1. A fresh read every 5 seconds
While a market is open, SpectralBot reads the live setup (price and probability as the market develops) every 5 seconds and compares it with what happened across more than 12,000 past BTC 15-minute markets.
2. Confidence, calibrated
Each read produces a confidence score: how likely the model thinks its side is to win. Scores are isotonically calibrated per time slot, so a stated 90% is meant to come true about 90% of the time at that point in the market. A raw model score doesn't have that property.
3. Edge against the live price
Confidence alone isn't enough. If the market already prices a side at 97¢, being 97% sure of it is worth nothing. SpectralBot also estimates edge, the gap between its confidence and the price, and requires a minimum edge before signalling. Why a 90% win rate can still lose money explains why.
4. Fixed rules, set in advance
The confidence threshold and the minimum edge are fixed ahead of time and applied the same way to every market. A signal is designated when it fires, before the market resolves. Nothing is picked out afterwards.
5. "First signal" is what gets scored
Within a market, the earliest signal that fires is the "first signal". That's the one the published results score against the market's final settlement. Some markets get no signal at all, because no read cleared both bars.
What it doesn't do
- It doesn't trade for you or connect to your Kalshi account. You decide whether to act on a signal, and you place any trade yourself on Kalshi.
- It doesn't promise outcomes. A signal is a calibrated probability, and some signals lose.
- It isn't personalized. Every subscriber sees the same signals.
Live and backtest results, with their methodology and disclosures, are on the performance section of the homepage.