{
  "what_if": "Participant orders fill before competitor orders at the same price, on both sides; quote at the competitor price instead of one tick inside",
  "blocked": null,
  "python_version": "3.12.3",
  "base_mean_edge": 109.27254289017199,
  "base_unseen": {
    "mean_edge": 98.28063388702982,
    "retail_edge": 126.01907227628422,
    "arb_edge": -27.73843838925235
  },
  "diff": {
    "mean": 47.859869056410425,
    "standard_error": 1.0281727297941858,
    "simulations": 4000,
    "seed_start": 50000,
    "retail": 56.70016212783635,
    "arb": -8.84029307142165
  },
  "by_spread": [
    {
      "spread": 1,
      "runs": 1016,
      "diff": 47.81785765661555,
      "standard_error": 2.6594672731265394
    },
    {
      "spread": 2,
      "runs": 998,
      "diff": 48.32466608308256,
      "standard_error": 2.001400493789408
    },
    {
      "spread": 3,
      "runs": 1017,
      "diff": 49.026854450867944,
      "standard_error": 1.825312511198579
    },
    {
      "spread": 4,
      "runs": 969,
      "diff": 46.20041807953684,
      "standard_error": 1.525828180258
    }
  ],
  "how": "Simulator https://github.com/danrobinson/prediction-market-challenge at commit b244530ee2a6247b5f930830eeca9dd950da0e7e; base code extracted verbatim from the supplied artifact, SHA256 baef955ec10728839f7c3357016ec9eb0550e7ec689bbd5171b1cdd5c9e37273. Original base scored from the unmodified clone with python3 -m orderbook_pm_challenge run ../base.py --simulations 600 --seed-start 1000 --workers 4 --json, then --simulations 4000 --seed-start 50000. No sandbox or dependencies. Scratch clone market.py PredictionMarket._best_order changes exactly two keys: asks use min(..., key=lambda order: (order.price_ticks, order.owner != PARTICIPANT, order.sequence)); bids use max(..., key=lambda order: (order.price_ticks, order.owner == PARTICIPANT, -order.sequence)). This prioritizes the participant only at equal prices and preserves FIFO within each owner. All arbitrage and retail matching paths use this function. First experiment changes only these keys and runs byte-identical base.py. Second experiment retains those keys and makes exactly three strategy edits: top(orders, comp, step) returns comp if 1 <= comp <= 99 else comp + step; add_b removes the and bid_t != cb guard; add_a removes the and ask_t != ca guard. This lets main quotes join and replenish at the competitor price. Missing-side boundary sentinels remain handled as before. The qbid/qask sizing functions, collateral and inventory logic, probes, belief filter and cancellation thresholds are unchanged, but naturally evaluate at the new target price; the two add-quantity guards are the stated exceptions. Top-level diff and by_spread describe the second experiment; priority-only results are reported in tried. Each uses all 4000 holdout seeds in order. Paired mean is statistics.mean(treated.total_edge - base.total_edge); standard_error is statistics.stdev(those differences)/sqrt(n), using sample standard deviation (n-1). Per-spread uses regime.competitor_spread_ticks and the same paired calculation within that group. Retail and arb are signed changes in the simulator edge components; negative arb means greater arbitrage losses. All runs succeeded; no seeds excluded. The original base artifact score field, not its older base_mean_edge field, contains the 109.27254289017199 reference. Additional legal join_idle candidate: Add every existing main-sized order (remaining quantity > PROBE + 1e-9) to self.ahead when its bid is strictly above the current competitor bid or its ask strictly below the current competitor ask; this certifies priority against any later same-tick replenishment. In top(), preserve the original sticky-ahead branch; otherwise let inside = comp + step, calculate qbid(inside) for bids or qask(inside) for asks, and return comp when it is in 1..99 and that inside desired size is <= 0.0, else return inside. Remove the bid_t != cb and ask_t != ca add-quantity guards, so fallback orders can join and replenish at the competitor price. All other base code and simulator rules stay unchanged. No parameters are fitted for join_idle. Selection among four legal candidates uses only training paired means; the retention candidate was already frozen before its holdout, and join_idle did not advance to holdout because it failed to beat that retention candidate on training.",
  "clean": "No probability, jump, retail-arrival or competitor-size information is revealed. Price priority, exogenous seeds/processes, flow sizes, event timing, competitor initial ladder and replenishment rule, collateral, quantization and edge accounting remain unchanged. Priority applies to arbitrage as well as retail. The arbitrageur still removes every stale order; there is no exemption from adverse selection. Changed fills can alter inventory, later actions and competitor depletion/replenishment, which are endogenous consequences of the requested matching change. The at-price variant deliberately changes the quoting target and permits replenishment there, so its gain is the joint priority-plus-repricing intervention, not the priority-only effect. No constants are tuned for either what-if. This is the requested fixed-design empirical ceiling, not a proof that arbitrary redesigned strategies cannot do better. Source excerpts are in sources.json; numerical comparisons and tests are local experiments.",
  "realizable": "Under the published FIFO rules a newly submitted order cannot displace an older competitor order at the same price. The full universal tie-priority gift is unavailable through the strategy API. A real strategy can gain FIFO priority by already resting at a tick before the competitor replenishes there, then preserving that order. The base already does this: self.ahead records main orders placed strictly inside competitor quotes, and top() retains sufficient ahead quantity when the competitor later joins. A strategy may also join behind existing competitor liquidity and wait for older lots to be consumed; it sacrifices intervening retail fills and loses accumulated queue age if it cancels. The ordinary-FIFO at-price control measures that simple joining policy without parameter tuning. The legal attempt extends that mechanism by retaining tiny surviving ahead positions and, optionally, tolerating an oversized ahead order rather than cancelling its queue position. It cannot top up at that price and inherit old priority; new quantity would join the back. The cost is committing collateral and leaving more stale inventory exposed to arbitrage, or keeping a less competitive price instead of moving one tick inside. Pre-positioning at additional future replenishment ticks could win more FIFO positions but would require collateral and expose those quotes to adverse selection before the competitor arrives. The selected attempt tests retention only; it does not claim to reproduce universal priority. An additional fixed, untuned control quotes at the competitor price under ordinary FIFO; its results and the incremental priority effect at those fixed prices are reported in tried. The join_idle candidate tests queue acquisition as well as retention: when the usual inside main quote has zero desired size, it joins the competitor queue and waits. It marks an existing order as ahead only when the public best competitor quote proves there is currently no competitor at that better price. It never assumes newly submitted equal-price quantity inherits another order's age. Waiting still reserves collateral and risks stale fills, and switching back to an inside quote may abandon queue age.",
  "attempt": {
    "mechanism": "Retain naturally acquired FIFO priority at the competitor best price in the original, unmodified simulator.",
    "selected_variant": "keep3",
    "selection": "Largest paired training mean on seeds 4000..7999 among small, keep3, keepall and join_idle. The three retention rules were frozen and selected first; keep3 then received one holdout evaluation. Counterfactual results motivated a separate fixed acquisition hypothesis, join_idle, evaluated on training only. Its training loss rejected it before holdout, leaving keep3 selected. No candidate parameters were fitted on seeds 50000..53999.",
    "training": {
      "small": {
        "score": {
          "mean_edge": 98.6212841640269,
          "retail_edge": 124.23777380580297,
          "arb_edge": -25.616489641791222
        },
        "diff": {
          "mean": -0.12275595912099997,
          "standard_error": 0.06494551503133494,
          "simulations": 4000,
          "seed_start": 4000,
          "retail": -0.09188357534817482,
          "arb": -0.030872383772350027
        },
        "by_spread": [
          {
            "spread": 1,
            "runs": 999,
            "diff": -0.19344865938818834,
            "standard_error": 0.2480507247529551
          },
          {
            "spread": 2,
            "runs": 1033,
            "diff": -0.12994281332758945,
            "standard_error": 0.02790342503381184
          },
          {
            "spread": 3,
            "runs": 993,
            "diff": -0.095444995683484,
            "standard_error": 0.05838494990545434
          },
          {
            "spread": 4,
            "runs": 975,
            "diff": -0.07052391679394897,
            "standard_error": 0.04506743758694934
          }
        ]
      },
      "keep3": {
        "score": {
          "mean_edge": 99.06287696992213,
          "retail_edge": 124.87261425069022,
          "arb_edge": -25.809737280783676
        },
        "diff": {
          "mean": 0.318836846774225,
          "standard_error": 0.11744334454933902,
          "simulations": 4000,
          "seed_start": 4000,
          "retail": 0.5429568695390752,
          "arb": -0.2241200227648
        },
        "by_spread": [
          {
            "spread": 1,
            "runs": 999,
            "diff": -0.08644675722802837,
            "standard_error": 0.2522751758101715
          },
          {
            "spread": 2,
            "runs": 1033,
            "diff": 0.12850976939225578,
            "standard_error": 0.10682729476031994
          },
          {
            "spread": 3,
            "runs": 993,
            "diff": 0.6509333182089627,
            "standard_error": 0.3544922647945048
          },
          {
            "spread": 4,
            "runs": 975,
            "diff": 0.5975182777476925,
            "standard_error": 0.14813964359513573
          }
        ]
      },
      "keepall": {
        "score": {
          "mean_edge": 98.77828443183458,
          "retail_edge": 125.59949588914662,
          "arb_edge": -26.8212114573315
        },
        "diff": {
          "mean": 0.034244308686675023,
          "standard_error": 0.21455748954785678,
          "simulations": 4000,
          "seed_start": 4000,
          "retail": 1.2698385079954753,
          "arb": -1.2355941993126252
        },
        "by_spread": [
          {
            "spread": 1,
            "runs": 999,
            "diff": -0.08279022977557592,
            "standard_error": 0.2521770330869786
          },
          {
            "spread": 2,
            "runs": 1033,
            "diff": 0.28548934015769617,
            "standard_error": 0.6070088901851167
          },
          {
            "spread": 3,
            "runs": 993,
            "diff": -0.3765823896718028,
            "standard_error": 0.500061299159112
          },
          {
            "spread": 4,
            "runs": 975,
            "diff": 0.30637999882430816,
            "standard_error": 0.18790373254955628
          }
        ]
      },
      "join_idle": {
        "score": {
          "mean_edge": 86.69346832300424,
          "retail_edge": 125.98151199151152,
          "arb_edge": -39.28804366852835
        },
        "diff": {
          "mean": -12.05057180014365,
          "standard_error": 0.4231517574744164,
          "simulations": 4000,
          "seed_start": 4000,
          "retail": 1.651854610360375,
          "arb": -13.702426410509474
        },
        "by_spread": [
          {
            "spread": 1,
            "runs": 999,
            "diff": -17.068800545064967,
            "standard_error": 0.904950117978275
          },
          {
            "spread": 2,
            "runs": 1033,
            "diff": -13.121244950545304,
            "standard_error": 0.8214447180019745
          },
          {
            "spread": 3,
            "runs": 993,
            "diff": -9.150316739684895,
            "standard_error": 0.6320245546191198
          },
          {
            "spread": 4,
            "runs": 975,
            "diff": -8.728251179112103,
            "standard_error": 0.9677781408734552
          }
        ]
      }
    },
    "edits": "Variants small, keep3, keepall all replace P['min_ahead'] in top() with 0.01. small keeps the existing size tolerance 1.5. keep3/keepall replace max(q * P['keep_hi'], 1.0) inside manage() with max(q * (3.0 or 1e100, respectively, if t == comp and o.order_id in self.ahead else P['keep_hi']), 1.0). Positive desired quantity q is still required. All other code is unchanged. No equal-price top-ups are added.",
    "score": {
      "mean_edge": 98.53329672951413,
      "retail_edge": 126.55717016111602,
      "arb_edge": -28.023873431600325
    },
    "diff": {
      "mean": 0.25266284248429993,
      "standard_error": 0.10793883644096303,
      "simulations": 4000,
      "seed_start": 50000,
      "retail": 0.5380978848318,
      "arb": -0.2854350423479749
    },
    "by_spread": [
      {
        "spread": 1,
        "runs": 1016,
        "diff": 0.28358261021436987,
        "standard_error": 0.09385129978392777
      },
      {
        "spread": 2,
        "runs": 998,
        "diff": -0.0377191826963927,
        "standard_error": 0.1548226784413644
      },
      {
        "spread": 3,
        "runs": 1017,
        "diff": 0.4586356067692235,
        "standard_error": 0.2554837680070487
      },
      {
        "spread": 4,
        "runs": 969,
        "diff": 0.30314011373178495,
        "standard_error": 0.3025456993053135
      }
    ],
    "failures": 0,
    "untuned_join_control": {
      "mechanism": "Same at-competitor-price code under ordinary FIFO; fixed diagnostic, not included in retention-rule selection",
      "score": {
        "mean_edge": -20.248279812295802,
        "retail_edge": 18.171276081870523,
        "arb_edge": -38.419555894150825
      },
      "diff": {
        "mean": -118.52891369932563,
        "standard_error": 2.3307002606258416,
        "simulations": 4000,
        "seed_start": 50000,
        "retail": -107.8477961944137,
        "arb": -10.681117504898475
      },
      "by_spread": [
        {
          "spread": 1,
          "runs": 1016,
          "diff": -63.74442844365197,
          "standard_error": 3.261801712509501
        },
        {
          "spread": 2,
          "runs": 998,
          "diff": -102.87014006292114,
          "standard_error": 4.1996941226002
        },
        {
          "spread": 3,
          "runs": 1017,
          "diff": -141.86515800319478,
          "standard_error": 4.8261657663969695
        },
        {
          "spread": 4,
          "runs": 969,
          "diff": -167.6058307807097,
          "standard_error": 5.475612622644718
        }
      ]
    },
    "rejected_join_candidate": {
      "edits": "Add every existing main-sized order (remaining quantity > PROBE + 1e-9) to self.ahead when its bid is strictly above the current competitor bid or its ask strictly below the current competitor ask; this certifies priority against any later same-tick replenishment. In top(), preserve the original sticky-ahead branch; otherwise let inside = comp + step, calculate qbid(inside) for bids or qask(inside) for asks, and return comp when it is in 1..99 and that inside desired size is <= 0.0, else return inside. Remove the bid_t != cb and ask_t != ca add-quantity guards, so fallback orders can join and replenish at the competitor price. All other base code and simulator rules stay unchanged. No parameters are fitted for join_idle.",
      "reason": "Training paired mean did not beat the previously selected retention variant; no holdout evaluation performed."
    }
  },
  "tried": [
    "Scorer verification, seeds 1000..1599: {\"mean_edge\": 109.27254289017199, \"retail_edge\": 135.356137274445, \"arb_edge\": -26.083594384290333}; 600 successes, zero failures; matches requester reference exactly.",
    "Unseen base, seeds 50000..53999: {\"mean_edge\": 98.28063388702982, \"retail_edge\": 126.01907227628422, \"arb_edge\": -27.73843838925235}; 4000 successes, zero failures; matches requester reference exactly.",
    "Unseen base by sampled competitor spread: [{\"spread\": 1, \"runs\": 1016, \"mean_edge\": 35.099921878704336, \"retail_edge\": 62.418918295369586, \"arb_edge\": -27.318996416670473}, {\"spread\": 2, \"runs\": 998, \"mean_edge\": 83.51205221971222, \"retail_edge\": 109.17986677895291, \"arb_edge\": -25.667814559215433}, {\"spread\": 3, \"runs\": 1017, \"mean_edge\": 122.97345583671554, \"retail_edge\": 152.8371980033301, \"arb_edge\": -29.86374216660187}, {\"spread\": 4, \"runs\": 969, \"mean_edge\": 153.82041508580312, \"retail_edge\": 181.90065087952496, \"arb_edge\": -28.080235793747264}].",
    "Priority alone, unchanged base code, seeds 50000..53999: {\"score\": {\"mean_edge\": 100.26022363830623, \"retail_edge\": 127.70085086153465, \"arb_edge\": -27.440627223225025}, \"diff\": {\"mean\": 1.9795897512764, \"standard_error\": 0.11741239655341132, \"simulations\": 4000, \"seed_start\": 50000, \"retail\": 1.681778585250425, \"arb\": 0.2978111660273251}, \"by_spread\": [{\"spread\": 1, \"runs\": 1016, \"diff\": 2.2110265353171257, \"standard_error\": 0.2567768281194803}, {\"spread\": 2, \"runs\": 998, \"diff\": 2.117902543427054, \"standard_error\": 0.19232485647757946}, {\"spread\": 3, \"runs\": 1017, \"diff\": 2.291416886801082, \"standard_error\": 0.309009895477506}, {\"spread\": 4, \"runs\": 969, \"diff\": 1.2672015820500517, \"standard_error\": 0.13218334635178478}]}; zero failures.",
    "Priority plus quoting at competitor price, seeds 50000..53999: {\"score\": {\"mean_edge\": 146.14050294344025, \"retail_edge\": 182.71923440412056, \"arb_edge\": -36.578731460674}, \"diff\": {\"mean\": 47.859869056410425, \"standard_error\": 1.0281727297941858, \"simulations\": 4000, \"seed_start\": 50000, \"retail\": 56.70016212783635, \"arb\": -8.84029307142165}, \"by_spread\": [{\"spread\": 1, \"runs\": 1016, \"diff\": 47.81785765661555, \"standard_error\": 2.6594672731265394}, {\"spread\": 2, \"runs\": 998, \"diff\": 48.32466608308256, \"standard_error\": 2.001400493789408}, {\"spread\": 3, \"runs\": 1017, \"diff\": 49.026854450867944, \"standard_error\": 1.825312511198579}, {\"spread\": 4, \"runs\": 969, \"diff\": 46.20041807953684, \"standard_error\": 1.525828180258}]}; zero failures.",
    "Legal at-competitor-price control with ordinary FIFO, no tuning, seeds 50000..53999: {\"score\": {\"mean_edge\": -20.248279812295802, \"retail_edge\": 18.171276081870523, \"arb_edge\": -38.419555894150825}, \"diff\": {\"mean\": -118.52891369932563, \"standard_error\": 2.3307002606258416, \"simulations\": 4000, \"seed_start\": 50000, \"retail\": -107.8477961944137, \"arb\": -10.681117504898475}, \"by_spread\": [{\"spread\": 1, \"runs\": 1016, \"diff\": -63.74442844365197, \"standard_error\": 3.261801712509501}, {\"spread\": 2, \"runs\": 998, \"diff\": -102.87014006292114, \"standard_error\": 4.1996941226002}, {\"spread\": 3, \"runs\": 1017, \"diff\": -141.86515800319478, \"standard_error\": 4.8261657663969695}, {\"spread\": 4, \"runs\": 969, \"diff\": -167.6058307807097, \"standard_error\": 5.475612622644718}]}; zero failures.",
    "Priority incremental effect holding at-competitor-price code fixed, relative to ordinary FIFO at those prices, seeds 50000..53999: {\"score\": {\"mean_edge\": 146.14050294344025, \"retail_edge\": 182.71923440412056, \"arb_edge\": -36.578731460674}, \"diff\": {\"mean\": 166.38878275573606, \"standard_error\": 2.9768318039805566, \"simulations\": 4000, \"seed_start\": 50000, \"retail\": 164.54795832225005, \"arb\": 1.840824433476825}, \"by_spread\": [{\"spread\": 1, \"runs\": 1016, \"diff\": 111.56228610026751, \"standard_error\": 4.954474360988445}, {\"spread\": 2, \"runs\": 998, \"diff\": 151.1948061460037, \"standard_error\": 5.619396492564223}, {\"spread\": 3, \"runs\": 1017, \"diff\": 190.89201245406272, \"standard_error\": 6.064277385174073}, {\"spread\": 4, \"runs\": 969, \"diff\": 213.80624886024654, \"standard_error\": 6.606754045534711}]}; zero failures.",
    "Training baseline, seeds 4000..7999: {\"mean_edge\": 98.74404012314791, \"retail_edge\": 124.32965738115115, \"arb_edge\": -25.585617258018875}; zero failures.",
    "Legal queue-retention training variant small, seeds 4000..7999: {\"score\": {\"mean_edge\": 98.6212841640269, \"retail_edge\": 124.23777380580297, \"arb_edge\": -25.616489641791222}, \"diff\": {\"mean\": -0.12275595912099997, \"standard_error\": 0.06494551503133494, \"simulations\": 4000, \"seed_start\": 4000, \"retail\": -0.09188357534817482, \"arb\": -0.030872383772350027}, \"by_spread\": [{\"spread\": 1, \"runs\": 999, \"diff\": -0.19344865938818834, \"standard_error\": 0.2480507247529551}, {\"spread\": 2, \"runs\": 1033, \"diff\": -0.12994281332758945, \"standard_error\": 0.02790342503381184}, {\"spread\": 3, \"runs\": 993, \"diff\": -0.095444995683484, \"standard_error\": 0.05838494990545434}, {\"spread\": 4, \"runs\": 975, \"diff\": -0.07052391679394897, \"standard_error\": 0.04506743758694934}]}; zero failures.",
    "Legal queue-retention training variant keep3, seeds 4000..7999: {\"score\": {\"mean_edge\": 99.06287696992213, \"retail_edge\": 124.87261425069022, \"arb_edge\": -25.809737280783676}, \"diff\": {\"mean\": 0.318836846774225, \"standard_error\": 0.11744334454933902, \"simulations\": 4000, \"seed_start\": 4000, \"retail\": 0.5429568695390752, \"arb\": -0.2241200227648}, \"by_spread\": [{\"spread\": 1, \"runs\": 999, \"diff\": -0.08644675722802837, \"standard_error\": 0.2522751758101715}, {\"spread\": 2, \"runs\": 1033, \"diff\": 0.12850976939225578, \"standard_error\": 0.10682729476031994}, {\"spread\": 3, \"runs\": 993, \"diff\": 0.6509333182089627, \"standard_error\": 0.3544922647945048}, {\"spread\": 4, \"runs\": 975, \"diff\": 0.5975182777476925, \"standard_error\": 0.14813964359513573}]}; zero failures.",
    "Legal queue-retention training variant keepall, seeds 4000..7999: {\"score\": {\"mean_edge\": 98.77828443183458, \"retail_edge\": 125.59949588914662, \"arb_edge\": -26.8212114573315}, \"diff\": {\"mean\": 0.034244308686675023, \"standard_error\": 0.21455748954785678, \"simulations\": 4000, \"seed_start\": 4000, \"retail\": 1.2698385079954753, \"arb\": -1.2355941993126252}, \"by_spread\": [{\"spread\": 1, \"runs\": 999, \"diff\": -0.08279022977557592, \"standard_error\": 0.2521770330869786}, {\"spread\": 2, \"runs\": 1033, \"diff\": 0.28548934015769617, \"standard_error\": 0.6070088901851167}, {\"spread\": 3, \"runs\": 993, \"diff\": -0.3765823896718028, \"standard_error\": 0.500061299159112}, {\"spread\": 4, \"runs\": 975, \"diff\": 0.30637999882430816, \"standard_error\": 0.18790373254955628}]}; zero failures.",
    "Training-selected legal variant keep3, unchanged simulator, seeds 50000..53999: {\"score\": {\"mean_edge\": 98.53329672951413, \"retail_edge\": 126.55717016111602, \"arb_edge\": -28.023873431600325}, \"diff\": {\"mean\": 0.25266284248429993, \"standard_error\": 0.10793883644096303, \"simulations\": 4000, \"seed_start\": 50000, \"retail\": 0.5380978848318, \"arb\": -0.2854350423479749}, \"by_spread\": [{\"spread\": 1, \"runs\": 1016, \"diff\": 0.28358261021436987, \"standard_error\": 0.09385129978392777}, {\"spread\": 2, \"runs\": 998, \"diff\": -0.0377191826963927, \"standard_error\": 0.1548226784413644}, {\"spread\": 3, \"runs\": 1017, \"diff\": 0.4586356067692235, \"standard_error\": 0.2554837680070487}, {\"spread\": 4, \"runs\": 969, \"diff\": 0.30314011373178495, \"standard_error\": 0.3025456993053135}]}; zero failures.",
    "Matching checks: original FIFO and priority variants both pass bid/ask tie ordering, participant FIFO, strictly better-priced competitor priority, retail partial-fill accounting and exhaustive stale-order arbitrage sweeps; six upstream market/engine unit tests pass under the priority variant.",
    "Joint what-if sensitivity: positive-gain runs 3811/4000; negative-gain runs 189/4000; median paired gain 24.62237686525; minimum -416.7427191512; maximum 991.3758818111; mean after removing the single largest paired gain 47.62393106872483.",
    "Legal join_idle candidate, original FIFO simulator, training seeds 4000..7999: {\"score\": {\"mean_edge\": 86.69346832300424, \"retail_edge\": 125.98151199151152, \"arb_edge\": -39.28804366852835}, \"diff\": {\"mean\": -12.05057180014365, \"standard_error\": 0.4231517574744164, \"simulations\": 4000, \"seed_start\": 4000, \"retail\": 1.651854610360375, \"arb\": -13.702426410509474}, \"by_spread\": [{\"spread\": 1, \"runs\": 999, \"diff\": -17.068800545064967, \"standard_error\": 0.904950117978275}, {\"spread\": 2, \"runs\": 1033, \"diff\": -13.121244950545304, \"standard_error\": 0.8214447180019745}, {\"spread\": 3, \"runs\": 993, \"diff\": -9.150316739684895, \"standard_error\": 0.6320245546191198}, {\"spread\": 4, \"runs\": 975, \"diff\": -8.728251179112103, \"standard_error\": 0.9677781408734552}]}; zero failures."
  ],
  "next": "The joint ceiling exceeds two points: first attribute the priority gift to initially resting versus replenished competitor lots, and measure how much can be reached by waiting in queue or pre-positioning before a replacement arrives. Use a collateral-limited ladder at predicted replacement ticks as the next concrete prototype; measure lost retail fills while waiting and adverse-selection costs. The legal retention attempt improves holdout edge; reproduce it first, then measure forward queue acquisition without changing matching rules. The join_idle acquisition candidate also failed to beat retention on training; do not spend a holdout evaluation on it. The ordinary-FIFO universal joining control loses edge badly; future builds need to acquire queue priority without abandoning price improvement indiscriminately."
}