Orphaned Stake MonitorV11

Delegator detection: for each stake address we take the latest delegation transaction. If it points at a retired pool and the stake key has not since been deregistered, the address counts as orphaned.
Balance calculation: the sum of all unspent UTxOs, plus lifetime rewards from both the reward and reward_rest tables, minus withdrawals. Including reward_rest matters: it holds refunds, treasury and reserve payouts, and leaving it out makes any wallet that withdrew one of those look overdrawn.
Delegator counts: "All" counts every registered stake address still delegated. "Above 5" counts those holding more than 5 ADA, which filters out dust and empty wallets.
Dormancy: three categories, not two. A delegator is dormant when their most recent still-unspent output and their most recent withdrawal both predate their pool's retirement certificate. Active means activity within the last 90 days. Stale sits between the two: activity after the retirement certificate, but none in the last 90 days. A dormant wallet is likely abandoned; an active wallet whose owner simply has not redelegated is a very different case; a stale wallet was touched after the pool died and has since gone quiet, and is shown as its own category rather than silently counting as neither. One overlap is possible: for a pool retired within the last 90 days, a wallet whose last activity falls between those two boundaries counts as both dormant and active, so the three categories can briefly sum to slightly more than the total until the retirement ages past the 90-day window.
Forgone rewards: an estimate, not a measurement. For each pool it is the orphaned stake multiplied by the actual network yield of each epoch since retirement (member and leader rewards over active stake, from db-sync), summed across those epochs. A flat annual rate can be applied instead as a manual override, and is labelled as such when in use. It assumes the stake sat orphaned for that whole period.
Trend history: snapshots begin with the first V11 refresher run. Nothing before that point exists or can be reconstructed, so the trend chart starts short and grows one point per epoch.
Refresh cycle: the heavy query runs on a schedule in the background and takes roughly 15 minutes. This page only ever reads the finished cache, so it always loads instantly.