"""One-shot reminder: review xhigh, and re-check a Poly-US-only rule. Scheduled for 2026-09-13 (about five weeks after the 2026-08-09 freeze). Emails a live summary read straight out of the forward-test log, then the Windows task that fires it is a one-off, so nothing recurs. WHY A SCHEDULED TASK AND NOT A CHAT REMINDER: Claude's own cron jobs are session-only and expire after 7 days, so they cannot carry five weeks. This runs off Task Scheduler on the box, which survives reboots and closed sessions. WHAT TO CHECK, AND THE TRAPS * xhigh was frozen 2026-08-09 (Kalshi-only, mid>=0.20, CLV in [0.05,0.12)). It needs ~480 bets. Anything below that is directional only. * Its discovery numbers (+15.1%, z+1.33) came from post-2026-07-05 data that was ALREADY the xlow forward-test window -- selection, not validation. Only fixtures after the freeze count. Expect the effect to shrink. * Poly-US-only was swept on 2026-08-09 and NOT frozen: the price-floor effect reproduced (+16.1% at minP>=0.20) but train and test disagreed in sign, and persistence hurt. Deliberately deferred to avoid a second frozen rule doubling the multiple-comparison exposure. * xhigh and xlow OVERLAP -- both fire on Kalshi CLV in the same band. If both look good, that is largely the same bets, not two pieces of evidence. * Better than freezing a Poly-US rule: split xhigh's OWN realised bets by venue. That answers "is the edge venue-dependent" for free. """ import collections import csv import json import math import statistics import sys from pathlib import Path sys.path.insert(0, r'E:\soccer\main') from daily_signals import send_email LOG = Path(r'E:\soccer\research\forward_test_unique_book_log.csv') CAPTURE = Path(r'E:\soccer\research\pin_fresh_capture.csv') ACC_CACHE = Path(r'E:\soccer\research\pin_accuracy_cache.json') def latest(): try: rows = [r for r in csv.DictReader(open(LOG, encoding='utf-8'))] return rows[-1] if rows else {} except Exception: return {} def _f(x): try: return float(x) except (TypeError, ValueError): return None def pinnapi_summary(): """Live read of the pinnapi-vs-API-Football comparison. Three separate questions, deliberately reported apart because they answered differently as of 2026-08-09 and conflating them caused real confusion: ACCURACY paired Brier per fixture. The only clean test -- paired, uses every fixture, no entry reconstruction. Needed ~580 fixtures. It FLIPPED SIGN (+0.0096 -> -0.0078) when n went 26 -> 77, which is why nothing below target should be believed. DIVERGENCE how often the two feeds produce a different bet. This is NOT evidence either is better -- only that the choice matters. BIAS API-Football rounds odds to 2dp (1.50 where pinnapi says 1.5917). Measured as a monotonic gradient: it underprices favourites and overprices long shots. This is a MECHANISM and was stable across sample sizes. """ out = {} if not CAPTURE.exists(): return out rows = list(csv.DictReader(open(CAPTURE, encoding='utf-8'))) out['snapshots'] = len(rows) # signal divergence, per FIXTURE (snapshot-level agreement is ~98% and # meaningless -- it is dominated by fixtures where neither feed fires) fx = collections.defaultdict(lambda: {'s': set(), 'f': set()}) for r in rows: k = (r['lc'], r['home'], r['away'], r['game_date']) if r.get('xlow_stale'): fx[k]['s'].add(r['xlow_stale']) if r.get('xlow_fresh'): fx[k]['f'].add(r['xlow_fresh']) sig = {k: v for k, v in fx.items() if v['s'] or v['f']} n = len(sig) or 1 out['sig_fixtures'] = len(sig) out['same'] = sum(1 for v in sig.values() if v['s'] and v['s'] == v['f']) out['af_only'] = sum(1 for v in sig.values() if v['s'] and not v['f']) out['pin_only'] = sum(1 for v in sig.values() if v['f'] and not v['s']) out['differ'] = sum(1 for v in sig.values() if v['s'] and v['f'] and v['s'] != v['f']) out['agree_pct'] = 100.0 * out['same'] / n # rounding bias by price bucket, signed (API-Football minus pinnapi) sg = collections.defaultdict(list) for r in rows: for t in ('h', 'a'): fr, st = _f(r.get(f'fair_fresh_{t}')), _f(r.get(f'fair_stale_{t}')) if fr is None or st is None: continue b = 'fav' if fr >= 0.5 else ('mid' if fr >= 0.25 else 'dog') sg[b].append(st - fr) for b in ('fav', 'mid', 'dog'): if sg[b]: out[f'bias_{b}'] = 100.0 * statistics.mean(sg[b]) # paired accuracy, recomputed from the cached settled results try: cache = json.loads(ACC_CACHE.read_text(encoding='utf-8')) except Exception: cache = {} last = {} for r in rows: k = f"{r['lc']}|{r['home']}|{r['away']}|{r['game_date']}" if k not in last or r['ts'] > last[k]['ts']: last[k] = r diffs = [] for k, r in last.items(): ftr = cache.get(k) if not ftr: continue fh, fa = _f(r['fair_fresh_h']), _f(r['fair_fresh_a']) sh, sa = _f(r['fair_stale_h']), _f(r['fair_stale_a']) if None in (fh, fa, sh, sa): continue yh, ya = 1.0 * (ftr == 'H'), 1.0 * (ftr == 'A') diffs.append(((sh - yh) ** 2 + (sa - ya) ** 2) - ((fh - yh) ** 2 + (fa - ya) ** 2)) if len(diffs) >= 10: m = statistics.mean(diffs) sd = statistics.pstdev(diffs) or 1e-9 se = sd / math.sqrt(len(diffs)) out.update(acc_n=len(diffs), acc_mean=m, acc_se=se, acc_z=m / se, acc_lo=m - 1.96 * se, acc_hi=m + 1.96 * se) return out def build_pin_html(p): if not p: return '

pinnapi vs API-Football

No capture data found.

' if 'acc_n' in p: verdict = ('pinnapi more accurate' if p['acc_lo'] > 0 else ('API-Football more accurate' if p['acc_hi'] < 0 else 'no measurable difference')) target = '' if p['acc_n'] >= 580 else ( f" — still {580 - p['acc_n']} short of the 580 target, " "so treat as directional") acc = (f"
  • Accuracy (paired Brier, n={p['acc_n']}): " f"{p['acc_mean']:+.5f} (SE {p['acc_se']:.5f}, z {p['acc_z']:+.2f}), " f"95% CI [{p['acc_lo']:+.5f}, {p['acc_hi']:+.5f}] → " f"{verdict}{target}
    • ") else: acc = '
  • Accuracy: too few settled fixtures to report.
    • ' bias = ' / '.join(f"{b} {p.get(f'bias_{b}', float('nan')):+.2f}c" for b in ('fav', 'mid', 'dog') if f'bias_{b}' in p) return f"""

    pinnapi vs API-Football

    Traps. The accuracy figure flipped sign (+0.0096 → −0.0078) when n went 26→77, so anything below target is noise. And a positive result justifies recalibrating the xlow band against the fresh line and re-testing — NOT switching the live feed, which stays on API-Football so the frozen forward tests keep running. Given the rounding gradient, a band tuned to a biased anchor will not transfer to an unbiased one.

    Re-run: python research\\pin_accuracy_test.py and python research\\pin_fresh_vs_stale.py

    """ def main(): r = latest() n = r.get('xhigh_n') or '0' pin_html = build_pin_html(pinnapi_summary()) body = f"""

    Reminder: review xhigh (and revisit Poly-US)

    Frozen 2026-08-09. Set as a five-week checkpoint.

    Where xhigh stands (from the forward-test log, run {r.get('run_date','?')})

    Read it with these in mind

    The Poly-US question that was deferred

    Swept 2026-08-09, not frozen: the price floor reproduced (+16.1% at minP≥0.20, needing only ~370 bets) but train and test disagreed in sign and persistence hurt every variant. The cheaper next step is to split xhigh's own realised bets by venue rather than freeze a second rule.

    Poly-US also pays a maker rebate where Kalshi charges a taker fee. If the floor effect survives, that is an execution reason to prefer it — a separate question from whether the signal is better.

    Re-run: python research\\rule_sweep.py and python research\\forward_test_unique_book.py

    {pin_html} """ ok = send_email('[REMINDER] Review xhigh + revisit Poly-US rule', body) print('sent' if ok else 'send failed') return 0 if ok else 1 if __name__ == '__main__': sys.exit(main())