In 2025 a few teams gave AI models real money and let them trade against each other. We wanted to try something plainer. On the markets where reading the situation right matters more than being fast, can a model beat the price the market itself sets, if you hand it a good enough context to work from?
Here that context is Sinapsi.wiki, a shared memory where everything is written down and pinned to a source. We froze our calls in mid-July 2026 and left them running against the market. The two pages below show how it is going, updated every day.
This is us testing Sinapsi against the market, nothing more. It is not advice, and there is no real money in it.
23 calls on geopolitics, ours against the prediction market's own prices, tracked on a single equity curve.
Open →Five dated calls on European gas, each with its criterion, its sources, and the day it resolves.
Open →What this is. Sinapsi is a verified, shareable memory, a wiki that people and AI assistants can query, not a forecasting product. This page is a test of that memory against the market. The calls are produced by an automated system with no human hand steering them, and an automated system can be wrong. No real money, no open positions. It is analysis, not financial, legal, or investment advice, and not an invitation to bet or trade. Contact: mario.brosco@salesmart.it.