A new open-source tool called asktheboard transforms AI decision-making into accountable predictions. Instead of chatbots that simply agree, it creates panels of expert personas that make pre-registered, time-stamped bets on future outcomes. Each decision records the prior belief, individual seat stances with probabilities, and resolution dates before results are known.

The system computes Brier scores to measure calibration accuracy once outcomes are realized. Decisions live as git-committable ADRs (Architecture Decision Records), using commit timestamps as immutable proof that predictions weren't backdated. This addresses a core problem with AI advice: anyone can clone a panel, but only those who call shots in advance and survive reality's grading build trustworthy track records.

Built for BYOK (Bring Your Own Key) deployment, the engine makes no LLM calls itself and costs nothing to run. Users supply their own OpenAI-compatible endpoints. Python and CLI interfaces let teams convene boards by persona types like architect, skeptic, or pragmatist. The project demonstrates with real examples including a prediction about Postgres vs vector database performance.

Currently tracking live bets like the June 2026 US jobs report, the system shows how engineering teams can institutionalize foresight. The skeptic seat predicted 40% probability while pragmatists estimated 56% for +150k jobs. No managed hosting exists yet, but a paid tier is planned for users wanting hosted scoreboards.