a Pj@sdZddlZddlZddlZddlZddlZddlZdZej ej e Z ej e dZej e dZhdZddd Zd Zefd d Zd dZedkreedS)aAccurate per-wallet P&L for Polymarket makers, from the FULL activity ledger. (read-only; touches no account and places nothing) WHY NOT /trades. Reconstructing a position from /trades alone is WRONG for exactly the wallets worth studying. Polymarket lets you SPLIT $1 of collateral into a Yes+No pair and MERGE it back, and neither is a trade -- so a maker who mints inventory instead of crossing a spread has trades that look like unexplained one-sided selling. pm_maker_study.py did this and produced $1.7M on 384 fills, which is not a real number. /activity carries every cash-affecting event: TRADE SPLIT MERGE REDEEM REWARD MAKER_REBATE YIELD so net cash flow over a closed history IS realised P&L, with no position arithmetic required at all. SIGNS. BUY -usdc, SELL +usdc, SPLIT -usdc (pay collateral, receive tokens), MERGE +usdc (return tokens, receive collateral), REDEEM +usdc (settlement). Still-open positions are valued from /positions.currentValue. SUBSIDY IS NOT THE STORY. Kalshi pays makers nothing, so income Polymarket pays for providing liquidity would not transfer to us -- but measured over six high-activity wallets it is 0.000%-0.351% of trade volume. Reported separately so it can be subtracted, but it does not change any ranking. CROSS-CHECK. /positions carries Polymarket's OWN cashPnl per open position. Two independent derivations that agree are worth far more than one that cannot be checked; `pnl_check_delta` is the disagreement and a large value means do not trust the row. Nzhttps://data-api.polymarket.comzpm_maker_study.csvzpm_maker_pnl.csv>Z MAKER_REBATEZREWARDZYIELDc Cst|D]z}ztj||dd}Wn(tyHtd|dYqYn0|jdkr`|S|jdkrpdStd|dqdS)N)paramsZtimeoutg333333?i)rangerequestsget ExceptiontimesleepZ status_codeZjson)ZurlrZtriesirrresearch/pm_maker_pnl.py_get.s     ricCs6d}d}d}t}d}d}||kr,d}qHttd|d|d}|durTd}qH|s\qH|D]} | d } zt| d p~d} WnttfyYq`Yn0|| d 7<| tvr|| 7}q`| d kr|| 7}|| d pd dkr| n| 7}q`| dkr || 8}q`| dvr`|| 7}q`|t |7}t |dkr<qHt dqd} d} ttd|ddpjg}|D]T}z4| t|dpd7} | t|dpd7} WnttfyYn0qp|t || dt |dt || |dt |t | dt | dt||dd|dd|ddt|d S)ajCash-flow P&L. `capped` means the history is INCOMPLETE -- do not use it. The offset cap is not a nuisance, it is a correctness boundary. With only the most recent N events, a SELL or REDEEM whose matching purchase falls outside the window has no cost basis and reads as pure profit, so a truncated wallet's P&L is biased UPWARD without limit. That is what produced an $8.8M row on the first pass. So a wallet is only scored when its ENTIRE history fits under the cap. That biases the sample toward smaller operators -- the tradeoff is an exact number instead of a fictional one. grFTz /activityi)userlimitoffsetNtypeZusdcSizerZTRADEZsideZSELLZSPLIT)MERGEREDEEMg?z /positions)rrZ currentValueZcashPnlrr) wallet pnl_spread pnl_subsidyZ pnl_totalZ trade_volumeZ open_valueZpm_cashpnl_openrowsZsplitsZmergesZredeemscapped) collectionsCounterrDATAr float TypeError ValueErrorSUBSIDYupperlenr r roundsumvaluesint)wZmax_rowsZcashZsubsidyZvoltypesZoffrdatuZopen_valZown_pnlposprrr wallet_pnl?sp    &         r4c svttjdkrttjdndttttdd}fdd|D}t t|ddg}t |dD]\}}t |d }|d |d <|d |d <|d |d <|d |d<|drt d|t||d ddfqn| |t d|t||d dd|d|d|dfqn|j dddttdddd<}tj|t|dd}|||Wdn1s0Yt dt|dtd d|D}t d!t|t|d"t|tt|df|r8td#d$|Dt|} td%d$|Dt|} t d&| t d'd"| d(d|D} | rrtd)d$| Dt| } t d*| dS)+Nrrzutf-8)encodingcs0g|](}t|dkrt|ddkr|qS)fills maker_shareg333333?)r+r").0rZ min_fillsrr s zmain..z maker-dominant wallets with >=z fills rr7avg_legs_per_eventall3_event_sharer6Z fills_sampledrz5 %2d/%d %s SKIPPED - history exceeds the offset capz: %2d/%d %s spread=%12.0f subsidy=%8.0f rows=%-6d COMPLETErrrcSs |d S)Nrr)r.rrrzmain..)keyr,r)newliner5r)Z fieldnamesz wrote z -> cSsg|]}|ddkr|qSrrrr8r.rrrr:r?z. profitable on SPREAD alone: %d of %d (%.0f%%)dcss|]}t|dVqdSr;Nr"rCrrr r?zmain..css|]}t|dVqdS)r<NrFrCrrrrGr?z@ their mean legs/match : %.2f (3.0 = full book like ours)z% their share of matches all3: %.0f%%cSsg|]}|ddkr|qSrBrrCrrrr:r?css|]}t|dVqdSrErFrCrrrrGr?z# LOSERS mean legs/match : %.2f)r'sysargvr+listcsvZ DictReaderopenIN_CSVprint enumerater4appendsortOUTZ DictWriterkeysZ writeheaderZ writerowsmaxr)) srcZwsoutr rr.ZfhZw_Zprofr/Zs3Zlosrr9rmainsR       "2&  rW__main__)r)__doc__rrKosrHr rr!pathdirnameabspath__file__ZHEREjoinrMrRr%rZ ACTIVITY_CAPr4rW__name__exitrrrrs"  J)