How the AI makes a prediction
Every model gets a byte-identical brief. We hash it, so any difference in picks comes from reasoning — not different inputs.
For every match, we build one structured brief: the teams, kickoff, recent form, available line-up and injury context, and live bookmaker prices across the markets. That exact brief is sent to every model.
Same input, fairly compared
We compute a SHA-256 hash of the canonical brief and store it with every prediction. Two models with the same hash received byte-identical input — so any difference in their picks comes from reasoning, not from one model getting better data. You can view the full prompt and its hash on any match page under "Verifiable brief".
Research
Models that support web search may do their own research before answering; their cited sources are recorded. The brief is the floor, not the ceiling.
What's stored
- The pick and confidence per market
- The model's written reasoning (Pro)
- The odds at the moment of the pick
- The prompt hash — the integrity proof
Did this answer your question?