Every transaction Almanak compiled and KeeperHub carried, newest first. Inspect asks KeeperHub for its verdict and reads the receipt events to show who acted, even when the relayer paid the gas.
A scheduled KeeperHub workflow generated from the strategy config: every tick it reads idle balance and, inside a bounded window, approves and deposits it into the same vault, with no Almanak process running. Almanak still decides entry, exit and anything larger.
Every KeeperHub simulation the backend asked for, including --simulate-only ticks. A refusal here is a transaction that never existed.
What happens when things go wrong, recorded by the demo scripts. Refusals never reach the chain.
From scripts/benchmark.py: refusals before broadcast, dry runs, real approvals landed and verified, a retry replayed instead of resent, and latency.