GeezerVison

… update cache

…e-from-All

…n't clobber

…eted columns

…g to All

cursor Bot left a comment

Route-count baseline conflict: staging moved to 984 on its own; merged truth
with this branch's 2 view routes is 986, confirmed by running the audit.

…ey-scope the load flush

Merge housekeeping: table_views migration renumbered 0273 -> 0274 (staging
shipped 0273_copilot_tool_permission_decision); route baseline 986 -> 992,
confirmed by running the audit.

Two review fixes on top:

handlePersistLayout routed by activeView, which derives from the URL and lags
a render after the resolve effect stamps an adopted or just-created view — a
write in that gap fell through to shared metadata. It now routes by
seededViewIdRef, the owner record itself, stamped synchronously. This also
sends layout for a just-created view to that view before the list refetch
resolves it.

The load-window flush persisted a full readLayout() snapshot, carrying keys
the user never touched (pins still unseeded while the slower detail query
loads) over stored metadata. The buffer now records which KEYS were touched;
values are still read live at flush, but only touched keys are written.

cursor Bot left a comment

table_views migration renumbered 0274 -> 0275; staging shipped its own 0274
(file_folder_cutover_reconcile). Regenerated via drizzle-kit, chain verified.

…their owner

Merge resolves the route baseline (staging dropped to 991; merged truth 993,
confirmed by audit).

Review fixes: routing layout by seededViewIdRef broke three ways, all because
the ref updates in a PARENT effect while the grid writes from CHILD effects
that run first. The grid now stamps every persist call with the owner it was
displaying (viewLayoutKey at write time; async paths already capture at
dispatch), so routing is independent of effect ordering:

- A view switch's seed reconcile no longer lands on the outgoing owner — the
  write carries the incoming key.
- A deep-linked view's first-load reconcile persists immediately instead of
  being buffered and dropped by adoption — an owner-stamped write is fully
  addressed regardless of resolve state.
- The terminal views-error path no longer stamps seededViewIdRef, so a later
  successful refetch still runs first-resolve adoption (with localWork keep);
  a new ownerResolvedRef marks settle for the All-buffer instead.

A create race now also ends on any navigation away, so a reconcile on the
destination can't fall back to the still-pending created id.

cursor Bot left a comment

The build outgrew the 15-minute cap over the last two days of merges:
10m02 (c6acc62), then 14m44 after the folders/desktop/library batch
(adc557a, 16s under the limit), then two consecutive timeouts on
7b1af41 after the outlook merge. Staging's own latest runs show the same
signature (one cancelled). GitHub labels a job timeout "cancelled", which is
why these read as cancellations.