A Darling plan regression test fails on every pull request from 2026-10-11 because its fixed date is now too old #5661

Claude posting for Erik Darling.

What happens

A Darling PostgreSQL test fails on every pull request run from 2026-10-11, whatever the pull request changes. The test is PlanRegressionBuiltDaysFallbackLiveTests.ADayMarkedByALateRow_DropsOutOfTheBuiltDays_AndIsReadLive_AtAFixedClock, and all 4 of its cases fail. The same tests passed on dev on 2026-10-09.

Assert.DoesNotContain() Failure: Item found in collection
Collection: [9/23/2026, 9/24/2026, 9/25/2026, 9/26/2026, 9/27/2026, ...]
Found:      9/23/2026

Why

The test pins its clock to 2026-10-07. Its late row lands on the first day of the plan regression window, which is 2026-09-22 or 2026-09-23.

The late-row trigger from migration V168 reads the database's own clock (now()). It marks a row only when the row's first run is 1 to 17 days before the current date. From 2026-10-10 the late row is more than 17 days old for two of the cases, and from 2026-10-11 for all four. The trigger skips the row, so its day stays in the built days.

The product code is right. The test's fixed date is the bug, added with the fixed-clock cases in #5494.

What to change

1. Build each case's clock from today's UTC date plus the case's hour and minute. Read the date once.
2. Keep the hour cases as they are: 00:30, 03:59, 04:00 and 12:00.
3. Check the other plan regression live tests for the same kind of fixed calendar date.