"""Re-run the Poly-US maker P&L settlement and EMAIL the result. WHY THIS EXISTS RATHER THAN A BARE SCHEDULED TASK. The 2026-08-24 read said Poly-US market making is NOT worth it -- but on 13 settled books, which is nowhere near enough to act on. A job whose output lands in a log nobody opens is not a re-check, so this mails it. TWO THINGS CHANGED THE DAY THE BASELINE WAS TAKEN, and both push the number being measured: * WINDOW_H went 6h -> 72h (user, 2026-08-24). Completion is the whole equation, and fill rate is 25.3% at 0-0.5h against 6.8% at 24h+, so a wider window plausibly makes completion WORSE while making coverage better. If completion has fallen, the widening is the first suspect. * The probe budget stopped being spent on in-play books it could never open, which should raise the number of books quoted per cycle. WHAT WOULD CHANGE THE VERDICT. Break-even was 74.2% completion against 53.8% actual. Poly-US pays far more per complete book than Kalshi (+9.09c vs +1.78c) but loses far more per partial (-26.12c vs -2.82c), because a partial here is a naked short at a price chosen for being attractive. So the question is only ever "did completion move", and the honest answer at n=13 was "unknown, leaning no". Usage: python research\\pus_maker_pnl_rerun.py [--no-email] """ import datetime as dt import subprocess import sys from pathlib import Path sys.path.insert(0, r'E:\soccer\main') sys.path.insert(0, r'E:\soccer\research') PY = sys.executable ROOT = Path(r'E:\soccer') LOG = ROOT / 'logs' / 'pus_maker_pnl_rerun.log' BASELINE = """2026-08-28 baseline -- SUPERSEDES the 08-24 read, which was WRONG. The 08-24 numbers (13 books, 53.8% completion, "not worth it") came from pus_live_state.json, which detect_fills stops writing whenever the venue 503s -- and it 503s for hours. It undercounted books badly. This now reads the VENUE's own settlement record (ACTIVITY_TYPE_POSITION_RESOLUTION) and uses afterPosition.realized, the venue's own P&L, filtered to books the maker actually quoted. venue settled books 62 -> 28 the maker quoted (rest are manual/futures) COMPLETE books n=19 mean +10.94c/book PARTIAL books n= 9 mean -20.78c/book ALL books n=28 mean +0.75c/book (realised +$0.21) completion 67.9% vs BREAK-EVEN 65.5% -> marginally ABOVE, not below WHAT IS NOT THE PROBLEM (measured 2026-08-24, still stands): Every leg was offered AT the touch; fill rate there 21.6%, better than Kalshi's 11.9%. Spread sits at the 1c tick floor. Quoting level costs nothing. THE GATE HYPOTHESIS IS DEAD, on both venues. It came from the same lossy data: thin books re-price more, re-priced legs missed the join, so they LOOKED like they failed to complete. On corrected Kalshi data completion by opening gross is flat (82.7% / 81.0% / 80.0%). Do not raise the gate on this evidence. WHAT TO CHECK ON THE RERUN: completion against 65.5%, and whether the 72h window (set 08-24) moved it. Margin is ~2pp on n=28 -- that is noise, so the question is still open, not settled.""" def run(cmd): p = subprocess.run([PY] + cmd, capture_output=True, text=True, encoding='utf-8', errors='replace', cwd=str(ROOT), timeout=1800) return (p.stdout or '') + (p.stderr or '') def main(): stamp = dt.datetime.now().strftime('%Y-%m-%d %H:%M') out = [f'=== Poly-US maker P&L re-run {stamp} ===', ''] out.append(run([str(ROOT / 'research' / 'pus_maker_pnl.py')])) out.append('') out.append('=' * 66) out.append(BASELINE) text = '\n'.join(out) try: LOG.parent.mkdir(parents=True, exist_ok=True) with LOG.open('a', encoding='utf-8') as fh: fh.write(text + '\n') except OSError: pass print(text) if '--no-email' in sys.argv: return 0 try: from daily_signals import send_email body = ('
'
                + text.replace('&', '&').replace('<', '<')
                + '
') send_email(f'[MM] Poly-US maker P&L re-run {stamp}', body) except Exception as e: print(f'email failed: {type(e).__name__}: {e}') return 0 if __name__ == '__main__': sys.exit(main())