""" Does PURE Kalshi-vs-Poly deviation make money, with NO Pinnacle anchor? NO. Question (2026-06-29): the unique/overlap edge is always selected by CLV-vs-pin_fair. Can you instead drop Pinnacle entirely and just trade the two books against EACH OTHER -- when Kalshi and Poly disagree, bet the cheaper one? Tested here. Answer: it needs pin_fair. The two-book midpoint is a worse fair than Pinnacle and carries no edge. Strategy A (PURE X-BOOK): fair := mean(kalshi_mid, poly_price) per side. Bet the CHEAPER book's side when its x-book CLV (vs that two-book avg) is in [lo,hi). Settle at the cheaper book's ask (Kalshi ask; Poly est-ask via near-term half-spread). One bet/fixture (max x-book CLV, first qualifying pre-KO snapshot, non-look-ahead). RESULTS on poly_backfill (975 resolved fixtures, 2026-06-29): A) PURE x-book deviation, NO Pinnacle (bet cheaper book): [3,12)% n=180 +1.4% z+0.54 | >=5% n=67 -23.1% z-0.60 | >=10% n=21 -52.4% z-1.16 -> flat-to-negative, and gets MONOTONICALLY WORSE as the deviation widens (opposite of a real edge). B) FALSIFICATION (bet the DEARER book): also loses (-29% to -42%). Deviation is non-directional noise -- neither leg is profitable. C) Crux: of 67 pure-dev bets, 57 are ones where Pinnacle ALSO says cheap=cheap, and even those lose (-9.6%). The two-book avg is contaminated: when both books are cheap vs Pinnacle, their midpoint is cheap too, so dev-from-midpoint never catches the real mispricing. D) Reference -- the live pin_fair signal on the SAME 975 fixtures: +11.0% ROI, 25.2% WR, n=421, z+1.26. Positive, on 6x the sample. VERDICT: Pinnacle (a SHARP book) is the truth anchor doing all the discriminating. Kalshi & Poly are both soft venues; their mutual disagreement says nothing about which is right. Pure cross-book deviation is NOT a standalone strategy. Don't reopen unless a genuinely sharp third reference (not two soft books) is available. Consistent with the unique/overlap mechanism finding (edge needs the 2nd book to CONFIRM Pinnacle). Caveat: pure-dev n is small (~63-67 in-band) because the two books usually price within ~1-2c; big divergences are rare and noisy -- itself part of the point. Read-only, trades nothing. Multiplicative de-vig. Mirrors kalshi_poly_unique_*.py. """ import csv, glob, os, math from datetime import datetime from collections import defaultdict DIR = r"E:\soccer\poly_backfill" _FMTS = ['%Y-%m-%d %H:%M:%S', '%m/%d/%Y %H:%M', '%m/%d/%Y %H:%M:%S', '%Y-%m-%d %H:%M'] def _pts(s): for f in _FMTS: try: return datetime.strptime(s, f) except (ValueError, TypeError): pass return None def _f(x): try: return float(x) except (TypeError, ValueError): return None def _result(rows): for r in reversed(rows): v = (r.get('ftr') or '').upper().strip() if v in ('H', 'D', 'A'): return v return None def fair_mult(o): r = [1.0/x for x in o]; s = sum(r); return [ri/s for ri in r] def half_spread(mid): if mid < 0.05: return 0.003 elif mid < 0.25: return 0.005 elif mid < 0.35: return 0.025 return 0.020 def load(): fixtures = defaultdict(list) for path in glob.glob(os.path.join(DIR, "*_polybackfill.csv")): with open(path, newline='', encoding='utf-8') as fh: for r in csv.DictReader(fh): fixtures[(r.get('game_date'), r.get('league'), r.get('home'), r.get('away'))].append(r) fx = [] for key, rows in fixtures.items(): gd = key[0] if not gd: continue pre = sorted([(_pts(r['timestamp']), r) for r in rows if (r.get('status') or '').upper() == 'PRE' and _pts(r['timestamp'])], key=lambda t: t[0]) if not pre: continue ftr = _result(rows) if ftr is None: continue fx.append(dict(gd=gd, ftr=ftr, pre=[r for _, r in pre])) fx.sort(key=lambda f: f['gd']) return fx SIDX = {'home': 0, 'away': 2} def pure_dev_bet(fx, lo, hi): """First pre-KO snapshot with a qualifying x-book deviation. Bet cheaper book. Returns dict or None.""" for r in fx['pre']: ph, pd_, pa = _f(r.get('pin_home')), _f(r.get('pin_draw')), _f(r.get('pin_away')) pin_ok = not (None in (ph, pd_, pa)) and ph > 1 and pd_ > 1 and pa > 1 pf_all = fair_mult((ph, pd_, pa)) if pin_ok else None cands = [] for side, sidx in SIDX.items(): k_mid = _f(r.get(f'kalshi_{side}_mid')); k_ask = _f(r.get(f'kalshi_{side}_ask')) p_px = _f(r.get(f'poly_{side}_price')) if k_mid is None or p_px is None: continue if not (0 < k_mid < 1 and 0 < p_px < 1): continue xfair = (k_mid + p_px) / 2.0 # cheaper book on this side if k_mid <= p_px: book, mid, px = 'kalshi', k_mid, (k_ask if k_ask and 0 < k_ask < 1 else k_mid) else: book, mid, px = 'poly', p_px, p_px + half_spread(p_px) xclv = (xfair - mid) / xfair if xfair else None if xclv is None or not (lo <= xclv < hi): continue won = (fx['ftr'] == 'H' and side == 'home') or (fx['ftr'] == 'A' and side == 'away') pin_clv = ((pf_all[sidx] - mid) / pf_all[sidx]) if pf_all else None cands.append(dict(xclv=xclv, side=side, book=book, px=px, won=won, pin_clv=pin_clv)) if cands: return max(cands, key=lambda d: d['xclv']) return None def stat(bets): n = len(bets) if not n: return None wins = sum(b['won'] for b in bets) net = sum(((1-b['px'])/b['px']) if b['won'] else -1.0 for b in bets) mu = sum(b['px'] for b in bets); var = sum(b['px']*(1-b['px']) for b in bets) z = (wins-mu)/math.sqrt(var) if var > 0 else 0.0 p = 0.5*math.erfc(z/math.sqrt(2)) return dict(n=n, wr=wins/n*100, roi=net/n*100, z=z, p=p) def line(tag, s): if not s: print(f" {tag:40s} (no bets)"); return print(f" {tag:40s} n={s['n']:3d} WR={s['wr']:4.1f}% ROIturn={s['roi']:+6.1f}% " f"z={s['z']:+.2f} p={s['p']:.3f}") def main(): fx = load() print(f"resolved fixtures with PRE snapshots: {len(fx)}\n") print("=== A) PURE X-BOOK DEVIATION (fair = mean of the two books, NO Pinnacle) ===") print(" band sweep on x-book CLV (bet cheaper book @ its ask):") for lo, hi in [(0.03, 0.12), (0.05, 0.12), (0.05, 0.20), (0.05, 1.0), (0.08, 1.0), (0.10, 1.0)]: bets = [b for b in (pure_dev_bet(f, lo, hi) for f in fx) if b] line(f"x-book CLV in [{lo:.2f},{hi:.2f})", stat(bets)) print("\n=== B) FALSIFICATION: same fixtures, bet the DEARER book instead ===") def dearer_bet(fx, lo, hi): for r in fx['pre']: cands = [] for side in SIDX: k_mid = _f(r.get(f'kalshi_{side}_mid')); p_px = _f(r.get(f'poly_{side}_price')) if k_mid is None or p_px is None or not (0= p_px: mid, px = k_mid, (_f(r.get(f'kalshi_{side}_ask')) or k_mid) else: mid, px = p_px, p_px+half_spread(p_px) xclv = (mid - xfair)/xfair # positive = this book DEARER if not (lo <= xclv < hi): continue won = (fx['ftr']=='H' and side=='home') or (fx['ftr']=='A' and side=='away') cands.append(dict(xclv=xclv, px=px, won=won)) if cands: return max(cands, key=lambda d: d['xclv']) return None for lo, hi in [(0.05, 0.12), (0.05, 1.0)]: bets = [b for b in (dearer_bet(f, lo, hi) for f in fx) if b] line(f"bet DEARER book, dev in [{lo:.2f},{hi:.2f})", stat(bets)) print("\n=== C) DOES IT NEED PINNACLE? split pure-dev bets by pin_fair agreement ===") bets = [b for b in (pure_dev_bet(f, 0.05, 1.0) for f in fx) if b] have_pin = [b for b in bets if b['pin_clv'] is not None] pin_agree = [b for b in have_pin if b['pin_clv'] > 0] # Pinnacle ALSO says cheap book is cheap pin_disag = [b for b in have_pin if b['pin_clv'] <= 0] # Pinnacle says cheap book is fair/expensive line("ALL pure-dev (x-book CLV>=5%)", stat(bets)) line(" + Pinnacle AGREES (pin_clv>0)", stat(pin_agree)) line(" + Pinnacle DISAGREES (pin_clv<=0)", stat(pin_disag)) print("\n=== D) reference: PIN_FAIR signal on same fixtures (the live strategy) ===") def pin_bet(fx, lo=0.05, hi=0.12): for r in fx['pre']: ph, pd_, pa = _f(r.get('pin_home')), _f(r.get('pin_draw')), _f(r.get('pin_away')) if None in (ph,pd_,pa) or not (ph>1 and pd_>1 and pa>1): continue pf = fair_mult((ph,pd_,pa)) cands = [] for side, sidx in SIDX.items(): k_mid = _f(r.get(f'kalshi_{side}_mid')); k_ask = _f(r.get(f'kalshi_{side}_ask')) if k_mid is None or not (0