a tgj%@sdZddlZddlZddlZddlmZdZdZdZ ddZ d Z d d Z d d Z ddZddZddZddZedkr~edS)aDoes gating on a book's GROSS ask-margin help, historically? Today's 7 paper books suggested: only post a 3-leg book if sum(asks) - 100 covers the ~2.4c it costs to complete the last leg (1c crossing + taker fee). That threshold was fitted to 7 points, so test the mechanism on the 641 settled events in kalshi_backfill.csv. THE TAUTOLOGY TRAP, stated up front: if all three legs fill, book P&L IS sum(asks) - 100. Gating on gross would then "work" by definition and prove nothing. The only real question is whether high-gross books still FILL -- a wide book is wide because it is illiquid, and gating toward wide books may select books nobody trades, leaving 2-of-3 partial books, which is the dangerous state. So the test here is fill behaviour vs gross, not P&L vs gross. KNOWN LOOK-AHEAD (do not paper over it): kalshi_backfill snaps the LAST candle at or before kickoff, but we would post hours earlier. Using that near-kickoff ask as our posting price and then counting trades over the whole pre-KO window counts fills at prices the book only reached later -- it inflates fill rates. Mitigated by reporting a NEAR window (last 3h) alongside the full window; the near window is the honest one. This is the same class of error that produced the 83%-vs-19% fill bug in the paper maker. MAKER-SELL SEMANTICS (verified against our own fills): taker_book_side='bid' means the taker BOUGHT, i.e. a maker SOLD. Those are our fills. N) defaultdictz&E:\soccer\research\kalshi_backfill.csvz E:\soccer\research\mm_trades.csv@cCsd|d|dS)NgQ?d)prr+E:\soccer\research\mm_gross_gate_history.py"r g@cCsP|sdSztj|dd}Wnty4YdS0|jr@|S|jtjjdS)NZz+00:00)tzinfo)dtdatetimeZ fromisoformatreplace ValueErrorr timezoneZutc)sdrrr_pt.s rc s,i}ttdd}t|D]z(fdddDfdddDWnttfydYqYn0td}|dustdd Dsqt fd d Dd }|t ks|d krqt d krqd|d|t d d d|d<qWdn1s0Y|S)Nutf-8encodingcs"i|]}|td|dqS)k_Z_askfloat.0rrrr =r zload_books..homedrawawaycs"i|]}|td|dqS)rZ_bidrrrrrr>r Zkickoffcss&|]}d|kodknVqdS)rrNr)rarrr Br zload_books..c3s|]}||VqdSNrr)asksbidsrrr%Dr rr?divftr)r*kor'r(r+spreadgrossZevent) openBACKFILLcsv DictReaderrKeyErrorrallvaluessum MAX_SPREAD_C)booksfhr,r-r)r'r(rr load_books8s*     0r:c Csdd|D}tt}ttdd}t|D]}|d}|d}|dkrRq2|d|}||}|durrq2|d d krq2t |d } | dus2| |krq2z8||||d df | t |d t |dfWq2t yYq2Yq20q2Wdn1s0Y|S)z@Per (event, leg-suffix): maker-sell trades, bounded pre-kickoff.cSsi|]\}}||dqS)r,r)rebrrrrYr zload_fills..rrZticker-rNZtaker_book_sideZbidZtsrZpricecount) itemsrlistr/TRADESr1r2rfindgetrappendrr) r8Zwantoutr9rZtkievr,trrr load_fillsWs,     8 ,rIcsZi}|D]F\}tfdd|D}t|dkr:q dd|DtdkrVq fdd|D}i}|D]0}dd||fD}|rt|nd ||<qptd d |Drq |\} } t|| |d d t|| |d d} t|| |d d t|| |d d} | | kr4| | fn| | f\} }| d|d|<q |S)apMap each event's 3 sides to their ticker suffixes by matching ask prices. Suffixes are team codes of varying length, so string-slicing them is the bug that mis-assigned 24/1060 legs before. Instead: the draw leg's suffix is the one Kalshi tags TIE/DRAW; home/away are resolved by which observed trade prices sit closest to that side's quoted ask. csh|]\}}|kr|qSrr)rr;r)rGrr {r zleg_suffixes..cSsg|]}|dvr|qS))ZTIEZDRAW)upperrrrr ~r z leg_suffixes..rcsg|]}|dkr|qS)rrr)r"rrrMr cSsg|]\}}}|qSrr)r_rrrrrMr Ncss|]}|duVqdSr&r)rvrrrr%r zleg_suffixes..r'r!r#rr )r?sortedlen statisticsZmediananyr5abs)r8fillsresr<ZsufsrestZmedrZpxr$cZd1Zd2r!r#r)r"rGr leg_suffixesqs*  00rYcCsD|tjtd}|D]*\}}}|r,||kr,q||dkrdSqdS)zFWould a resting sell at our_ask have been hit? Trade price >= our ask.)Zhoursg& .>TF)r Z timedeltaNEAR_H)ZtradesZour_askr,Z near_onlylorHrrNrrrfilleds  r\cst}t|t|}tt|dt|ddD]rPdtddnd}g}|D]H\|tfdd d D}|d |d q`td |tdddddddddddddddddd gd}|D]J\fdd|D}|s"qtdd |D}tdd |D}tdd |D} g} |D]d} | dd kr| | d n@| dd!kr`| d"t d# } | | d d$t | q`| rt | ntd%} d&krdd'dd(n d)dd(}td|ddt|d*d|d*d|d*d| d*d| d+d( qtq8td,tdd-ddd.d/dd0d1dd2d1dd3d/ g}|D]\|tfd4d d D}|d krֈd }n0|d!krd d$t t d# }nd}|d ||d5qd6D]fd7d|D}d8d|D} | sRq$td9d |Dt|d:}tdd;dt|d<dt | d=d>t| d?d>|d@dA q$dS)BNz settled books; z* with all 3 legs resolvable in trade data )TFzNEAR window (last gz h, honest)z(FULL pre-KO window (look-ahead inflated)c3s4|],}t|fd|dVqdS)r'r,Nr\r)r<rGrUnearsfxrrr%szmain..r r.)r.nfrGr<r`z--- z z gross bucketz>16s r8z>6sz3/3z2/3z<=1zmean net c/contractz>21s))r))r)@)rdr)r@)reics,g|]$}|dkr krnq|qSr.rrx)hir[rrrMr zmain..css|]}|ddkrdVqdS)rarKrNrrgrrrr%r css|]}|ddkrdVqdSrarNrrgrrrr%r css|]}|ddkrdVqdS)rarNrrgrrrr%r rarKrkr<r(r)nanrcz to rXz< Z6dz20.2fz>GATE SIMULATION (near window): post only books with gross >= TTz books postedz>13szmean netz>10stotalz 2/3 strandedc3s4|],}t|fd|ddVqdS)r'r,TNr^r)r<rGrUr`rrr%s)r.ranet)rcrrrkg@rKcsg|]}|dkr|qSrfrrg)rmrrrMr cSs g|]}|ddur|dqS)roNrrgrrrrMr css|]}|ddkrdVqdSrjrrgrrrr%r rz6.1fZ13dz9.2fzc z9.1fz12.1f%)r:rIrYprintrQrZr?r6rDminr5FEErRZmeanr)r8ZsmapZlblZrowsraZbucketsr]Zn3Zn2Zn1ZnetsrhZworstZmnnameroZstrandr)rmr<rGrUrir[r_r`rmains|  ,,@0   "&rv__main__)__doc__r1rr rR collectionsrr0rArZrtr7rr:rIrYr\rv__name__rrrrs"     A