PR 18076 landing regression evidence

Original authorized head: 13a80daa18db183b112c64a906d8741871a09a1c
Prepared head: b12fb5c41ea6d3e18f277186b1e7d01757a75635

Real Chromium browser tests mount the actual pooled-usage component, Base UI popover, reset hook, dialog and styles with synthetic Alpha account data. No provider RPC is sent.

Before: original head leaves Alpha details open when every segment disappears. Intermediate repair 3eb9f719b1 hides the popup, but it resurfaces on restoring the segments even after moving the pointer away. The same intermediate head retains the source-account popup when nextCreditId is removed, replacing the active reset-capable trigger. Both strengthened assertions failed before the final repair.

After: the popup closes through Base UI normal lifecycle. Restoring all segments leaves it closed until a new click. Toggling a source account from credit-1 to no credit and back to credit-2 dismisses each replaced trigger; a fresh click shows current details, and Use reset opens a live confirmation dialog. The existing hover transfer, empty reset plate, pending, success and failure cases pass. Five browser tests pass; scoped lint, web typecheck, formatting and diff checks pass.

The existing published hover/refill screenshots and recordings remain representative for the unchanged visible-segment flow. These focused tests verify the disappearance and credit-update paths; this was not a full Usage route or real reset request.