A forecast is only worth something if it was made before the outcome and cannot be edited after. Commit a claim about the future with an explicit deadline and the one public source that will settle it. The deadline must be in the future, so a past event cannot be logged as a prediction; the claim and source are frozen on chain. The prediction cannot be resolved until its deadline has passed. Then anyone can ask a round of GenLayer validators to read the committed source and rule whether the claim came true. The judgement is source relative: Canon grades a claim against the one source its forecaster named, not an independent oracle, and records the evidence fingerprint the round read. You cannot grade your own prediction, and the same claim cannot be registered twice.