For five months we traded daily temperature contracts on Kalshi using a four-model forecast ensemble, and published every position in real time. In September 2026 we shut it down. The strategy needed a win rate in the high eighties just to break even, the markets proved too thin to scale into, and our most recent stretch fell below that line. Rather than keep selling something we no longer believed would pay for itself, we stopped.
Every trade we ever placed is still published - wins, losses and stop-outs alike, reconstructed from the exchange's own settlement records. We did not take the record down when the numbers turned against us.
Nothing here is investment advice. Past performance does not guarantee future results, and prediction-market trading carries real risk of loss.
How it works
A Telegram message lands - the city, the price, our win-probability, and the exact market link to open.
Tap our link straight to the exact contract on Kalshi (or find it on Coinbase / Robinhood) - you're on it in seconds.
Place the trade, set it down, get on with your day. We tell you the result when it settles.
No charts. No forecasting. No trading experience. If you can read a message and tap a button, you can do this.
Reconstructed directly from Kalshi's settlement and fill records - every win, loss, and stop-loss exit. The full per-trade ledger lives on a dedicated page.
The bot continuously scans 40+ US cities every 5 minutes. The figures above are the exchange-verified record since Apr 23, 2026 (v1.0 methodology start); see the full per-trade ledger at /track.
| Date | City | Threshold | NO Fill | P&L | Result |
|---|---|---|---|---|---|
| Loading recent trades… see the full ledger at /track. | |||||
The /track page shows every trade, reconstructed from Kalshi settlement records - with every stop-loss exit counted at its real cash result.
3rd Eyes isn't a hype account - and it isn't a single AI prompt. It's an engineered pipeline: a four-model weather ensemble, per-station bias correction learned from history, an ensemble-spread filter, a +EV pricing gate, confidence grading, and a pre-set stop-loss on every trade. Built and run by two engineers, brothers, with a combined 17 years moving and protecting production data. That discipline is why the public ledger - reconstructed from Kalshi's own settlement records - shows a win rate of the rate published at /track you can verify trade by trade.
Databricks Certified Data Engineer. Builds the forecast pipeline - multi-model weather ingestion, the scoring engine, and the signal logic that finds mispriced contracts. Day job: enterprise data engineering (Spark, Azure Synapse, SAP Datasphere).
Principal Backup Engineer & Solution Architect at a major financial institution - enterprise backup/restore, RTO/RPO, and cyber resilience & recovery. Owns reliability and data integrity: the systems that keep the engine running 24/7 and the public ledger trustworthy.
𝕏 @Sunilpokhrel007We trade every signal ourselves. Every signal we publish runs through our own accounts on auto-execute - we don't send you a trade we haven't taken. Our money is on the same line as yours.
We ran this as a live, publicly verified experiment from April to September 2026. It is now closed to new customers, and the trading engine has been switched off.
Why we stopped. The strategy needed a win rate in the high eighties just to break even, and the market for these contracts is too thin to size into. Our most recent stretch fell below that break-even line. Rather than keep selling something we no longer believed would pay for itself, we shut it down.
The record stays public. Every trade we ever placed, wins and losses alike, remains at /track exactly as it was reconstructed from the exchange's own settlement records. We did not take it down when the numbers turned against us, and we are not taking it down now.
Existing subscriptions have been cancelled and no further payments will be taken. Questions: support@3rdeyes.io.
3rd Eyes was a live experiment in trading weather prediction markets, run in public with an exchange-verified record. It is closed. The full ledger stays up, unedited, because a track record you only publish while it flatters you is not a track record.
Read the full record →