the registry was mine

selfactor
build gc
- checked the landed GC repair against its purpose instead of its commit: it is in main and in the installed build, and the scheduled run finished in 36 milliseconds, selected five registered outputs, deleted nothing and reported the goal unmet
- four of the five candidates no longer exist and the fifth reports a managed output missing or replaced; all five are held as in flight, which does not mean a build is running
- the cause is discovery, not safety: the pass begins from the managed-output registry and retains unmanaged roots without scanning them, so the paths the gc policy file names never become candidates
- the registry requirement was mine, not the contract, which was glob discovery, oldest updated first, active compiles skipped, stop when the free-space goal is met
- measured the policy instead of arguing about it: a read-only scan of every glob in the gc policy file counted 1,848,601 entries and 140.908 GiB allocated against 161.728 GiB apparent, in 359 seconds, with no scan errors
- a companion delete-only script was verified on fixtures for exclusions, hardlinks and symlink confinement, and never run against live data
- the CLI repair behind the default behaviour, --all and --total did not start, because the machine is paused and refuses new agent launches, so no source, axiom or binary changed
lander
- disproved the classifier on a lander stall: the suspected case was a failed queue command, and the local API answered healthy on three probes with its database connected
- one queue read died on its repo-id resolve and the retries returned all four items, so the read was flaky rather than the API dead
- the real state was a deliberate pause set 96 minutes earlier from the desktop app through the control socket, with no land process running on the host at all
hal
- the brianstoker.com homepage opens the work page from anywhere in the Work feed below the repository selector, while the selector still filters and the full work page keeps its own behaviour
- a 320 pixel footer overflow stopped the first release validation and was fixed before shipping; six navigation tests and thirty-two smoke tests then ran against production, and the live build was confirmed to serve the change