{"credit":"Taylor (gtmcknight), github.com/gtmcknight/amm-sim@4812a6382311624e1506aca131ca35c378f930d9 research/solved (replica and LEANBASE).","arms":[{"name":"PIN-live","mode":"live","definition":"At the first retail split callback of a step, infer the rival pre-retail spot by inverting the router allocation from our fill. Compare it with the previous inferred rival spot and its direction; when that live rule identifies an arb-to-band-edge move, set LEANBASE price belief to the corresponding implied price (spot/0.997 or spot*0.997); otherwise clip belief to [spot*0.997, spot/0.997].","report":{"mean_diff":null,"se":null,"n":0}},{"name":"PIN-ORACLE","mode":"oracle","definition":"Same belief update as PIN-live, with the arb-to-band-edge decision supplied by the hidden true price label.","report":{"mean_diff":null,"se":null,"n":0}}],"arb_signal":{"share_first_splits_exact_band_edge":null,"live_precision":null,"live_recall":null},"best_live":null,"ceiling":null,"advance":false,"cost":{"storage_slots":0,"gas_estimate":null,"note":"Python replica only; no Solidity implementation or gas measurement."},"reference_zb64":null,"seeds_used":{"choose":[],"report":[]},"not_done":["The required inversion and candidate arms were not implemented or run; no paired seed results, label precision/recall, best arm, oracle ceiling, or arm source bundle can be reported. The supplied split-test artifact is a different quotient consistency experiment and does not supply this task\u2019s per-step pre-retail rival-spot inference."]}