chadtmiller.com/knotless

Switch repository

Loading…

What changes

harness/shots/world.mjs waits for the index by checking that no page shows today's UTC date: a record that has not been re-dated yet shows the day it was really made. The world's runner-minutes fixture frees minutes on 2026-11-02T00:00:00Z, and the settings page shows that as "Nov 2, 2026". So on 2 Nov 2026 the wait never held, and e2e failed on main and on every pull with "timed out waiting for the index". That date is now one constant, freeing, and the wait strips its rendered day from the pages before it looks for today's.

Why

e2e runs on every merge-queue batch, so nothing could land. main's e2e passed at 23:40Z on 1 Nov and fails on the same tree at 00:54Z on 2 Nov.

Keeping the fixture's date and leaving it out of the check means fixtures.json and the shots stay as they are, and nothing else in the world names a date that a calendar day can reach. The wait still uses today's date by design: what it detects is a record still carrying its real creation time.

How it was checked

node --check. The e2e and shots checks on this pull run it.

Shortcomings

On 2 Nov 2026 itself, a page showing an un-re-dated record dated that day would not hold the wait back; the wait's other conditions still do.

Changeset: none (test harness only).

🤖 Generated with Claude Code

This pull makes the e2e world's index wait ignore the rendered day that runner minutes free up. It does that by pulling the fixture date into one freeing constant and stripping its formatted day from the pages before looking for today's date. The fixture record and the wait now share one date, and the formatting matches the old today formatting exactly, so "Nov 2, 2026" is removed as it renders. The cost is the one the description names: on that single day an un-re-dated record would not hold the wait back, and the wait's other conditions still cover it. I only read the diff and did not run it.