Two consecutive e2e runs of the same commit on #1419 failed different tests, and each passed what the other failed.
| run |
failed |
passed |
| 1 |
billing-categories.spec.ts:62 — 60s timeout, both attempts |
contactmomenten, xwiki |
| 2 |
contactmomenten.spec.ts:50 + xwiki-integration.spec.ts:49 — assertions |
billing-categories |
309 and 308 passed respectively.
Why this is worth a look
contactmomenten list renders seeded contactmoment tickets loops every row asserting it matches /interaction/i, with the comment "the demo seed writes twelve contactmoment tickets". It failed on nth(16) — so on that run the tab held more than twelve rows, and at least one was not an interaction ticket.
That is a data-shape difference between two runs of identical code, which points at non-deterministic seeding or a filter that does not always apply, rather than at either spec.
What it is not
It was raised on #1419 (wrapping two repair steps in runAsSystem), so the obvious suspicion is that the fix makes migrations write and changes the seed. Measured live, it does not account for this:
MigrateAvgVerzoekenToOrDsar 0 migrated, 1 skipped, 0 unmappable, 0 failed (of 1 source objects)
MigratePosBookkeepingToShillinq early return, no output
At most one object on a fresh instance — it cannot produce 16+ extra ticket rows. And a deterministic consequence of a code change would fail the same test every run, not alternate.
development e2e is 3/3 green in recent runs, so the suite is not simply broken — which is what makes this worth chasing rather than dismissing.
Suggested starting points
nth(16) and "twelve seeded" disagree — capture what the extra rows actually are (the CI trace artifact is attached to the run).- Check whether the Contactmomenten quick-filter is applied before the assertion, or whether the test can observe a pre-filter render.
billing-categories timing out at 60s on both attempts in run 1 suggests the suite is close to its time budget; a slow run may push marginal specs over.
#1419 was admin-merged on this evidence: 2250 unit tests green, phpcs 0, phpmd 0, and the failures are non-deterministic and too large to be caused by a one-object migration.
Observed 2026-08-25, runs on fix/repair-steps-need-a-system-identity.
Two consecutive e2e runs of the same commit on #1419 failed different tests, and each passed what the other failed.
billing-categories.spec.ts:62— 60s timeout, both attemptscontactmomenten.spec.ts:50+xwiki-integration.spec.ts:49— assertions309 and 308 passed respectively.
Why this is worth a look
contactmomenten list renders seeded contactmoment ticketsloops every row asserting it matches/interaction/i, with the comment "the demo seed writes twelve contactmoment tickets". It failed onnth(16)— so on that run the tab held more than twelve rows, and at least one was not an interaction ticket.That is a data-shape difference between two runs of identical code, which points at non-deterministic seeding or a filter that does not always apply, rather than at either spec.
What it is not
It was raised on #1419 (wrapping two repair steps in
runAsSystem), so the obvious suspicion is that the fix makes migrations write and changes the seed. Measured live, it does not account for this:At most one object on a fresh instance — it cannot produce 16+ extra ticket rows. And a deterministic consequence of a code change would fail the same test every run, not alternate.
developmente2e is 3/3 green in recent runs, so the suite is not simply broken — which is what makes this worth chasing rather than dismissing.Suggested starting points
nth(16)and "twelve seeded" disagree — capture what the extra rows actually are (the CI trace artifact is attached to the run).billing-categoriestiming out at 60s on both attempts in run 1 suggests the suite is close to its time budget; a slow run may push marginal specs over.#1419 was admin-merged on this evidence: 2250 unit tests green, phpcs 0, phpmd 0, and the failures are non-deterministic and too large to be caused by a one-object migration.
Observed 2026-08-25, runs on
fix/repair-steps-need-a-system-identity.