"s1":{"n":"115","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"116","sum_d":"2.5322396334000206","sum_d2":"169.34663027278427","mean_diff":"0.021829652012069142","se":"0.11265214677990446","retail_diff":"-0.790653741730182","arb_diff":"0.8124833937137907"},"s3":{"n":"138","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s4":{"n":"131","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}},"seconds_per_simulation":{"value":"0.27176597608089653","basis":"1600 simulations, 434.83 s (fastest machine in the batch)"},"flags":["null result: H overall +0.0051 (SE 0.0261); s2 +0.0218 (SE 0.1127)","diagnostic runs on seeds 1000..1099 and a smoke test on 1000..1007 were outside T, H and finalist ranges and excluded from statistics (reported by the item)","Python 3.14.4 reported; B baseline edges match the majority exactly","screening only; no promotion"],"promotion":false,"comparator_check":{"B_mean_edge_tuning_overall":"85.16265316504199","B_mean_edge_holdout_overall":"104.4137531881468","matches_majority_baseline":true},"arithmetic_recomputed_ok":true},{"key":"grid-resolution","idea":"Spread 2: raise the within-interval posterior grid from 8 bins/tick (cap 48) to 32 bins/tick (cap 192); 16/96 rejected on tuning.","mechanism_changed":"nb = max(2, min(48, round(wid*8))) becomes, under s==2, wid*32 with cap 192; prior kernel, floor, normalization and likelihood unchanged.","target_spreads":["2"],"code_sha256":"22806a5844bd249e9a7b89b2b34b49292f7553e56a37c5781861bc9d30f4f36b","code_bytes_verified_against_code_sha256":true,"code_utf8_length":"46723","incumbent_code_sha256":"bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae","incumbent_matches_brief":true,"simulator_commit":"b244530ee2a6247b5f930830eeca9dd950da0e7e","complete":true,"failures":{"count":"0","detail":"count 0; exceptions []; failed_flags 0; missing_seeds []"},"input_result_locator":"artifacts/goal/batches/r3-deepen-current/items/grid-resolution/result.json (also .imd/reads/artifacts/item-grid-resolution)","input_artifact_sha256":"91390b3932027df0bfc5f1130c0221055bcab2aebb6da36794b69909209e9a01","code_url":"https://api.imd.fun/artifacts/91390b3932027df0bfc5f1130c0221055bcab2aebb6da36794b69909209e9a01","tuning":{"seeds":"310000..310199","selected_variant":"C32 (32 bins/tick, cap 192)","variants_overall":{"C16 (16/tick, cap 96)":{"mean_diff":"-0.0034490533470004836","se":"0.016570714948185367","n":"200"},"C32 (32/tick, cap 192), sha256 22806a58...":{"mean_diff":"6.560040300004205e-05","se":"0.017804355310354902","n":"200"}},"selected":{"overall":{"n":"200","sum_d":"0.01312008060000841","sum_d2":"12.616404567773618","mean_diff":"6.560040300004205e-05","se":"0.017804355310354902","retail_diff":"-0.0032051986605006856","arb_diff":"0.003270799060500056","retail_arb_unit":"mean_per_simulation","B_mean_edge":"85.16265316504199"},"s1":{"n":"47","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"57","sum_d":"0.01312008060000841","sum_d2":"12.616404567773618","mean_diff":"0.0002301768526317265","se":"0.06286896492367915","retail_diff":"-0.01124631108947609","arb_diff":"0.011476487931579144"},"s3":{"n":"53","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s4":{"n":"43","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}}},"paired":{"seeds":"410000..410499","screening_only":true,"overall":{"n":"500","sum_d":"5.307132093500122","sum_d2":"260.1071242919925","mean_diff":"0.010614264187000243","se":"0.03228448152680194","retail_diff":"0.02168030183599953","arb_diff":"-0.011066037658599984","retail_arb_unit":"mean_per_simulation","B_mean_edge":"104.4137531881468"},"s1":{"n":"115","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"116","sum_d":"5.307132093500122","sum_d2":"260.1071242919925","mean_diff":"0.04575113873707001","se":"0.13957107053319331","retail_diff":"0.09344957687930831","arb_diff":"-0.04769843818362062"},"s3":{"n":"138","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s4":{"n":"131","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}},"seconds_per_simulation":{"value":"1.027647920968011","basis":"holdout C32; B holdout 0.8899; C32 is about 15% slower than B"},"flags":["null result: H overall +0.0106 (SE 0.0323); s2 +0.0458 (SE 0.1396)","C16 variant hash not reported in the result","screening only; no promotion"],"promotion":false,"comparator_check":{"B_mean_edge_tuning_overall":"85.16265316504199","B_mean_edge_holdout_overall":"104.4137531881468","matches_majority_baseline":true},"arithmetic_recomputed_ok":true},{"key":"probe-attribution","idea":"Spread 2: give 0.01 probe orders identities and a cancel ledger so fill-derived bounds use truly missing probes only.","mechanism_changed":"Spread 2: probe orders carry identities; a side/tick ledger and exact CancelOrder identities let fill-derived bounds use genuinely missing probes instead of all tiny residuals.","target_spreads":["2"],"code_sha256":"ff72a56eaa8f77734cabf4d902a2b12fdbe8a9dd71bfee5bff147842cccf1902","code_bytes_verified_against_code_sha256":true,"code_utf8_length":"48302","incumbent_code_sha256":"bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae","incumbent_matches_brief":true,"simulator_commit":"b244530ee2a6247b5f930830eeca9dd950da0e7e","complete":true,"failures":{"count":"0","detail":"failures []; failed_seeds []; missing_seed_ranges []; B_failed 0, C_failed 0"},"input_result_locator":"artifacts/goal/batches/r3-deepen-current/items/probe-attribution/result.json (also .imd/reads/artifacts/item-probe-attribution)","input_artifact_sha256":"23dc453b7a19305f3bf8fdd996ccb1254aec80bf1758b00c60d9b4071474ffc1","code_url":"https://api.imd.fun/artifacts/23dc453b7a19305f3bf8fdd996ccb1254aec80bf1758b00c60d9b4071474ffc1","tuning":{"seeds":"310000..310199","selected_variant":"identity-ledger (only variant)","variants_overall":{"identity-ledger, sha256 ff72a56e...":{"mean_diff":"0.007944665855499942","se":"0.006846101451125877","n":"200"}},"selected":{"overall":{"n":"200","sum_d":"1.5889331710999883","sum_d2":"1.8780139252596018","mean_diff":"0.007944665855499942","se":"0.006846101451125877","retail_diff":"0.0006601114615000014","arb_diff":"0.007284554394000009","retail_arb_unit":"mean_per_simulation","B_mean_edge":"85.162644703073"},"s1":{"n":"47","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"57","sum_d":"1.5889331710999883","sum_d2":"1.8780139252596018","mean_diff":"0.02787602054561383","se":"0.02396818583425939","retail_diff":"0.0023161805666666716","arb_diff":"0.0255598399789474"},"s3":{"n":"53","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s4":{"n":"43","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}}},"paired":{"seeds":"410000..410499","screening_only":true,"overall":{"n":"500","sum_d":"-8.636520615399991","sum_d2":"45.45091898180984","mean_diff":"-0.017273041230799983","se":"0.013474794997363019","retail_diff":"-0.027284591635000084","arb_diff":"0.010011550404000038","retail_arb_unit":"mean_per_simulation","B_mean_edge":"104.39217582931519"},"s1":{"n":"115","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"116","sum_d":"-8.636520615399991","sum_d2":"45.45091898180984","mean_diff":"-0.07445276392586199","se":"0.057956134313759226","retail_diff":"-0.11760599842672449","arb_diff":"0.04315323450000016"},"s3":{"n":"138","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s4":{"n":"131","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}},"seconds_per_simulation":{"value":"1.396860032439498","basis":"1400 simulations, 1955.60 s"},"flags":["null result: T +0.0079 (SE 0.0068); H -0.0173 (SE 0.0135); only 2 of 200 tuning pairs and 7 of 500 holdout pairs differ","source flag: one cited source did not hold on re-read (merge brief); sources list not supplied, affected claim unknown; paired numbers are local measurements","comparator mismatch: B baseline matches horizon-drift's machine, not the eight-item majority (see 'Comparator mismatches'); Python 3.10.12","screening only; no promotion"],"promotion":false,"comparator_check":{"B_mean_edge_tuning_overall":"85.162644703073","B_mean_edge_holdout_overall":"104.39217582931519","matches_majority_baseline":false},"arithmetic_recomputed_ok":true},{"key":"tail-quadrature","idea":"Spread 2: replace B's six ad-hoc jump nodes (second moment 1.35) with normalized 10-node Gauss-Hermite nodes; 16-node rejected on tuning.","mechanism_changed":"Spread 2: the six ad-hoc jump nodes are replaced by hard-coded normalized probabilists' Gauss-Hermite nodes and weights (10 nodes); lam_j and sig_j unchanged.","target_spreads":["2"],"code_sha256":"851f2f2d5b9a5bad8f140c635ac97a901ea99b083d95291bef447c89678b0849","code_bytes_verified_against_code_sha256":true,"code_utf8_length":"47077","incumbent_code_sha256":"bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae","incumbent_matches_brief":true,"simulator_commit":"b244530ee2a6247b5f930830eeca9dd950da0e7e","complete":true,"failures":{"count":"0","detail":"failures []; missing_seeds []; duplicate_seeds 0; label mismatches 0"},"input_result_locator":"artifacts/goal/batches/r3-deepen-current/items/tail-quadrature/result.json (also .imd/reads/artifacts/item-tail-quadrature)","input_artifact_sha256":"27431b07fb1bf1dbc667c1271dfda472459232cb96b864bee872488278f6e1c4","code_url":"https://api.imd.fun/artifacts/27431b07fb1bf1dbc667c1271dfda472459232cb96b864bee872488278f6e1c4","tuning":{"seeds":"310000..310199","selected_variant":"GH-10","variants_overall":{"GH-10, sha256 851f2f2d...":{"mean_diff":"0.19368208455749983","se":"0.17908544851731953","n":"200"},"GH-16 (hash not reported)":{"mean_diff":"0.18235667532050026","se":"0.17875585191876073","n":"200"}},"selected":{"overall":{"n":"200","sum_d":"38.736416911499965","sum_d2":"1283.952145227558","mean_diff":"0.19368208455749983","se":"0.17908544851731953","retail_diff":"0.3394260737694996","arb_diff":"-0.14574398922099993","retail_arb_unit":"mean_per_simulation","B_mean_edge":"85.16265316504199"},"s1":{"n":"47","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"57","sum_d":"38.736416911499965","sum_d2":"1283.952145227558","mean_diff":"0.6795862616052626","se":"0.6276890544048377","retail_diff":"1.1909686798929813","arb_diff":"-0.511382418319298"},"s3":{"n":"53","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s4":{"n":"43","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}}},"paired":{"seeds":"410000..410499","screening_only":true,"overall":{"n":"500","sum_d":"-59.27167895730018","sum_d2":"8825.810279852767","mean_diff":"-0.11854335791460037","se":"0.18800486089311258","retail_diff":"0.2713849839179995","arb_diff":"-0.3899283418426","retail_arb_unit":"mean_per_simulation","B_mean_edge":"104.4137531881468"},"s1":{"n":"115","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"116","sum_d":"-59.27167895730018","sum_d2":"8825.810279852767","mean_diff":"-0.5109627496318981","se":"0.8119942623835076","retail_diff":"1.169762861715515","arb_diff":"-1.6807256113905171"},"s3":{"n":"138","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s4":{"n":"131","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}},"seconds_per_simulation":{"value":"0.8624150770902633","basis":"mean over all simulations"},"flags":["sign flip between tuning and holdout: T +0.194 (SE 0.179); H -0.119 (SE 0.188); s2 H -0.511 (SE 0.812)","source flag: one cited source did not hold on re-read (merge brief); sources list not supplied, affected claim unknown; paired numbers are local measurements","T run in two invocations (310000-310001, then 310002-310199) and H in two (410000-410348, 410349-410499) after a shell timeout; no seed repeated (reported by the item)","screening only; no promotion"],"promotion":false,"comparator_check":{"B_mean_edge_tuning_overall":"85.16265316504199","B_mean_edge_holdout_overall":"104.4137531881468","matches_majority_baseline":true},"arithmetic_recomputed_ok":true},{"key":"probe-extremes","idea":"Spread 3: relocate the same number of 0.01 probes to the ticks nearest the upper/lower 0.1 posterior quantiles (clamped variant selected).","mechanism_changed":"Spread 3: the same counts of 0.01 probes are moved to the ticks nearest the upper and lower 0.1 quantiles of the current tick posterior, clamped to buy>=f and sell<=f+1.","target_spreads":["3"],"code_sha256":"c2c076dd69826effc18840314068a1155792c93416d565f7e6a82641deca02c9","code_bytes_verified_against_code_sha256":true,"code_utf8_length":"47973","incumbent_code_sha256":"bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae","incumbent_matches_brief":true,"simulator_commit":"b244530ee2a6247b5f930830eeca9dd950da0e7e","complete":true,"failures":{"count":"0","detail":"failures [] in T and H"},"input_result_locator":"artifacts/goal/batches/r3-deepen-current/items/probe-extremes/result.json (also .imd/reads/artifacts/item-probe-extremes)","input_artifact_sha256":"5e00c9698814138ab644d73c7f3697317ca50a60a15208b7083d4c8d0e5fa813","code_url":"https://api.imd.fun/artifacts/5e00c9698814138ab644d73c7f3697317ca50a60a15208b7083d4c8d0e5fa813","tuning":{"seeds":"310000..310199","selected_variant":"C1 (PE_ALPHA=0.1, clamped)","variants_overall":{"C1 clamped, sha256 c2c076dd...":{"mean_diff":"-0.0003998508510002807","se":"0.00035809102913807487","n":"200"},"C2 no clamp (hash not reported; identical per-seed totals to C1 on T)":{"mean_diff":"-0.0003998508510002807","se":"0.00035809102913807487","n":"200"}},"selected":{"overall":{"n":"200","sum_d":"-0.07997017020005615","sum_d2":"0.0051354977095459204","mean_diff":"-0.0003998508510002807","se":"0.00035809102913807487","retail_diff":"-1.1406865000651579e-06","arb_diff":"-0.0003987101675000115","retail_arb_unit":"mean_per_simulation","B_mean_edge":"85.16265316504199"},"s1":{"n":"47","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"57","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s3":{"n":"53","sum_d":"-0.07997017020005615","sum_d2":"0.0051354977095459204","mean_diff":"-0.001508871135850116","se":"0.0013489275732850923","retail_diff":"-4.304477358736445e-06","arb_diff":"-0.0015045666698113643"},"s4":{"n":"43","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}}},"paired":{"seeds":"410000..410499","screening_only":true,"overall":{"n":"500","sum_d":"-0.03486687440005465","sum_d2":"8.175777953202527e-05","mean_diff":"-6.97337488001093e-05","se":"1.7830911628508366e-05","retail_diff":"-7.947093999050026e-07","arb_diff":"-6.893903820005653e-05","retail_arb_unit":"mean_per_simulation","B_mean_edge":"104.4137531881468"},"s1":{"n":"115","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"116","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s3":{"n":"138","sum_d":"-0.03486687440005465","sum_d2":"8.175777953202527e-05","mean_diff":"-0.00025265851014532356","se":"6.211664063628644e-05","retail_diff":"-2.8793818837137772e-06","arb_diff":"-0.0002497791239132483"},"s4":{"n":"131","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}},"seconds_per_simulation":{"value":"0.8628","basis":"reported scalar"},"flags":["negligible negative effect: H overall -0.0000697 (SE 0.0000178); 62 of 500 holdout pairs nonzero; magnitude is far below any ranking relevance","source flag: one cited source did not hold on re-read (merge brief); sources list not supplied, affected claim unknown; paired numbers are local measurements","label note from the item: groups use regime.competitor_spread_ticks; the strategy's own s label was not logged separately","H interrupted by a 10-minute tool timeout after seed 410347 and the suffix 410348..410499 run in a second process; no seed repeated (reported by the item)","numbers are stored as strings in this item's result and are carried through unchanged","screening only; no promotion"],"promotion":false,"comparator_check":{"B_mean_edge_tuning_overall":"85.16265316504199","B_mean_edge_holdout_overall":"104.4137531881468","matches_majority_baseline":true},"arithmetic_recomputed_ok":true},{"key":"posterior-payoff","idea":"Spread 3: replace the diffusion AB Gaussian moment collapse with a posterior-cell-weighted average of expected losses (v2 cell width selected).","mechanism_changed":"Spread 3: diffusion AB Gaussian moment collapse replaced by a weighted average of diffusion expected losses over existing posterior cells; jump AB and downstream policies unchanged.","target_spreads":["3"],"code_sha256":"94d8fd887742b27444643264bc7309a241c377abfc7b9d0fa3d8a11e42d85f68","code_bytes_verified_against_code_sha256":true,"code_utf8_length":"46934","incumbent_code_sha256":"bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae","incumbent_matches_brief":true,"simulator_commit":"b244530ee2a6247b5f930830eeca9dd950da0e7e","complete":true,"failures":{"count":"0","detail":"no failures in tuning_B, tuning_v1, tuning_v2, holdout_B, holdout_selected; missing_seeds []"},"input_result_locator":"artifacts/goal/batches/r3-deepen-current/items/posterior-payoff/result.json (also .imd/reads/artifacts/item-posterior-payoff)","input_artifact_sha256":"0fbc810ac55efec9ab4437c06c1a5256754aaf95e126a2302d95b424ae269529","code_url":"https://api.imd.fun/artifacts/0fbc810ac55efec9ab4437c06c1a5256754aaf95e126a2302d95b424ae269529","tuning":{"seeds":"310000..310199","selected_variant":"v2 (cell width sqrt(sp^2+0.5*var_t/10000))","variants_overall":{"v1 (cell width sp; hash not reported)":{"mean_diff":"-0.09468707148199995","se":"0.06852515983399698","n":"200"},"v2, sha256 94d8fd88...":{"mean_diff":"-0.07754086789900097","se":"0.06420810385781751","n":"200"}},"selected":{"overall":{"n":"200","sum_d":"-15.508173579800195","sum_d2":"165.285205159354","mean_diff":"-0.07754086789900097","se":"0.06420810385781751","retail_diff":"-69.03017023460019","arb_diff":"53.52199665489999","retail_arb_unit":"sum_over_pairs (item reports signed sums, not means)","B_mean_edge":"85.16265316504199"},"s1":{"n":"47","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"57","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s3":{"n":"53","sum_d":"-15.508173579800195","sum_d2":"165.285205159354","mean_diff":"-0.2926070486754754","se":"0.24150847618733215","retail_diff":"-69.03017023460019","arb_diff":"53.52199665489999"},"s4":{"n":"43","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}}},"paired":{"seeds":"410000..410499","screening_only":true,"overall":{"n":"500","sum_d":"-53.75575245900014","sum_d2":"1060.0132601949692","mean_diff":"-0.10751150491800028","se":"0.06500297197923088","retail_diff":"-202.05755171170003","arb_diff":"148.30179925030004","retail_arb_unit":"sum_over_pairs (item reports signed sums, not means)","B_mean_edge":"104.4137531881468"},"s1":{"n":"115","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s2":{"n":"116","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"},"s3":{"n":"138","sum_d":"-53.75575245900014","sum_d2":"1060.0132601949692","mean_diff":"-0.3895344381086967","se":"0.23443546775500412","retail_diff":"-202.05755171170003","arb_diff":"148.30179925030004"},"s4":{"n":"131","sum_d":"0.0","sum_d2":"0.0","mean_diff":"0.0","se":"0.0","retail_diff":"0.0","arb_diff":"0.0"}},"seconds_per_simulation":{"value":"0.946","basis":"reported scalar"},"flags":["negative on both sets: T -0.0775 (SE 0.0642); H -0.1075 (SE 0.0650, about -1.65 SE); s3 H -0.390 (SE 0.234)","retail/arb differences are reported as signed sums over pairs (they add to sum_d), not per-simulation means; carried as given with the unit marked","ran via the CLI (orderbook-pm run ... --json) rather than the engine API; paired by seed (reported by the item)","screening only; no promotion"],"promotion":false,"comparator_check":{"B_mean_edge_tuning_overall":"85.16265316504199","B_mean_edge_holdout_overall":"104.4137531881468","matches_majority_baseline":true},"arithmetic_recomputed_ok":true}]}
```

## Inputs

| item | input result locator | artifact SHA-256 (supplied and recomputed) |
|---|---|---|
| retreat-crossing | artifacts/goal/batches/r3-deepen-current/items/retreat-crossing/result.json | 459d3ec15ee1323593c23f1933500cb42ac662c51e5e2aa89f9b1befee3640d1 |
| side-jump-budget | artifacts/goal/batches/r3-deepen-current/items/side-jump-budget/result.json | a53f876973198c8c1f235f17d201084f6ba7a97c6ad758b0b1229192ff731420 |
| stock-entry-gate | artifacts/goal/batches/r3-deepen-current/items/stock-entry-gate/result.json | 6846a33e73fa0c4685a2a7e5072ec59ba300f454ab95612e9ab7a0eade8274fe |
| horizon-drift | artifacts/goal/batches/r3-deepen-current/items/horizon-drift/result.json | 4838d029fa2e454634bfdb47f0e5cb17bed8c690f227d165df619da3a95f56e4 |
| variance-floor | artifacts/goal/batches/r3-deepen-current/items/variance-floor/result.json | c762056de359e467be3b18327480640b00074a3846dbaa27644be28cd11e1019 |
| grid-resolution | artifacts/goal/batches/r3-deepen-current/items/grid-resolution/result.json | 91390b3932027df0bfc5f1130c0221055bcab2aebb6da36794b69909209e9a01 |
| probe-attribution | artifacts/goal/batches/r3-deepen-current/items/probe-attribution/result.json | 23dc453b7a19305f3bf8fdd996ccb1254aec80bf1758b00c60d9b4071474ffc1 |
| tail-quadrature | artifacts/goal/batches/r3-deepen-current/items/tail-quadrature/result.json | 27431b07fb1bf1dbc667c1271dfda472459232cb96b864bee872488278f6e1c4 |
| probe-extremes | artifacts/goal/batches/r3-deepen-current/items/probe-extremes/result.json | 5e00c9698814138ab644d73c7f3697317ca50a60a15208b7083d4c8d0e5fa813 |
| posterior-payoff | artifacts/goal/batches/r3-deepen-current/items/posterior-payoff/result.json | 0fbc810ac55efec9ab4437c06c1a5256754aaf95e126a2302d95b424ae269529 |

Other references: incumbent B artifact https://api.imd.fun/artifacts/db22aff7167a4798251ed3d20432509f30e511e4958799aba5b17b41bb39b4b6; wave 3 report https://github.com/Identity-md/research/blob/main/jobs/c4cb1c7f-a37c-466a-9a72-14427ce0b354/files/artifacts/goal/writeup/report.md (named in the brief, not re-read for this merge); challenge page https://www.paradigm.xyz/puzzles/prediction-market-challenge. The current standing figures in the brief (3rd of 186 at 104.84) are the brief's own and were not re-checked here.