{ "key": "f-retail-clock", "mech": "retail clock", "approach": "Blind overlay counts filled sides over the previous 100 steps (buy>0 plus sell>0), divides this rate by 0.253, the midpoint of the simulator randomization band [0.154,0.352], and clips to tuned size bounds. Warm up for 100 observations. Resize only N placements; retain prices and cancellation actions. Cap resized orders using a conservative collateral ledger through cancellations and placements. Own fill counts are a censored proxy and include arbitrage; actual retail arrivals are not observable.", "tried": [ "Three bound settings: [0.9,1.1], [0.75,1.25], [0.5,1.5], on all four spreads.", "The initial pilot failed on wide bounds at seed 320001. A cancel-all-only guard later failed on mild bounds at seed 320196. The final repair caps every placement with a conservative action-by-action collateral ledger. Earlier revision observations are excluded from the final tuning table; cached N observations are reused.", "Selected the highest overall paired tuning mean; no spread gating or parameter changes after selection.", "Checked low/high-rate scaling and collateral clipping with synthetic StepStates and the simulator market." ], "next": "Keep N: this screen does not establish an improvement. Own fills mix retail and arbitrage and depend on quoting exposure; investigate an estimator that accounts for those limitations in a separately assigned item.", "spreads_acted": [ 1, 2, 3, 4 ], "off_equals_n": { "passed": true, "seeds": { "start": 310000, "end": 310009, "inclusive": true }, "callbacks": 20000, "method": "One engine run per seed, forwarding each actual N state to separate N and disabled-overlay instances and asserting action equality every step. Disabled early-return path unchanged by both collateral repairs." }, "code": "from collections import deque as _ClockDeque\nfrom dataclasses import replace as _clock_replace\nfrom orderbook_pm_challenge.types import PlaceOrder as _ClockPlace, CancelAll as _ClockCancelAll, CancelOrder as _ClockCancel, Side as _ClockSide\nfrom orderbook_pm_challenge.utils import quantize_down as _clock_floor\n\nclass Strategy(_N):\n # Midpoint of the documented randomized retail arrival-rate band.\n clock_prior = (0.154 + 0.352) / 2.0\n clock_lo = 0.9\n clock_hi = 1.1\n clock_enabled = True\n clock_spreads = (1, 2, 3, 4)\n\n def on_step(self, state):\n actions = super().on_step(state)\n if not self.clock_enabled or self.s not in self.clock_spreads:\n return actions\n if not hasattr(self, '_retail_clock_fills'):\n self._retail_clock_fills = _ClockDeque(maxlen=100)\n if state.step > 0:\n # Only aggregate side quantities are observable: count filled sides.\n self._retail_clock_fills.append(\n int(state.buy_filled_quantity > 0) +\n int(state.sell_filled_quantity > 0))\n if len(self._retail_clock_fills) < 100:\n return actions\n rate = sum(self._retail_clock_fills) / 100.0\n scale = min(self.clock_hi, max(self.clock_lo, rate / self.clock_prior))\n # Track a conservative collateral budget through the action sequence.\n # Existing sell-order cash reservations are not exposed in StepState;\n # individual cancellations therefore need not release their collateral.\n # CancelAll restores the exact starting cash and inventory budget.\n free = max(0.0, state.free_cash)\n yes = max(0.0, state.yes_inventory - sum(\n o.remaining_quantity for o in state.own_orders\n if o.side == _ClockSide.SELL))\n reservations = {\n o.order_id: (round(o.price_ticks / 100.0 * o.remaining_quantity, 10), 0.0)\n if o.side == _ClockSide.BUY else (0.0, 0.0)\n for o in state.own_orders}\n result = []\n for a in actions:\n if isinstance(a, _ClockCancelAll):\n free = max(0.0, state.cash)\n yes = max(0.0, state.yes_inventory)\n reservations.clear()\n result.append(a)\n elif isinstance(a, _ClockCancel):\n cash_refund, yes_refund = reservations.pop(a.order_id, (0.0, 0.0))\n free += cash_refund\n yes += yes_refund\n result.append(a)\n elif isinstance(a, _ClockPlace):\n q = _clock_floor(a.quantity * scale, 0.01)\n price = a.price_ticks / 100.0\n capacity = (free / price if a.side == _ClockSide.BUY\n else yes + free / (1.0 - price))\n q = min(q, _clock_floor(max(0.0, capacity - 1e-8), 0.01))\n if q < 0.01:\n continue\n if a.side == _ClockSide.BUY:\n cash_used = round(price * q, 10)\n covered = 0.0\n else:\n covered = min(yes, q)\n cash_used = round((1.0 - price) * (q - covered), 10)\n free = max(0.0, free - cash_used)\n yes -= covered\n if a.client_order_id is not None:\n reservations[a.client_order_id] = (cash_used, covered)\n result.append(_clock_replace(a, quantity=q))\n else:\n result.append(a)\n return result\n", "code_sha256": "8bc98c28c0086a79616b2611c92c395c41ba3900931c216332eefd12b087da07", "assembled_sha256": "bd3791eaad2b71e7e03fa777c309b77307d7c9cf6565cefd44afcaac7ef283ec", "tuning": { "seeds": { "start": 320000, "end": 320199, "inclusive": true }, "incumbent_code_sha256": "bf0d179f3e47d28819f6fa7e0d25548c2ced43374c216d36460b00a26a26ffae", "selected": "mild_safe2", "selection_rule": "Maximum paired overall mean difference over all 200 tuning seeds; frozen before any holdout run.", "variants": [ { "name": "mild_safe2", "lo": 0.9, "hi": 1.1, "code_sha256": "8bc98c28c0086a79616b2611c92c395c41ba3900931c216332eefd12b087da07", "paired": { "n": 200, "sum_d": -90.29857212240023, "sum_d2": 4188.587739156367, "mean_diff": -0.45149286061200117, "se": 0.3228258775237606, "incumbent_mean": 98.47071831456549, "candidate_mean": 98.0192254539535, "retail_diff": 0.44666970058000016, "arb_diff": -0.8981625611965002 }, "by_spread": { "1": { "n": 50, "sum_d": -13.995089903400062, "sum_d2": 579.2078202930278, "mean_diff": -0.27990179806800125, "se": 0.4845745322781932, "incumbent_mean": 25.711261479664, "candidate_mean": 25.431359681596, "retail_diff": 0.7010581343319986, "arb_diff": -0.9809599323880005 }, "2": { "n": 50, "sum_d": -17.887509487300026, "sum_d2": 540.6996399134017, "mean_diff": -0.35775018974600054, "se": 0.46699227804664895, "incumbent_mean": 61.219908510538, "candidate_mean": 60.862158320792, "retail_diff": 0.39710946525000124, "arb_diff": -0.7548596550519998 }, "3": { "n": 48, "sum_d": -11.175551633200058, "sum_d2": 510.45967869497554, "mean_diff": -0.23282399235833454, "se": 0.4744620610813396, "incumbent_mean": 130.8342429047854, "candidate_mean": 130.60141891242708, "retail_diff": 0.13947414117291723, "arb_diff": -0.37229813355625035 }, "4": { "n": 52, "sum_d": -47.240421098500086, "sum_d2": 2558.2206002549624, "mean_diff": -0.908469636509617, "se": 0.9738868248097938, "incumbent_mean": 174.37579815333268, "candidate_mean": 173.46732851682307, "retail_diff": 0.5332846415500002, "arb_diff": -1.4417542780115384 } } }, { "name": "medium_safe2", "lo": 0.75, "hi": 1.25, "code_sha256": "7b2a778d31e9ca7af92bbdd61a606f84bcaa88b5e4226d5c348bc4ebbec8e0fd", "paired": { "n": 200, "sum_d": -182.65343786710028, "sum_d2": 10959.940415848332, "mean_diff": -0.9132671893355014, "se": 0.5207534410181903, "incumbent_mean": 98.47071831456549, "candidate_mean": 97.55745112523, "retail_diff": 1.1748860473175, "arb_diff": -2.08815323668 }, "by_spread": { "1": { "n": 50, "sum_d": 6.673378658799955, "sum_d2": 896.322366579946, "mean_diff": 0.13346757317599908, "se": 0.6045513387579834, "incumbent_mean": 25.711261479664, "candidate_mean": 25.844729052839998, "retail_diff": 1.500054194753999, "arb_diff": -1.3665866215680007 }, "2": { "n": 50, "sum_d": -35.545736672600015, "sum_d2": 871.1349006289439, "mean_diff": -0.7109147334520003, "se": 0.5875806198152249, "incumbent_mean": 61.219908510538, "candidate_mean": 60.508993777086, "retail_diff": 1.0829590840840002, "arb_diff": -1.7938738175679998 }, "3": { "n": 48, "sum_d": -74.48671563600003, "sum_d2": 1182.8007681685592, "mean_diff": -1.5518065757500006, "se": 0.6877898463926282, "incumbent_mean": 130.8342429047854, "candidate_mean": 129.28243632903542, "retail_diff": 0.7142655011958335, "arb_diff": -2.266072077022917 }, "4": { "n": 52, "sum_d": -79.29436421730017, "sum_d2": 8009.682380470884, "mean_diff": -1.5248916195634648, "se": 1.7247168509823094, "incumbent_mean": 174.37579815333268, "candidate_mean": 172.85090653376923, "retail_diff": 1.3758038743115388, "arb_diff": -2.9006954938865395 } } }, { "name": "wide_safe2", "lo": 0.5, "hi": 1.5, "code_sha256": "6ec77bb251aa4939f1aebba5c24b6e931ae916d607c70cc0006b6eb0312a7291", "paired": { "n": 200, "sum_d": -433.74922764020005, "sum_d2": 28975.322483843513, "mean_diff": -2.168746138201, "se": 0.8392780835599122, "incumbent_mean": 98.47071831456549, "candidate_mean": 96.3019721763645, "retail_diff": 2.1789323024015, "arb_diff": -4.3476784406140006 }, "by_spread": { "1": { "n": 50, "sum_d": 10.730767574599996, "sum_d2": 2924.0775341524086, "mean_diff": 0.21461535149199992, "se": 1.092044431087129, "incumbent_mean": 25.711261479664, "candidate_mean": 25.925876831156003, "retail_diff": 2.539559469992, "arb_diff": -2.324944118490001 }, "2": { "n": 50, "sum_d": -76.6121045649, "sum_d2": 2750.405801025886, "mean_diff": -1.532242091298, "se": 1.0366778787891957, "incumbent_mean": 61.219908510538, "candidate_mean": 59.68766641924, "retail_diff": 2.4461271763859997, "arb_diff": -3.978369267727999 }, "3": { "n": 48, "sum_d": -193.22867020700002, "sum_d2": 4509.367260412874, "mean_diff": -4.025597295979167, "se": 1.2860935215424014, "incumbent_mean": 130.8342429047854, "candidate_mean": 126.80864560880626, "retail_diff": 0.9629795800875008, "arb_diff": -4.988576876110417 }, "4": { "n": 52, "sum_d": -174.63922044290004, "sum_d2": 18791.471888252345, "mean_diff": -3.358446546978847, "se": 2.6200407726978407, "incumbent_mean": 174.37579815333268, "candidate_mean": 171.01735160635386, "retail_diff": 2.697675159946153, "arb_diff": -6.056121706896154 } } } ], "superseded_revision_rows": 595 }, "paired": { "seeds": { "start": 430000, "end": 430499, "inclusive": true }, "n": 500, "complete": true, "missing_seeds": [], "overall": { "n": 500, "sum_d": 11.583124723699964, "sum_d2": 94798.94306796261, "mean_diff": 0.023166249447399926, "se": 0.6164045815354549, "incumbent_mean": 95.54201511346659, "candidate_mean": 95.56518136291399, "retail_diff": 1.175439528368601, "arb_diff": -1.1522732789392003 }, "by_spread": { "1": { "n": 131, "sum_d": -2.1959440067000293, "sum_d2": 40399.23624530289, "mean_diff": -0.016762931348855186, "se": 1.5402068676754244, "incumbent_mean": 38.277936657049615, "candidate_mean": 38.26117372570076, "retail_diff": 2.308708432766413, "arb_diff": -2.3254713641366407 }, "2": { "n": 120, "sum_d": -107.75560350150003, "sum_d2": 7841.424356210039, "mean_diff": -0.8979633625125003, "se": 0.736439679411548, "incumbent_mean": 77.8558643979175, "candidate_mean": 76.957901035405, "retail_diff": 1.0727727075750018, "arb_diff": -1.970736070090001 }, "3": { "n": 133, "sum_d": 104.56779314780002, "sum_d2": 33874.662598209026, "mean_diff": 0.7862240086300754, "se": 1.3873851660724037, "incumbent_mean": 146.55140363646615, "candidate_mean": 147.33762764509626, "retail_diff": 0.6335602255729328, "arb_diff": 0.15266378299097716 }, "4": { "n": 116, "sum_d": 16.966879084099997, "sum_d2": 12683.619868240661, "mean_diff": 0.14626619900086205, "se": 0.9749923379289188, "incumbent_mean": 120.02204692465259, "candidate_mean": 120.16831312365345, "retail_diff": 0.6231269360491392, "arb_diff": -0.4768607370232764 } } }, "failures": [ { "phase": "superseded_revision_on_tuning_seeds", "seed": 320001, "variant": "wide", "error": "orderbook_pm_challenge.market.OrderBookError: Insufficient free cash for order collateral", "resolution": "Replaced incomplete collateral guards with an action-by-action ledger; the same three bound settings were then screened under safe2 revision IDs. Preflight outcomes excluded from tuning statistics; every (seed, revision ID) ran once; cached N observations reused." }, { "phase": "superseded_revision_on_tuning_seeds", "seed": 320196, "variant": "mild_safe", "error": "orderbook_pm_challenge.market.OrderBookError: Insufficient free cash for order collateral", "resolution": "Replaced incomplete collateral guards with an action-by-action ledger; the same three bound settings were then screened under safe2 revision IDs. Preflight outcomes excluded from tuning statistics; every (seed, revision ID) ran once; cached N observations reused." } ], "check_values": [ 17.116002, 20.017111, 2.390311, 241.484752 ], "python_version": "3.12.3", "seconds_per_simulation": 1.4677585896634553 }