NegRisk Arbitrage Bounds Break: Four Bots Worth Cloning

NegRisk Arbitrage Is Real — But Read the Fine Print

The most important finding today isn't a bot. It's a research paper that cuts directly to the heart of a strategy several operators are already running. Polymarket's NegRisk architecture makes arbitrage bounds mathematically observable: if you can buy a complete set of mutually exclusive YES outcomes for less than $1, that's a deterministic profit. The paper formalizes this — but its key contribution is the distinction between *payoff-space* no-arbitrage and *protocol-executable* no-arbitrage. Position transformations available to traders on-chain determine what's actually capturable. Violations may exist in theory but remain unexploitable until settlement. If you're running or building NegRisk arb, this paper is required reading before you size up.

---

New Bots: Four Worth Cloning

**polymarket-edge-model** runs a Random Forest calibration model against live CLOB prices to identify mispriced markets, then paper trades the edge. The calibration-first approach is sound — most retail bots skip this step entirely. Worth examining the feature set and recalibration frequency before deploying capital.

**polymarket-weather-edge** is the most specific bot this week. It bets NO on temperature markets when a Gaussian model built from live METAR observations and Open-Meteo forecast distributions flags the YES price as overpriced. Narrow scope, clear data pipeline, testable thesis. This is the kind of domain-specific edge that doesn't get crowded out quickly.

**arbitrage** does exactly what the NegRisk paper describes — scans multi-outcome market groups for complete-set purchases below $1 and calculates post-fee profit. Flag: read the paper above before running this live. The gap between observable and executable is real and the bot may not account for it.

**plytrademoch** bets on underdogs in high-volume, unanchored markets based on documented favorite-longshot bias. Crucially, it uses a measurement-focused framework rather than a black-box model, which means you can audit the edge rather than trust it blindly. The strategy is theoretically grounded — longshot bias in prediction markets has academic support — but verify the "unanchored" market filter, since that selection criterion is doing a lot of work.

**postbot** is monitor-only. It tracks live post counts via the xtracker API for social-media-related markets and trades on deviations from expected trajectories. Interesting signal source, but the xtracker dependency is a single point of failure and the market category is narrow.

---

Security: No Malware, But Three Star-Farmed Repos

No confirmed malware this week. However, three repos warrant explicit callouts: a copy-trading Telegram bot, a whale alert Telegram bot, and a keyword-stuffed trading bot pack — all flagged for fake GitHub stars. Star-farming is the oldest credibility laundering trick in the ecosystem. None of these repos were confirmed malicious on static analysis, but Telegram bots requesting wallet permissions or private keys should be treated as hostile until proven otherwise. Don't clone, don't run, don't connect.

---

Alpha Signals

Beyond the NegRisk paper, two other research signals are worth tracking. An LLM-based merger arbitrage system using long-context reasoning over technical documents outperforms mixed-topic benchmarks on high-stakes deal outcomes — directly applicable to M&A markets on Polymarket. The second paper addresses market fragmentation: an agentic pipeline clusters overlapping and contradictory Polymarket questions using NLP over contract text. For operators running cross-market positions, hidden contradictions between related markets are an underappreciated risk.

---

What to Watch

Spain's block of Polymarket and Kalshi over licensing is the regulatory story of the week — 1,083 points on HN signals this is getting mainstream attention. Jurisdiction risk is no longer hypothetical. The Iran missile story threatening traders is a reminder that anonymity assumptions are thinner than most operators believe. And the "Nothing Ever Happens" NO-buyer bot hitting 471 points suggests the base-rate trading thesis has broken into general awareness — expect more competition in low-probability event markets.