Selling hope at 5–20¢.

Shipped 16 Oct, 03:03 · research/longshot/FINDINGS.md §28–32

Skip recurring markets

• No bet in recurring markets: daily and weekly crypto prices, “up or down”, temperatures, tweet and post counts (Polymarket's recurring tag or the question)
• Skipped recurring markets go to the shadow log and are settled, like Incheon's skips
• Alerts and the health tile judge Jakarta's bets only; plan: losers at 0.85× what prices implied (was 0.70×)
• Everything else as Incheon: volume, quiet and fallen skips, order, caps, 2.5% size, exits
• 16 Oct: entry log — every bet keeps what the market looked like when we went in (price, spread, volume, hours left, rank) and how the fill came (size, how far buyers paid through us), to find losing bets before entry
• 16 Oct: Polymarket V2 markets are read (ids from positionIds), ahead of the ~2 Nov switch

Shipped 7 Oct, 16:56 · research/longshot/FINDINGS.md §20–26

Bet selection: skip crowded, quiet and fallen markets

• No bet in a market that has already traded $100k or more (crowded markets lose)
• No bet if the market had no trade in the last 2 hours (a buyer in a quiet market is often informed)
• No bet if the Yes is more than 15¢ below its highest trade of the last 7 days (fallen contenders come back)
• 2.5% of equity per position (was 5%)
• Every market closing within 7 days is listed, not only the 2,000 busiest
• Alerts retuned: the edge score expects losers at 0.70× the price-implied rate and judges bets since Incheon only
• Skipped markets are logged and settled, per rule, in the daily Telegram report
• 9 Oct: the Incheon book — wallet restarted at $500 from Incheon's ship time with Incheon's own bets; the 11 legacy bets were archived (followed to settlement outside the wallet) so they no longer lock capital or skew the stats

Shipped 5 Oct, 22:23 · research/longshot/FINDINGS.md §16–18; bot/polymarket/alerts.py

Security layer: alerts and holds replace the 50% stop

• Fast alerts (hours): picked off, fill burst, correlated shock, loss cluster, operations, drawdown from peak
• Slow alerts (days–weeks): edge score (losers vs research and no-edge rates), return per closed bet, markout trend
• Added 6 Oct: locked capital — share of equity in bets > 3 days past their end date (amber at 15%, never holds); such bets are held, not sold (§19)
• Hold — no new bets, open ones run on — on a red drawdown (−50% from peak), picked off or edge score, or any 2 reds; released only by you (/resume in Telegram, /hold to hold by hand)
• The old permanent stop at 50% of the start is gone; Alerts section on the dashboard, Telegram on every change

Shipped 5 Oct, 21:23 · research/longshot/FINDINGS.md §15

Better long shots first: 10–20¢ before 5–10¢

• Among markets closing within 7 days, quotes go to Yes at 10–20¢ before 5–10¢, then most traded (was: most traded only)
• Category stays out of it: the category ranking flipped between 2025 and 2026

Shipped 5 Oct, 08:20 · research/longshot/FINDINGS.md §12–14

Correlated bets share one cap: crypto direction, then topic

• Crypto price markets with a direction share one cap across all coins: at most 3 “dip/below” and 3 “reach/above” positions per week (was 3 per coin, up to 12)
• Crypto markets without a direction (price ranges) keep 3 per coin-week
• Extended 5 Oct: other markets share a cap of 3 per topic-week across events — topic = the market's most specific Polymarket tag (e.g. one election, one league), so 5 bets on one election night can't happen again

Shipped 4 Oct, 14:16 · research/longshot/FINDINGS.md §11

Close positions whose capital earns more elsewhere

• Buy the Yes back (taker fee paid) once the profit left in a position — the Yes ask per share — is below 1%/day × days left × capital
• In practice: slow positions close at once, fast ones near the end when Yes is ~0
• Positions in a converted negRisk group are held (their shares were converted)
• First run closes the 5 slow bets left from before Barcelona (~$130 back to recycle)

Shipped 3 Oct, 23:00 · research/longshot/FINDINGS.md §10

Convert multi-outcome positions

• When 2+ of our No positions sit in one mutually exclusive (negRisk) event, convert them: cash back now = shares − largest position, same final payoff
• Released cash buys new positions; it is repaid first from that event's payouts

Shipped 3 Oct, 10:28 · research/longshot/FINDINGS.md §9

Size with equity, cap crypto clusters

• 5% of equity at risk per position (was $25 fixed), min $5
• At most 3 positions per crypto coin closing the same week
• Hold to resolution: no early exits, never a stop-loss

Shipped 3 Oct, 09:16 · research/longshot/FINDINGS.md §8

Fast markets first

• Quote slots go to markets closing within 7 days first, most traded first; then the rest

Shipped 2 Oct, 15:31 · research/longshot/FINDINGS.md §1, §7

Launch: the paper maker book

• Sell “Yes” at 5–20¢ with resting orders, all categories, never buy Yes
• $25 at risk per position, at most 3 per event
• Markets ending within 45 days; no quotes in the last 3h or once a live event starts
• Book stops for good at 50% of the $500 start