# Merge 03 of batch r3-deepen-current: evidence table for 10 items

**No promotion.** All ten items are screening results against the incumbent B (code SHA-256 `bf0d179f…ffae`). Each used tuning seeds 310000..310199 and holdout seeds 410000..410499 (n=500). None ran finalist seeds 900000..903999. The scoreboard best stays B. Each row below measures one policy against B, so nothing is pooled across policies. Every item reports `complete: true` with no failures or missing seeds.

## Verification done in this merge
- I recomputed the SHA-256 of each item's `code` field (UTF-8 bytes) and it equals its `code_sha256` for all 10. The code is not duplicated here. Full code stays in each input result (`code` field), located by the `input_result_locator` in the table.
- Every row's `incumbent_code_sha256` equals B (`bf0d179f…`). There are no comparator mismatches. I did not re-download B, so the B hash is checked only as what each item recorded.
- Each item's own flag arrays (`failures`, `failed_flags`) are empty. I did not re-run any simulation.
- `artifact_sha256` is the SHA-256 given in this task's input metadata for each result file. `code_url` is `https://api.imd.fun/artifacts/<that digest>`. The digest is of the result JSON, which contains the code, not of the bare code string.
- Spread labels: items state they verified the label against the sampled competitor spread; I did not re-check this.

## Source-flag caveat
The requester says some cited sources did not hold when re-read: s3-replenish (1 source), s3-recollateralize (2 sources), s3-order-ledger (1 source). The `sources.json` files were not among my inputs, so I cannot say which citations failed. I also found no URL quoted in those three results other than the simulator tree and the incumbent artifact (in s3-order-ledger). Their numbers come from local paired simulation. I take only the numbers from them and treat any literature or background claim in them as unsupported. None of the three shows a gain either way, so the caveat changes no conclusion.

## Screening summary per target spread (mean paired d, SE in brackets; edge units per simulation)
Holdout n per spread: s1=115, s2=116, s3=138, s4=131. Tuning n per spread: s1=47, s2=57, s3=53, s4=43. "z" is the holdout mean divided by its SE. Retail/arb are the holdout differences for the target spread (retail+arb equals d).

| item | target s | tuning d (SE) | holdout d at target s (SE) | holdout overall d (SE) | z | holdout retail / arb diff (target s) |
|---|---|---|---|---|---|---|
| s2-bank-cash-cap | 2 | +0.278 (0.451) | +0.063 (0.923) | +0.015 (0.213) | +0.07 | -0.1035 / +0.1662 |
| s3-replenish | 3 | +0.017 (0.064) | -0.038 (0.066) | -0.011 (0.018) | -0.59 | -0.06968 / +0.03119 |
| s3-premium-payback | 3 | +0.000 (0.000) | +0.000 (0.000) | +0.000 (0.000) | n/a | +0 / +0 |
| s2-late-uncovered | 2 | +0.011 (0.012) | -0.001 (0.002) | -0.000 (0.000) | -0.70 | +3.774e-06 / -0.001481 |
| s3-recollateralize | 3 | -0.034 (0.031) | +1.183 (1.134) | +0.327 (0.313) | +1.04 | +0.909 / +0.2744 |
| s2-flow-exposure | 2 | -0.221 (0.252) | -0.301 (0.240) | -0.070 (0.056) | -1.26 | -0.02141 / -0.2799 |
| s3-censor-cap | 3 | +0.002 (0.001) | +0.030 (0.030) | +0.008 (0.008) | +1.03 | -0.0003705 / +0.03071 |
| s2-flow-jump-mask | 2 | +0.056 (0.150) | +0.351 (0.221) | +0.081 (0.051) | +1.59 | +2.603e-06 / +0.3509 |
| s3-order-ledger | 3 | +0.001 (0.004) | +0.035 (0.032) | +0.010 (0.009) | +1.11 | -0.002046 / +0.03726 |
| s1-two-cent-commit | 1 | -0.017 (0.017) | +0.031 (0.084) | +0.007 (0.019) | +0.37 | -0.04125 / +0.07262 |

**Per spread.**
- **Spread 1:** only s1-two-cent-commit. Holdout +0.031 (SE 0.084) at s1, so no effect.
- **Spread 2:** four items.
  - s2-flow-jump-mask is the only positive, at +0.351 (SE 0.221), z=1.59, all on the arb side.
  - s2-bank-cash-cap gives +0.063 (SE 0.923).
  - s2-late-uncovered gives -0.001 (SE 0.002).
  - s2-flow-exposure is negative at -0.301 (SE 0.240), all arb.
- **Spread 3:** five items.
  - s3-recollateralize is +1.183 (SE 1.134) at s3, driven by a single seed (see flags).
  - s3-order-ledger is +0.035 (SE 0.032).
  - s3-censor-cap is +0.030 (SE 0.030).
  - s3-replenish is -0.038 (SE 0.066).
  - s3-premium-payback is exactly 0, because its selected variant never changed an outcome.
- **Spread 4:** no item targeted it.
- **Verdict:** no item exceeds 2 SE on its own spread. Overall holdout gains are at most +0.33, against SE 0.3 or more for the largest, so none can be ranked. A final ranking needs thousands of paired seeds on 900000..903999, run by a measure item, which has not happened.

## What the results learned
- **Estimator tweaks are inert to slightly negative.** The flow estimator changes (censor-cap, order-ledger, flow-exposure, flow-jump-mask) moved edge by less than 0.1 per simulation overall. The flow-jump-mask item says s2 ask sizing barely uses the flow estimate, only the 1-tick YES stock. It suggests applying the masked estimate to s2 ask sizing as a next step.
- **Cash and inventory gates mostly never fire.** premium-payback changed nothing, and replenish and recollateralize changed only a few seeds. recollateralize says cash binds rarely on s3.
- **Heavy tails.** recollateralize's holdout sum_d is 163.31, of which seed 410361 contributes +156.44. Without it the other four changed seeds sum to +6.87 (item's own figures). bank-cash-cap's holdout sum_d2 is about 11372 on a mean of 0.015, so a few seeds carry the variance.
- **Nothing here attacks the arbitrage loss directly.** The screening shows none of these ten is a lever for the gap to first place (about -2.6 arbitrage and about -2.2 retail points).

## Per-row notes (flags)
- **s2-bank-cash-cap**: Caps cash resting in the tick-1 penny-bank bid to 10% of equity (cash + Y*phat). BANK_FRAC 0.10 chosen over 0.25 on tuning. Flags: holdout run interrupted by tool timeout after seed 410344; suffix 410345..410499 rerun with same frozen code, no seed replaced (item note); retail/arb differences reported as per-seed means (retail_diff_mean/arb_diff_mean), same quantity as retail_diff/arb_diff elsewhere; holdout sum_d2 11372 indicates heavy-tailed per-seed d
- **s3-replenish**: Tags stock purchases (m-prefix client IDs) and subtracts still-resting purchases from the stock-target shortfall before sizing new ones. Flags: item flagged by the requester: 1 cited source did not hold when re-read; sources.json not supplied to this merge so the exact source is unidentified. Result is a measured simulation, not dependent on the quote
- **s3-premium-payback**: Gates the ammo "premium-to-phat" purchase on whether the premium cost is covered by estimated remaining retail edge. Selected C2 (stochastic demand bound); the other variant was mean-demand bound. Flags: selected C2 identical to B on every one of 700 tuning+holdout seeds (all d exactly 0): gate never changed behaviour
- **s2-late-uncovered**: Late (H<300) and phat<0.05: scales the terminal uncovered-ask allowance above the ordinary budget by retail_gain/(retail_gain+jump_loss). No flags.
- **s3-recollateralize**: Cancel and repost a kept inside-rival ask whose size is uncovered, once YES inventory can cover it; selected C2 fires only when cash binds. Flags: item flagged by the requester: 2 cited sources did not hold when re-read; sources.json not supplied so unidentified; holdout +0.327 dominated by one seed: 410361 d=+156.44 of sum_d +163.31 (item text); only 5 of 138 s3 seeds differ from B; repost counts (61 C2 / 797 C1) not reconciled against an independent count (item text)
- **s2-flow-exposure**: Replaces the binary flow-exposure increment of 1 with 1/last_flow_correction (capacity-fractional exposure) in the spread-2 flow estimator. No flags.
- **s3-censor-cap**: Lowers the cap on the censoring correction in the spread-3 flow estimate from 3.0 to 1.5 (B cap never binds; max factor ~1.823). Flags: holdout run interrupted by tool timeout after 783 records; resumed, every (seed,strategy) once (item note); cap chosen on tuning from 1.5 vs 1.25
- **s2-flow-jump-mask**: Spread-2 flow estimator drops an observation when the rival ask rose at all or the rival bid moved >1 tick (variant CB; CA was ask-only). Flags: B alone ran once on seeds 310000..310002 as a smoke test before tuning; output discarded, those seeds then rerun in tuning (item note); variant CB chosen on tuning by higher mean
- **s3-order-ledger**: Spread-3 demand estimator uses a per-order ledger of quantity reductions on tracked ask IDs instead of aggregate sell_filled_quantity. Flags: item flagged by the requester: 1 cited source did not hold when re-read; the only URLs in the result are the simulator tree and incumbent artifact (both verified-looking), sources.json not supplied
- **s1-two-cent-commit**: Spread 1: caps a new 2-cent ask size at K*fl_rate*H/0.02 shares (K=1 chosen; K=0.5 worse on tuning). Flags: two variants tried (K=1 chosen, K=0.5); holdout nonzero only on s1

## Full evidence table (JSON)
The numbers are strings copied unrounded from the item results (n, sum_d, sum_d2, mean_diff, se, retail_diff, arb_diff). `tuning` and `paired` hold groups `overall` and `s1`..`s4` (paired = holdout 410000..410499). `retail_diff` and `arb_diff` are the per-seed means (bank-cash-cap's `retail_diff_mean` and `arb_diff_mean`). Where the tuning block covers a selected variant, it is for the variant whose code is cited, and the item's other variant(s) are in its own result. For s2-bank-cash-cap the code carries BANK_FRAC=0.10 and the 0.25 variant is in its result. Tuning and holdout seeds do not overlap.

```json
{
 "incumbent_B_sha256": "bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae",
 "rows": [
  {
   "key": "s2-bank-cash-cap",
   "idea": "Caps cash resting in the tick-1 penny-bank bid to 10% of equity (cash + Y*phat). BANK_FRAC 0.10 chosen over 0.25 on tuning.",
   "target_spreads": [
    2
   ],
   "code_sha256": "904b85c90dd501666d730be86e2407746a2115b3cdf3bc40caf0ee33d15c1283",
   "incumbent_code_sha256": "bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae",
   "comparator_matches_B": true,
   "code_bytes_verified": true,
   "complete": true,
   "failures": [],
   "tuning_seeds": "310000..310199",
   "holdout_seeds": "410000..410499",
   "tuning": {
    "overall": {
     "n": "200",
     "sum_d": "15.84227",
     "sum_d2": "654.499282",
     "mean_diff": "0.079211",
     "se": "0.128114",
     "retail_diff": "0.02612",
     "arb_diff": "0.053092"
    },
    "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": "15.84227",
     "sum_d2": "654.499282",
     "mean_diff": "0.277935",
     "se": "0.451292",
     "retail_diff": "0.091648",
     "arb_diff": "0.186287"
    },
    "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": {
    "overall": {
     "n": "500",
     "sum_d": "7.27008",
     "sum_d2": "11372.050574",
     "mean_diff": "0.01454",
     "se": "0.213492",
     "retail_diff": "-0.024021",
     "arb_diff": "0.038561"
    },
    "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": "7.27008",
     "sum_d2": "11372.050574",
     "mean_diff": "0.062673",
     "se": "0.923279",
     "retail_diff": "-0.103539",
     "arb_diff": "0.166212"
    },
    "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": "0.8689",
   "input_result_locator": "artifacts/goal/batches/r3-deepen-current/items/s2-bank-cash-cap/result.json (.imd/reads/artifacts/item-s2-bank-cash-cap)",
   "artifact_sha256": "095404d0e0d180ce7a23df164b8f7fbed3214d35d509ea04a59790a2558d4ac5",
   "code_url": "https://api.imd.fun/artifacts/095404d0e0d180ce7a23df164b8f7fbed3214d35d509ea04a59790a2558d4ac5",
   "flags": [
    "holdout run interrupted by tool timeout after seed 410344; suffix 410345..410499 rerun with same frozen code, no seed replaced (item note)",
    "retail/arb differences reported as per-seed means (retail_diff_mean/arb_diff_mean), same quantity as retail_diff/arb_diff elsewhere",
    "holdout sum_d2 11372 indicates heavy-tailed per-seed d"
   ],
   "protocol_violations": []
  },
  {
   "key": "s3-replenish",
   "idea": "Tags stock purchases (m-prefix client IDs) and subtracts still-resting purchases from the stock-target shortfall before sizing new ones.",
   "target_spreads": [
    3
   ],
   "code_sha256": "13fc73b52cbdfcf7c8033d863c759e289e7b2a61876d6e6b2d8f9617c7de5d3c",
   "incumbent_code_sha256": "bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae",
   "comparator_matches_B": true,
   "code_bytes_verified": true,
   "complete": true,
   "failures": [],
   "tuning_seeds": "310000..310199",
   "holdout_seeds": "410000..410499",
   "tuning": {
    "overall": {
     "n": "200",
     "sum_d": "0.8991560260997602",
     "sum_d2": "11.190155378133028",
     "mean_diff": "0.004495780130498801",
     "se": "0.016764787944754316",
     "retail_diff": "-0.004324676340999929",
     "arb_diff": "0.008820456472500009"
    },
    "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": "0.8991560260997602",
     "sum_d2": "11.190155378133028",
     "mean_diff": "0.016965208039618117",
     "se": "0.06367694741227271",
     "retail_diff": "-0.016319533362263883",
     "arb_diff": "0.033284741405660415"
    },
    "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": {
    "overall": {
     "n": "500",
     "sum_d": "-5.311662379399934",
     "sum_d2": "82.02061382576726",
     "mean_diff": "-0.010623324758799868",
     "se": "0.01812494890450146",
     "retail_diff": "-0.019230708051600147",
     "arb_diff": "0.008607383293200009"
    },
    "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": "-5.311662379399934",
     "sum_d2": "82.02061382576726",
     "mean_diff": "-0.03849030709710097",
     "se": "0.06578391802892262",
     "retail_diff": "-0.06967647844782662",
     "arb_diff": "0.031186171352173945"
    },
    "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": "1.3895931366086005",
   "input_result_locator": "artifacts/goal/batches/r3-deepen-current/items/s3-replenish/result.json (.imd/reads/artifacts/item-s3-replenish)",
   "artifact_sha256": "b88bd341fb497524f2ec16b7010c693e5dd6823b330db9baa7e7d88e3bc8fb1a",
   "code_url": "https://api.imd.fun/artifacts/b88bd341fb497524f2ec16b7010c693e5dd6823b330db9baa7e7d88e3bc8fb1a",
   "flags": [
    "item flagged by the requester: 1 cited source did not hold when re-read; sources.json not supplied to this merge so the exact source is unidentified. Result is a measured simulation, not dependent on the quote"
   ],
   "protocol_violations": []
  },
  {
   "key": "s3-premium-payback",
   "idea": "Gates the ammo \"premium-to-phat\" purchase on whether the premium cost is covered by estimated remaining retail edge. Selected C2 (stochastic demand bound); the other variant was mean-demand bound.",
   "target_spreads": [
    3
   ],
   "code_sha256": "c40fc2edccf131afd789157002de0c07ad6b5b0c3990ba6e424bf5f298b4200d",
   "incumbent_code_sha256": "bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae",
   "comparator_matches_B": true,
   "code_bytes_verified": true,
   "complete": true,
   "failures": [],
   "tuning_seeds": "310000..310199",
   "holdout_seeds": "410000..410499",
   "tuning": {
    "overall": {
     "n": "200",
     "sum_d": "0.0",
     "sum_d2": "0.0",
     "mean_diff": "0.0",
     "se": "0.0",
     "retail_diff": "0.0",
     "arb_diff": "0.0"
    },
    "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": "0.0",
     "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": {
    "overall": {
     "n": "500",
     "sum_d": "0.0",
     "sum_d2": "0.0",
     "mean_diff": "0.0",
     "se": "0.0",
     "retail_diff": "0.0",
     "arb_diff": "0.0"
    },
    "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": "0.0",
     "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": "0.9264713299836148",
   "input_result_locator": "artifacts/goal/batches/r3-deepen-current/items/s3-premium-payback/result.json (.imd/reads/artifacts/item-s3-premium-payback)",
   "artifact_sha256": "93c81955d1f538564bb67deec643d64e59ffba682e4e0f6628238340413f5968",
   "code_url": "https://api.imd.fun/artifacts/93c81955d1f538564bb67deec643d64e59ffba682e4e0f6628238340413f5968",
   "flags": [
    "selected C2 identical to B on every one of 700 tuning+holdout seeds (all d exactly 0): gate never changed behaviour"
   ],
   "protocol_violations": []
  },
  {
   "key": "s2-late-uncovered",
   "idea": "Late (H<300) and phat<0.05: scales the terminal uncovered-ask allowance above the ordinary budget by retail_gain/(retail_gain+jump_loss).",
   "target_spreads": [
    2
   ],
   "code_sha256": "e065ea32b5ea8048e2eba688ab3924c3781cd2a0a2c921045d72e6a6ae966466",
   "incumbent_code_sha256": "bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae",
   "comparator_matches_B": true,
   "code_bytes_verified": true,
   "complete": true,
   "failures": [],
   "tuning_seeds": "310000..310199",
   "holdout_seeds": "410000..410499",
   "tuning": {
    "overall": {
     "n": "200",
     "sum_d": "0.6132396374999978",
     "sum_d2": "0.45368865723949336",
     "mean_diff": "0.0030661981874999886",
     "se": "0.0033692682315963527",
     "retail_diff": "0.006102119061499991",
     "arb_diff": "-0.003035920874500002"
    },
    "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": "0.6132396374999978",
     "sum_d2": "0.45368865723949336",
     "mean_diff": "0.010758590131578908",
     "se": "0.011834953426027427",
     "retail_diff": "0.021410944075438566",
     "arb_diff": "-0.010652353945614042"
    },
    "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": {
    "overall": {
     "n": "500",
     "sum_d": "-0.17130929729997035",
     "sum_d2": "0.060252538164865914",
     "mean_diff": "-0.00034261859459994073",
     "se": "0.0004911801028580245",
     "retail_diff": "8.755622000649055e-07",
     "arb_diff": "-0.0003434941565999754"
    },
    "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": "-0.17130929729997035",
     "sum_d2": "0.060252538164865914",
     "mean_diff": "-0.00147680428706871",
     "se": "0.002120782221469108",
     "retail_diff": "3.773975000279765e-06",
     "arb_diff": "-0.0014805782612067905"
    },
    "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": "0.4131623820774257",
   "input_result_locator": "artifacts/goal/batches/r3-deepen-current/items/s2-late-uncovered/result.json (.imd/reads/artifacts/item-s2-late-uncovered)",
   "artifact_sha256": "58219f8f21fc34c3711c66d527a069ba37d8099fce1925ddb3229c6e80d15869",
   "code_url": "https://api.imd.fun/artifacts/58219f8f21fc34c3711c66d527a069ba37d8099fce1925ddb3229c6e80d15869",
   "flags": [],
   "protocol_violations": []
  },
  {
   "key": "s3-recollateralize",
   "idea": "Cancel and repost a kept inside-rival ask whose size is uncovered, once YES inventory can cover it; selected C2 fires only when cash binds.",
   "target_spreads": [
    3
   ],
   "code_sha256": "958419fb66f7d42609e75c67dd4f143d435e31d3815f49ab2f107296720c3bab",
   "incumbent_code_sha256": "bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae",
   "comparator_matches_B": true,
   "code_bytes_verified": true,
   "complete": true,
   "failures": {
    "missing_seeds": [],
    "failed_or_exception": []
   },
   "tuning_seeds": "310000..310199",
   "holdout_seeds": "410000..410499",
   "tuning": {
    "overall": {
     "n": "200",
     "sum_d": "-1.7906260918999806",
     "sum_d2": "2.6336071230181277",
     "mean_diff": "-0.008953130459499904",
     "se": "0.008109761185308898",
     "retail_diff": "0.0",
     "arb_diff": "-0.00895313045950001"
    },
    "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": "-1.7906260918999806",
     "sum_d2": "2.6336071230181277",
     "mean_diff": "-0.033785397960376995",
     "se": "0.030555512948051193",
     "retail_diff": "0.0",
     "arb_diff": "-0.0337853979603774"
    },
    "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": {
    "overall": {
     "n": "500",
     "sum_d": "163.31138959329996",
     "sum_d2": "24505.3287948289",
     "mean_diff": "0.32662277918659993",
     "se": "0.31305583881470794",
     "retail_diff": "0.25087562325840007",
     "arb_diff": "0.07574715592879999"
    },
    "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": "163.31138959329996",
     "sum_d2": "24505.3287948289",
     "mean_diff": "1.1834158666181156",
     "se": "1.1339948545370986",
     "retail_diff": "0.9089696494869568",
     "arb_diff": "0.27444621713333334"
    },
    "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": "see item result (dict of timings; first-10 pairs ~0.3-0.8 s/sim)",
   "input_result_locator": "artifacts/goal/batches/r3-deepen-current/items/s3-recollateralize/result.json (.imd/reads/artifacts/item-s3-recollateralize)",
   "artifact_sha256": "e932d858242ddc513e0e162d243b6506367d2e2c755d035b2efda93bc6444031",
   "code_url": "https://api.imd.fun/artifacts/e932d858242ddc513e0e162d243b6506367d2e2c755d035b2efda93bc6444031",
   "flags": [
    "item flagged by the requester: 2 cited sources did not hold when re-read; sources.json not supplied so unidentified",
    "holdout +0.327 dominated by one seed: 410361 d=+156.44 of sum_d +163.31 (item text); only 5 of 138 s3 seeds differ from B",
    "repost counts (61 C2 / 797 C1) not reconciled against an independent count (item text)"
   ],
   "protocol_violations": []
  },
  {
   "key": "s2-flow-exposure",
   "idea": "Replaces the binary flow-exposure increment of 1 with 1/last_flow_correction (capacity-fractional exposure) in the spread-2 flow estimator.",
   "target_spreads": [
    2
   ],
   "code_sha256": "5b16ae2cd01882c82c2b7a0a2cdaa75e06ddc18789f163b383b2e8d3313af950",
   "incumbent_code_sha256": "bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae",
   "comparator_matches_B": true,
   "code_bytes_verified": true,
   "complete": true,
   "failures": [],
   "tuning_seeds": "310000..310199",
   "holdout_seeds": "410000..410499",
   "tuning": {
    "overall": {
     "n": "200",
     "sum_d": "-12.56970040359998",
     "sum_d2": "204.72590889546055",
     "mean_diff": "-0.0628485020179999",
     "se": "0.07158224739463911",
     "retail_diff": "0.017620477084500052",
     "arb_diff": "-0.08046897910249987"
    },
    "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": "-12.56970040359998",
     "sum_d2": "204.72590889546055",
     "mean_diff": "-0.22052105971228034",
     "se": "0.25153291017367163",
     "retail_diff": "0.06182623538421071",
     "arb_diff": "-0.28234729509649076"
    },
    "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": {
    "overall": {
     "n": "500",
     "sum_d": "-34.949684876500136",
     "sum_d2": "778.5695787900195",
     "mean_diff": "-0.06989936975300028",
     "se": "0.055773900053975124",
     "retail_diff": "-0.004967749278800056",
     "arb_diff": "-0.06493162047379997"
    },
    "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": "-34.949684876500136",
     "sum_d2": "778.5695787900195",
     "mean_diff": "-0.3012903868663805",
     "se": "0.239946204414547",
     "retail_diff": "-0.021412712408620933",
     "arb_diff": "-0.27987767445603434"
    },
    "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": "1.4010781692676912",
   "input_result_locator": "artifacts/goal/batches/r3-deepen-current/items/s2-flow-exposure/result.json (.imd/reads/artifacts/item-s2-flow-exposure)",
   "artifact_sha256": "9aa791787134864cbf537242ba556fa077d69e6c6f5efa1a671dfea8dc54b026",
   "code_url": "https://api.imd.fun/artifacts/9aa791787134864cbf537242ba556fa077d69e6c6f5efa1a671dfea8dc54b026",
   "flags": [],
   "protocol_violations": []
  },
  {
   "key": "s3-censor-cap",
   "idea": "Lowers the cap on the censoring correction in the spread-3 flow estimate from 3.0 to 1.5 (B cap never binds; max factor ~1.823).",
   "target_spreads": [
    3
   ],
   "code_sha256": "e95e7b93ae696671334262cf672b2048c0e652561e254ae2f35b496da612cefe",
   "incumbent_code_sha256": "bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae",
   "comparator_matches_B": true,
   "code_bytes_verified": true,
   "complete": true,
   "failures": {
    "tuning_failed": 0,
    "holdout_failed": 0,
    "missing_seeds": [],
    "exceptions": [],
    "notes": "Holdout run was interrupted by a tool timeout after 783 records (B for 410391 done); resumed with C for 410391 then seeds 410392..410499; every (seed,strategy) record appears exactly once. Spread label = B's own self.s from the first StepState (verified: equals regime competitor_spread_ticks on all 200 T and 500 H seeds)."
   },
   "tuning_seeds": "310000..310199",
   "holdout_seeds": "410000..410499",
   "tuning": {
    "overall": {
     "n": "200",
     "sum_d": "0.1211102271999529",
     "sum_d2": "0.003702950548811432",
     "mean_diff": "0.0006055511359997645",
     "se": "0.0003019872253092549",
     "retail_diff": "0.0",
     "arb_diff": "0.0006055511350002796"
    },
    "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": "0.1211102271999529",
     "sum_d2": "0.003702950548811432",
     "mean_diff": "0.002285098626414206",
     "se": "0.0011149794388805154",
     "retail_diff": "0.0",
     "arb_diff": "0.0022850986226425644"
    },
    "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": {
    "overall": {
     "n": "500",
     "sum_d": "4.18710480710013",
     "sum_d2": "16.65780109577295",
     "mean_diff": "0.00837420961420026",
     "se": "0.00816236473123924",
     "retail_diff": "-0.00010224528219998774",
     "arb_diff": "0.008476454896800021"
    },
    "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": "4.18710480710013",
     "sum_d2": "16.65780109577295",
     "mean_diff": "0.030341339181885002",
     "se": "0.029569676064322373",
     "retail_diff": "-0.0003704539210144483",
     "arb_diff": "0.030711793104347903"
    },
    "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": "see item result (dict of timings; first-10 pairs ~0.3-0.8 s/sim)",
   "input_result_locator": "artifacts/goal/batches/r3-deepen-current/items/s3-censor-cap/result.json (.imd/reads/artifacts/item-s3-censor-cap)",
   "artifact_sha256": "fc53dce055d3d51588bda8b18399fb69c33ca37143e947189d90807a419f4e5d",
   "code_url": "https://api.imd.fun/artifacts/fc53dce055d3d51588bda8b18399fb69c33ca37143e947189d90807a419f4e5d",
   "flags": [
    "holdout run interrupted by tool timeout after 783 records; resumed, every (seed,strategy) once (item note)",
    "cap chosen on tuning from 1.5 vs 1.25"
   ],
   "protocol_violations": []
  },
  {
   "key": "s2-flow-jump-mask",
   "idea": "Spread-2 flow estimator drops an observation when the rival ask rose at all or the rival bid moved >1 tick (variant CB; CA was ask-only).",
   "target_spreads": [
    2
   ],
   "code_sha256": "91ce2de42cda0eb65638767a546b272549638ee7cf7fb94ed45a9650e5959db7",
   "incumbent_code_sha256": "bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae",
   "comparator_matches_B": true,
   "code_bytes_verified": true,
   "complete": true,
   "failures": {
    "tuning": [],
    "holdout": [],
    "exceptions": [],
    "missing_seeds": []
   },
   "tuning_seeds": "310000..310199",
   "holdout_seeds": "410000..410499",
   "tuning": {
    "overall": {
     "n": "200",
     "sum_d": "3.2198962761000067",
     "sum_d2": "71.60100155957086",
     "mean_diff": "0.016099481380500033",
     "se": "0.04239950071391896",
     "retail_diff": "-0.011044900727497975",
     "arb_diff": "0.027144382107497478"
    },
    "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": "3.2198962761000067",
     "sum_d2": "71.60100155957086",
     "mean_diff": "0.0564894083526317",
     "se": "0.14958077540972065",
     "retail_diff": "-0.03875403764035923",
     "arb_diff": "0.09524344599122969"
    },
    "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": {
    "overall": {
     "n": "500",
     "sum_d": "40.703359575799965",
     "sum_d2": "663.9596087743716",
     "mean_diff": "0.08140671915159993",
     "se": "0.051457556174251844",
     "retail_diff": "6.039209949904034e-07",
     "arb_diff": "0.08140611523159791"
    },
    "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": "40.703359575799965",
     "sum_d2": "663.9596087743716",
     "mean_diff": "0.35089103082586176",
     "se": "0.2206840199352422",
     "retail_diff": "2.6031077595689567e-06",
     "arb_diff": "0.3508884277224098"
    },
    "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": "0.5433347434997559",
   "input_result_locator": "artifacts/goal/batches/r3-deepen-current/items/s2-flow-jump-mask/result.json (.imd/reads/artifacts/item-s2-flow-jump-mask)",
   "artifact_sha256": "fbd2c0ef8a258d64f04248e339e5e400f6ded3cb13f287718f45540542689b76",
   "code_url": "https://api.imd.fun/artifacts/fbd2c0ef8a258d64f04248e339e5e400f6ded3cb13f287718f45540542689b76",
   "flags": [
    "B alone ran once on seeds 310000..310002 as a smoke test before tuning; output discarded, those seeds then rerun in tuning (item note)",
    "variant CB chosen on tuning by higher mean"
   ],
   "protocol_violations": []
  },
  {
   "key": "s3-order-ledger",
   "idea": "Spread-3 demand estimator uses a per-order ledger of quantity reductions on tracked ask IDs instead of aggregate sell_filled_quantity.",
   "target_spreads": [
    3
   ],
   "code_sha256": "565fbb98d04ca3b4fbf12b77fc555b5aeb7376f73995191d7d007dbffe0422f0",
   "incumbent_code_sha256": "bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae",
   "comparator_matches_B": true,
   "code_bytes_verified": true,
   "complete": true,
   "failures": {
    "records": [],
    "exceptions": [],
    "missing_seeds": {
     "tuning": [],
     "paired": []
    },
    "failed_flags": {
     "B": {
      "false": 700,
      "true": 0
     },
     "C": {
      "false": 700,
      "true": 0
     }
    }
   },
   "tuning_seeds": "310000..310199",
   "holdout_seeds": "410000..410499",
   "tuning": {
    "overall": {
     "n": "200",
     "sum_d": "0.06397669199984435",
     "sum_d2": "0.04991024737227553",
     "mean_diff": "0.00031988345999922174",
     "se": "0.0011196035538331244",
     "retail_diff": "0.0",
     "arb_diff": "0.00031988345850007425"
    },
    "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": "0.06397669199984435",
     "sum_d2": "0.04991024737227553",
     "mean_diff": "0.0012071073962234783",
     "se": "0.004252252055682591",
     "retail_diff": "0.0",
     "arb_diff": "0.001207107390566318"
    },
    "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": {
    "overall": {
     "n": "500",
     "sum_d": "4.8591749550001175",
     "sum_d2": "19.285849941144352",
     "mean_diff": "0.009718349910000234",
     "se": "0.00878115735901359",
     "retail_diff": "-0.0005645668033999982",
     "arb_diff": "0.010282916714200006"
    },
    "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": "4.8591749550001175",
     "sum_d2": "19.285849941144352",
     "mean_diff": "0.03521141271739216",
     "se": "0.03179687954894194",
     "retail_diff": "-0.002045531896376805",
     "arb_diff": "0.03725694461666669"
    },
    "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": "0.6188083949964494",
   "input_result_locator": "artifacts/goal/batches/r3-deepen-current/items/s3-order-ledger/result.json (.imd/reads/artifacts/item-s3-order-ledger)",
   "artifact_sha256": "ad39b02ccf9f9ff80e04962d0aa6bdf927b7b18959ca7dc379e8cfc983db4bb3",
   "code_url": "https://api.imd.fun/artifacts/ad39b02ccf9f9ff80e04962d0aa6bdf927b7b18959ca7dc379e8cfc983db4bb3",
   "flags": [
    "item flagged by the requester: 1 cited source did not hold when re-read; the only URLs in the result are the simulator tree and incumbent artifact (both verified-looking), sources.json not supplied"
   ],
   "protocol_violations": []
  },
  {
   "key": "s1-two-cent-commit",
   "idea": "Spread 1: caps a new 2-cent ask size at K*fl_rate*H/0.02 shares (K=1 chosen; K=0.5 worse on tuning).",
   "target_spreads": [
    1
   ],
   "code_sha256": "be4d12fdc949cea6dab1e38f20926b42b782d584e7d80a96a5d9d9f42fa45d41",
   "incumbent_code_sha256": "bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae",
   "comparator_matches_B": true,
   "code_bytes_verified": true,
   "complete": true,
   "failures": [],
   "tuning_seeds": "310000..310199",
   "holdout_seeds": "410000..410499",
   "tuning": {
    "overall": {
     "n": "200",
     "sum_d": "-0.8217439519999914",
     "sum_d2": "0.6752631226485641",
     "mean_diff": "-0.004108719759999957",
     "se": "0.0041087197599999565",
     "retail_diff": "-0.004108719759500161",
     "arb_diff": "0.0"
    },
    "s1": {
     "n": "47",
     "sum_d": "-0.8217439519999914",
     "sum_d2": "0.6752631226485641",
     "mean_diff": "-0.017483913872340242",
     "se": "0.017483913872340242",
     "retail_diff": "-0.017483913870213454",
     "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": "0.0",
     "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": {
    "overall": {
     "n": "500",
     "sum_d": "3.6078820405000016",
     "sum_d2": "92.24362687881194",
     "mean_diff": "0.007215764081000003",
     "se": "0.019225233215675083",
     "retail_diff": "-0.009486790194399988",
     "arb_diff": "0.01670255427539999"
    },
    "s1": {
     "n": "115",
     "sum_d": "3.6078820405000016",
     "sum_d2": "92.24362687881194",
     "mean_diff": "0.03137288730869567",
     "se": "0.0838301455503026",
     "retail_diff": "-0.04124691388869559",
     "arb_diff": "0.07261980119739127"
    },
    "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": "0.0",
     "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": "0.637",
   "input_result_locator": "artifacts/goal/batches/r3-deepen-current/items/s1-two-cent-commit/result.json (.imd/reads/artifacts/item-s1-two-cent-commit)",
   "artifact_sha256": "9a63b7ec09d7a73262f865119923608f77df467fcbf12cbdad7c6a804501d05e",
   "code_url": "https://api.imd.fun/artifacts/9a63b7ec09d7a73262f865119923608f77df467fcbf12cbdad7c6a804501d05e",
   "flags": [
    "two variants tried (K=1 chosen, K=0.5); holdout nonzero only on s1"
   ],
   "protocol_violations": []
  }
 ]
}
```