{"key":"cutfn-1","cut_rule":{"features":["prior belief sd before the first own trade"],"bins_bps":["<7","[7,15)",">=15"],"cuts":[0.82,0.9,0.95],"decision":"call arb iff posterior pi exceeds the bin cut"},"cost_table":[{"bin":"prior_sd_bps [7,15)","n":888814,"cost_arb_called_retail":9.0,"cost_retail_called_arb":1.0,"cut":0.9},{"bin":"prior_sd_bps >=15","n":915435,"cost_arb_called_retail":19.0,"cost_retail_called_arb":1.0,"cut":0.95}],"settings_tried":[{"setting":"constant cut 0.90","n":256,"mean_diff":0.0,"se":0.0},{"setting":"fill-size bins: 0.82/0.90/0.95","n":256,"mean_diff":-0.0345,"se":0.0147},{"setting":"fill-size bins: 0.95/0.90/0.82","n":256,"mean_diff":-0.0411,"se":0.0167},{"setting":"prior-sd bins: 0.82/0.90/0.95","n":256,"mean_diff":-0.0122,"se":0.008},{"setting":"arb-reading-distance bins: 0.82/0.90/0.95","n":256,"mean_diff":-0.0278,"se":0.014}],"confirm_paired":{"n":256,"mean_diff":0.0061,"se":0.0092},"holdout_vs_lattice_qm":{"n":2048,"mean_diff":0.3231120211,"se":0.0108508999,"sum":661.7334191233,"sumsq":707.4187500461,"worst5":[[10942,-1.7213492895],[11487,-1.6483963688],[10395,-1.4694172416],[10982,-1.4177883156],[10885,-1.3345858494]]},"holdout_vs_thr09":{"n":2048,"mean_diff":0.0002201012,"se":0.0033094149,"sum":0.4507673514,"sumsq":45.9146393486,"worst5":[[11137,-0.9453457456],[10886,-0.7433505179],[10822,-0.6728473365],[10627,-0.6553340531],[11526,-0.5903134381]]},"misread_rate":{"base":0.288834005,"mine":0.3048761692},"price_error_bps":{"base":13.2771527568,"mine":13.2626479178},"gas_estimate":null,"code_zb64":"eNrtV1Fv2zYQzrN+xaF5iJzKiuzE8eLEBbq2ax/aNUj7lKJwaImKOUuiKtFx8u93R0qyZSnNhmIdhumARDJ5PH68u+94unrz8vWHN24c7P1z4qGceh49B+NR7anldHy6NxgNxyejE+94ONrzBsfD0/EeeHs/QVa5YhnAnoiD/lpmS5616z01/x+VffBXKkz6A5hHIgkgzYTM+nlAw5Z1uVJwoxVmqYyE/+CmDzcw57kIOKgFh1xGdzyAjKc4yyAUEc+BoZ1slYBQwHKrtn7yeZG9Z0oJn9/AWqgF3HyLp9/i2XLA+yP3j1wmN65lfUbTkVGDVOaKEyoQOawSf8GSWx648BKt44ZZrkCuE1AZQ0wyxEHUTx0NL12wnFsaj7hjEazSgCkOPovwPUeoG+OIMgfP/WXo4P8zz9GnwLcRhDir3WJhqiQBywII+J1gSsgEfRHJNYxhnuYOjPv9wci80mqmIOIM8ZlBF+hYxg+W3l0m0YMGM2f+EmEjtnONG72X8Owgx1OtOIQRu6XDJ1JByvIcHa4kete1fjz+O8H9N/g/Ph5t8f9kgPwfoX7H/58hz549+5WI74CPVOivubhdKEwwls2PMq6YiIgZmgOqRkqdMIapGxIV/Mp4zESSm1TmIc944m/WGiUXPlL2C2SdZumRZqilqWk4nk+Io5qTcKRJaR5IpgdDSMA69RgBcZYpq86/l2hbJCkxnWXcEA4B+jLJkWg+nTvMZAwsxPN8WrO0ImfuWugpS8SpzBTETC3K96L0DbHUwZWll9NBoZguit2l9pcD714ZlYxjNSpU0oVw4C3Wnat3Hx14/3r2Qf//ZFmWHyHdYVMz7Zq53sQClICHWElmugDaOY9CB4vFbL7C7dg9/uFTOYBv+EKlZPq7THixlkTCFGiZK6uhHIce0Nv31UigcEhB32gqxKWqOdwN0wXnf2MRltvKCGmaAj2l9S+mMNisCcuhDZBqkZ/QEq82ERFMcrwbyVt9TDdP4RDsgeuVsMJ5r3cOC9GmeVTXZL1ezbxc4qLPWG1ro5T2M0wZyCgh7VgkdkBOGDhw7PV6degV/ECEIWau3WtM6120TiKz2GBDetmRdBB2r7kA3URVXy4nMMecWVo7kzTxnT2jtIjMxh0tLiNDJmOAYwC3lZ/w2gZeA/k+MVNgagYCaWXHMuARxCJH4/6iN0EGUOFV2ATgtc0yucILk+oF4cWTBiK5bfduikEw58JYpxRrRHhuCsLsjvLGrWcOnamJb+ZsLdGGYxnzROUtYYsKLxpP8yQvATQ0eUrUke4tV/YBVc8Daie81sCS7guabYIjiWMH7v4CuG2ARbDQcg/jbEaf640OqczYBW5UieMeBvfE4/2T6qVhOkCTbJ7btYQoCNWSrPtQ3BhzpIu/mECK1KlaxUT2aWuRYG2lSr+7+nLJeVoe9/LL5Os5cn4miBgzUY6bEf1Kww0j16XiFqtawok4UfEavVKEKlvIA12Ae6WnqjPH7N7G6xH70zOkSl8XaHIZlWh61hRwtR5v7KkT1QQEl5SR0Ug0/3aHMDEMGQc72VwS3+CWGfPxjsvmlGfIqpIPpKLbRm3Da7GxoD62SlX6pVP17OysNVnRMBZrvQhr4jVcTOnAx94jybvL1CZRImP0wsDta9PfsaXr5XWbnTZym2yMxb1aZXyCtMANUkHRKTJ007I8JxQYc4mJiZUPr1ts258qRE/C0gldJXPrNN0tYnO3tKU3ghtgUC+/iK9wWLE7Fe37FVHCII6KSwFbGywaGX80Cpsri6NHhCoUncYeplTmQXmV5N8yZRf1k9w64PXqQT3j1LRuBKtcfQFjt8hrWzd09cnBqJylNm/nqimbDH1G+jptHCIuXPdE0dwHXWkoxibea/Zg+j4aWmMzjA0lD2757lVHpdDYLS9En4h64qED3r3a9bnRxDvq72BLeelhfp+WJnZxmAo8XKDHUP8Qe8fJZmxajlkNulU6L0jnqGXd0e46M/2A7XvQ6O+Ky98EZrIL0jQUTWr69GWMPmj2Ic3dC1N6yUUFtQaa5s53YDbauPYisYXk7bDZ6TyK5MUPIKna2+fbzbCO/II6h0EVX1Ki4mj4sPFL8yxmpVfvuyu6mJf6pG7fqZ8v8ep7t9JZUyexufN7pVPigunnsB7WVQovNDQrkyHWlJBtbnG5XKX22kFDDuF3tqqKoUlhoLfzNTF3yvigrSsXv4/iNLQxeZytX1uxw3K/ypJie8va66STTjrppJNOOumkk0466aSTTjrppJP/lfwJMke0hA==","code_sha256":"d4ba752731822ebe92eedc8e7f26c72111360c5541ca8637b2827d16f9aeac2a","notes":"Tuning used seeds 60000..60255 and confirmation used 60256..60511. The table stores the observed prior-sd bins; costs are normalized loss-ratio weights (retail-called-arb=1), so cut=C_arb_called_retail/(C_arb_called_retail+C_retail_called_arb), not currency edge. The <7 bps branch was unobserved on tuning calls. On the one 2048-seed holdout, the feature cut retains the constant-cut improvement over lattice-QM but is statistically neutral versus constant 0.9; it should not replace that simpler rule on this evidence. This is a Python-only policy, so gas was not measured."}