Irreversibility is usually captured by a comparison between the process that happens and corresponding "reverse process". In last decades, this has been extensively studied through fluctuation relations. Here we revisit relations from standpoint, suggested decades ago Watanabe, should involve prediction retrodiction on unique process, rather than two processes. We identify necessary sufficient ...