Scanning for markets that are already decided
Somewhere in the long tail, a market's outcome has already happened and the price hasn't noticed. An LLM read the rules and the news for thousands of markets, hunting for exactly that. The bounty didn't cover the ammunition.
The hypothesisOur proven edge class is information latency on public truth. Generalize it: for every market ending soon whose book hasn't converged, ask an LLM one classification question — is this outcome ALREADY determined by events that have happened, per the official resolution rules? — with dated headlines attached and the price withheld. Buy the determined side where the ask still allows it.
The test
Hourly, for unconverged markets inside a closing window: the LLM received today's date, the market's resolution criteria, and fresh dated news headlines, and answered DETERMINED or UNDETERMINED with confidence and cited evidence. High-confidence determinations with a sanely-priced ask became would-enter signals; every verdict was also scored for pure accuracy at resolution, separately from the trading result — so the model's judgment and the trade's economics could fail independently.
The result
The model's verdicts were mostly accurate; the economics were not. Genuinely determined-but-mispriced markets are vanishingly rare: a 0.32% hit rate that declined toward zero week over week, because the long tail is obscure precisely where it is undetermined and efficient precisely where it is not. Full accounting: $33.48 of LLM spend against $16.54 of hypothetical edge at standard entry size — a negative-margin factory — and all four profitable determinations were just-finished sports games, a species a free scoreboard feed catches mechanically for nothing.
Conclusion
"The market hasn't noticed yet" is almost never true of anything an LLM can notice from public headlines — the overlap between machine-discoverable and not-yet-priced is nearly empty, and where it isn't, cheaper instruments than a language model already cover it. The scan survives at token cadence as a tripwire; as a strategy it's retired. The general rule earned here: before building an information edge, price the information — ours cost twice what it found.