After Discover showed failed batch j1, another session could finish a newer batch. The deck then reported neither an active nor a failed batch, as the server suppresses a failure a later batch supersedes, but the page kept following j1 and its error stayed shown (or, before j1's first poll, "Finding new side quests…" did, and the error arrived later). Each new read of the deck is now reconciled: one reporting neither batch stops following a job known to have failed, by its poll or by the deck's previous read. A followed job that succeeded is still left followed, so its end is read as before. The replenish-failure test's stub now reports its failed batch on the deck, as the server does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>