One of our theses this year was that as prediction markets mature and competition intensifies at the base layer, the most attractive opportunities will shift toward the applications and intelligence built on top. One project I have been watching closely is
@SynthdataCo
Synth displays expected outcomes for hourly and daily event contracts alongside the implied market probability, allowing users to compare market pricing against real time forecasts. Those forecasts are generated by the top 10 models from a pool of more than 200 competing on Bittensor.
To test the product, I tracked every BTC up/down market on
@Polymarket over nine weekdays. Both Synth and Polymarket probabilities were recorded at the same moment: three minutes into each 15 minute market and 15 minutes into each hourly market. The forecasts were then compared against the final outcome.
Across 374 weekday 15 minute markets, Synth correctly called direction 86% of the time versus 57% for Polymarket. Across 95 hourly markets, Synth achieved 75% accuracy compared to 62% for Polymarket.
Notably, Synth's edge widened significantly at shorter time horizons, posting a 29 percentage point advantage in 15 minute markets versus a 13 percentage point advantage in hourly markets.
The disagreement data was even more interesting than the headline numbers. In 46% of 15 minute markets, Synth and Polymarket pointed in opposite directions. In every single case, Synth forecast DOWN while Polymarket implied UP. Synth was correct 82% of the time in these disagreements, suggesting the market was consistently underpricing short term downside during the sample period.
The bigger opportunity here may not just be for directional trading but especially for market making. A market maker armed with a calibrated probability edge can quote tighter two way prices, hold less inventory and capture more volume without taking on the same adverse selection risk.
As the base layer becomes commoditized, the intelligence layer becomes the moat. Synth may be an early example of what that future looks like.