My Robot Keeps Its Receipts

Description and use

The system records forecasts in a tamper‑proof ledger, enforces a public deadline via drand, escalates invalid submissions, and grades predictions with Brier scores, rewarding honest uncertainty.

Problem

How to record and evaluate forecaster predictions while ensuring the forecaster knows when it doesn’t know and preventing manipulation of the record or timing.

Outcome

The ledger records forecaster predictions in a tamper‑proof way, enforces a public deadline, escalates invalid submissions, and grades honest uncertainty, providing a fair evaluation of forecasters over many calls.

Workflow

• 1. Call – submit a prediction record with symbol, kind, close_date, and probability. 2. Commit – seal the record’s tip before the market opens on the close date, using drand’s latest round to enforce the deadline. 3. Resolve – after the close, a reader fetches the closing price and writes it as a row. 4. Grade – code proposes the outcome, the owner seals it, and every call is scored with a Brier score, including escalations for invalid or late submissions.

Evidence

One record. Every result is a row on a hash chain. Edit a row and the chain breaks where you edited it.

Source

A clock nobody controls. drand publishes public randomness on a fixed schedule.

Source

Every call is scored with a Brier score: the square of the gap between the probability you gave and what happened.

Source