a ij+@sdZddlZddlZddlZddlZddlmZddlmZej ddej ddddl Z ddl Z dZdZd d Zd d Zd dZddZddZedkredS)a'Is API-Football's "Pinnacle" actually Pinnacle -- and how far behind is it? We have an INDEPENDENT Pinnacle reference sitting on disk: football-data.co.uk publishes Pinnacle opening (PSH/PSD/PSA) and CLOSING (PSCH/PSCD/PSCA) odds for every match. Our live pipeline reads a different source entirely: API-Football's `/odds` endpoint, bookmaker id 4, labelled "Pinnacle". The two have never been compared. TWO QUESTIONS, and they are separable: 1. IS IT REALLY PINNACLE? If API-Football's line is a genuine Pinnacle feed, its last pre-kickoff quote should sit close to Pinnacle's published CLOSE. A large systematic gap would mean the label is wrong -- that we have been computing "CLV vs Pinnacle" against something else entirely, which would undermine every result in this project that references Pinnacle. 2. WHAT DOES THE STALENESS COST? We measured API-Football's line changing only every ~5.8h ([[project-pin-line-staleness]]). Even if the source is genuine, a slow relay means our last pre-kickoff read is an OLD Pinnacle price. This measures, in cents, how far it lands from the real close -- which is the error sitting inside every CLV number we compute. Question 1 is answered by the shape of the distribution (centred and tight = genuine), question 2 by its width. A biased centre would indicate a different book; a wide but unbiased spread indicates lag. Comparison is on DEVIGGED probabilities so bookmaker margin differences cannot masquerade as price differences. Usage: python research\af_vs_real_pinnacle.py N) defaultdict)datetimezE:\soccer\mainzE:\soccer\researchzE:\soccer\live_data\*.csvzE:\soccer\league_data\*.csvc Cs(z t|WSttfy"YdS0dS)N)float TypeError ValueError)xr)E:\soccer\research\af_vs_real_pinnacle.py_f0s r cs(dd|D}t|fdd|DS)NcSsg|] }d|qS)g?r.0rrrr 8zdevig..csg|] }|qSrrr srr r :r)sum)orrrr devig7src sji}ttD]T}z4t|dddd }t|D]fdddD}fddd D}t|r6td d |Dszq6d pd}d }dD]2}zt || }WqWqt yYq0q|sq6|t dt df}t|rtdd |Drt|nd t|f||<q6Wd n1s>0YWqtybYqYq0q|S)z6(date, home_fd, away_fd) -> (open_probs, close_probs).utf-8replacenewlineencodingerrorscsg|]}t|qSrr getr crrr r Drzload_fd..)ZPSHZPSDZPSAcsg|]}t|qSrrrr rr r Er)ZPSCHZPSCDZPSCAcss|]}|dkVqdSNrr vrrr Frzload_fd..ZDateN)z%d/%m/%Yz%d/%m/%yZHomeTeamZAwayTeamcss|]}|dkVqdSr!rr#rrr r%Sr)globLEAGUEopencsv DictReaderallrstriprstrptimedaterTnormr Exception) outpfhopZcldZdtZfmtkeyrr r load_fd=s4  "(. r8c Csi}ttD]j}t|ddddB}t|D]$}|dpDddkrPq2t|d}|dus2|d krpq2t|d t|d t|d g}t|r2td d|Dsq2|d}zt |ddd }Wnt t fyYq2Yn0|tt|dtt|df}| d} ||vsH| ||dkr2t|| f||<q2Wdq1sp0Yq|S)zM(date, home_fd, away_fd) -> (last pre-KO AF pinnacle probs, hours before KO).rrrrZstatusZPREelapsedNrZpin_homeZpin_drawZpin_awaycss|]}|dkVqdSr!rr#rrr r%frzload_af..Z game_date z%Y-%m-%dhomeawaygN@r")r&LIVEr(r)r*rupperr r+rr-r.rrr/r0DSZ fdco_namer) Zbestr3r4rZelrgdr6r7hrsrrr load_afZs0(   4rBc Cs<t}t}t|t|@}tdt|tdt|tdt|dt|dkrltddSggg}}}tt}|D]r}||\}} ||\} } || tdD]B\} } | | | | d}|||t ||| |qq| td td t |d d t |d d |D]*\}}td|ddt |d d q:tdtdtdt |dd|tt|ddd|tt|dddt|dd tdtdd|Dt|dddtdd|Dt|ddd td!t |d"d#td$dS)%Nz,football-data fixtures with Pinnacle close: z3live_data fixtures with an API-Football pin quote: zmatched on (date, home, away):  z-too few matched fixtures to conclude anything)r;Zdrawr<dzDQ1 -- IS IT PINNACLE? (signed: AF minus real Pinnacle close, cents)z mean bias z+.3fz c median rz Z5sz mean zY A near-zero centre means the same book. A persistent bias would mean a different one.z5 Q2 -- WHAT DOES THE LAG COST? (|AF - close|, cents)z median z.2fzc p75 g?zc p90 g?zc max z within 1c: css|]}|dkrdVqdSr!rr rrr r%rzmain..z.0fz% beyond 3c: css|]}|dkrdVqdS)r"Nrr rrr r%r%z) (our last pre-KO read is a median of z.1fzh before kickoff)z For scale: the xlow band is 5-12% of fair. On a 25c leg that is 1.25-3.0c, so an error of this size is a material fraction of the signal being traded.)r8rBsetprintlenrlistappend enumerateabssort statisticsZmeanZmedianitemsintmaxr)fdZafkeysZsignedZabsdZlagsZper_sidek_closeZafprAiZsider6rr$rrr mainvsN      $* & rZ__main__)__doc__r)r&rPsys collectionsrrpathinsertZ daily_signalsr?Znonprice_edge_testr/r=r'r rr8rBrZ__name__rrrr s&  -