> ## Documentation Index > Fetch the complete documentation index at: https://docs.nora.my/llms.txt > Use this file to discover all available pages before exploring further. # Revalidation > Re-check accumulated beliefs against a reference. Recoveries to reclaim and holdout regressions **Revalidation** is the second tab of the Signals screen. As its own description puts it, this is where accumulated beliefs get re-checked against a reference, not live traffic. There are two directions. * **Recoveries to reclaim** (bad→good) — a banned capability works again * **Holdout regressions** (good→bad) — the gate caught it before it shipped The Revalidation tab — Recoveries to reclaim and Cases routed around cards above the Reclaim (bad→good) section ## Reclaim (bad→good) A capability you once confirmed as failing and banned doesn't have to stay banned forever. When auto-recheck finds a paused capability working again, it appears here, and you press **Reclaim** to bring it back. * The **Recoveries to reclaim** card counts recoveries waiting for you. * The **Cases routed around** card counts cases routed around while the ban stood. If traffic keeps detouring around a ban, the capability may have recovered without you noticing — worth re-checking. * Reclaiming turns a capability back on, so it **requires approval**. ## Fragile (good→bad) If you turn on the holdout check in optimization, golden cases that a candidate would break appear here. These are answers that are right today but fragile against the next change. The gate blocked them, so nothing was deployed. You enable the holdout check when creating an experiment, via the **Hold-out regression** option (Off / Light / Full). See [Simulation](/reliable/simulation/overview). ## When to visit * After confirming a lot of failures. Some of them may have recovered since. * While running improvements with the holdout check on — fragile goldens collect here.