{"sources":[{"url":"https://api.imd.fun/artifacts/3d001532cce4c2885307bb007a5170f77dd4b2f8151d041696e04a1c60a0f475","quote":"Returned PORT is the supplied decoded PORT with only the three anchor literals replaced by W[NW..NW+2], and PT arguments in strat_init/strat_after replaced by W[NW+3:]. Vector layout: [1], 19 feature multipliers, 19 feature offsets, W1 row-major, b1, W2 row-major, b2, three anchor constants, then TH_VALUES with named constants replaced from params.constants. Scaling is unchanged. IB used its own port with its frozen 178 head weights."},{"url":"https://api.imd.fun/artifacts/afd5183e9ff3235334e041c00b0cd19079599a2c6464db95f8c1586712105557","quote":"ha_port.py = RS2 port + T toxicity anchor (S[16] mean, S[17] count; cap 674, k_lin 0.712696014807650816, k_quad 1.295240753826310656) + 19-8-2 head whose weights are TH[HEAD0:HEAD0+178] (HEAD0=len(RS2 TH)+3). Same strat_init/strat_after API; plug into the replica STRATEGY block via import. Weights zero => exactly T; HA weights with new columns zero => HA."},{"url":"https://api.imd.fun/artifacts/48cd6a10f39fc22a79c5f12398474bd65282803d59314819acd586f187015808","quote":"run_market(W,seed,mean_size,rate,sigma,n_steps); run_many(W,seeds,procs=8,n_steps=10000) -> rows [edge,retail_edge,arb_edge,retail_volume]. Each policy runs alone against 30 bps Vanilla."},{"url":"https://raw.githubusercontent.com/benedictbrady/amm-challenge/6b6921f8d463314aa7c96234e1be25a25245f704/amm_competition/competition/match.py","quote":"    def _build_configs(self) -> list[amm_sim_rs.SimulationConfig]:\n        \"\"\"Build simulation configs with optional variance.\"\"\"\n        import numpy as np\n\n        configs = []\n        for i in range(self.n_simulations):\n            rng = np.random.default_rng(seed=i)"}]}