Changes touching this path

• the git-side ingest asks the seal for the tier it guards on and for the one it reuses under, so an entry overclaiming Internal can no longer buy a silent demotion nor ride into the next tree as this position own recording: #2203 asked for every premise to be re-verified on the tree first and the body named one reader where there are two decisions a few lines apart over one field, the demotion refusal taking demotes(&old_entry.1, &vis) and the Act::Reuse decision comparing old_entry.1 with the derived tier and then carrying that entry oid AND its tier forward, beside a third use that decides nothing, parent_tree going over whole to ingest_change as the base every untouched path is carried on. the two decisions get two answers because they are two questions and proximity is not an argument, that being what produced the bound #2187 had to refute: the guard is about the tier a path is recorded at, a path in disagreement is recorded two ways, so both are weighed and the narrower stands, the seal supplying the half that was missing while the entry keeps the withholding half it has at public_delta_tree, because asking the seal alone would have deleted a refusal rather than repaired one — an entry narrower than its seal is exactly what withholds that path from the projection, so ingesting over it widens the recording that was doing the withholding. the reuse gets the object alone, its question being whether a fresh put_sealed would produce what is already on disk, which the entry is not a recording of, and a disagreement there needs no refusal because withholding a reuse only means doing the work; it refuses where the projection drops, because dropping a path here drops a break-glass commit edit out of the very change that stands for that commit. the comparison is == and not the deposit lanes discriminant (#2187), the holder list here being exactly who could open the object a reuse declines to re-seal. the cost is measured at this site rather than inherited from #2196 already-open projection object: the arm pays 2 object_gets for a path the parent tree records and 0 for one it does not, read as a difference over a real ingest_commit, and the entry spelling cost the same pair. red under mutation, counts read each time: the seal term dropped from the guard (62 passed and 1 failed), the entry term dropped (62 passed and 1 failed), the reuse comparison reading the entry (61 passed and 2 failed), Act::Reuse carrying the entry tier (62 passed and 1 failed), the tier taken by a second store read (62 passed and 1 failed, object_gets reading 4 where 3 belongs), the overclaiming fixture seal made unopenable (62 passed and 1 failed, the vacuity control firing) and the underclaiming fixture entry made honest (62 passed and 1 failed, the refusal correctly stopping), each restored to 63 passed and 0 failed. the site is a census row for the first time, the field having been reached through .1 which that census header names as outside its needle, so its silence was never evidence, and ADR 0012 takes a thirteenth amendment recording the two answers, the measurement and the carry this does not repair. no migration, no wire or format byte moves and no host behaviour moves, but which git commits ingest moves on the client, so this rides the next release and owes no deploy. the workspace suite is green (4137 passed over 132 binaries, 8 ignored) (#2203) 231f38e6 · dbf3dbe6…

Renames are not followed. loot's tree maps a path to an address, so a rename is a delete and an add. This list is the history of the name, not of the bytes.