The auto-filed fallback lane routes every regression to the lane that may not own it

The re-lane instruction is not a triage step. The banner says "This is a FALLBACK, not a finding — nothing says the defect is Track T's. Re-lane it before working it." That is correct and it is also the whole difficulty: deciding the lane requires diagnosing the cause, which is the work the ticket is asking for. So the cheapest honest action for a T seat reading it is to decline it — correctly — and the ticket stays where it is. Nobody is being careless; the step that would move it costs the same as fixing it.

Why it is invisible from any single seat. A T seat sees a queue full of tickets it correctly does not own. An A or P seat sees nothing at all. Only a reader of every lane sees that these are the same tickets. That is also why this was found by accident: frankS hit regression-test-core-test-generic-nested-inline-specialize while proving an unrelated Makefile row, and it is red at HEAD and green on the pin.

Measured (9945b2514)

open regression-* in ranked folders        20
still at the track: T fallback             12
oldest                                     2026-09-02  (7 days)
all at prio                                70
re-lanes by any agent, ever                 0
top 9 of `ready --track T` that are these   7
of those, visible in ready --track A / P    0

What a fix has to decide, and why it is not a patch

A better GUESS is the wrong direction and the tool already says so: the banner records that guessing a lane from the job's src is what "sent three reds in one job to the wrong lane". So do not make the filer smarter about lanes.

Options, none of them costed here:

• A — file with no track: at all and let ready surface unlaned regressions in EVERY lane's queue. Honest about what is known; collides with bug-t-lane-attribution-has-two-instruments-that-disagree, which is about the ranker's fallbacks for exactly that state, so the two want deciding together.
• B — keep track: T but exclude auto-filed regressions from ready --track T's ranked body and print them as a separate unlaned block, so T's queue stops being dominated by work T does not own.
• C — leave it and accept that regressions are found by the tstate report rather than by any ready. That is the status quo and it is a decision worth making explicitly rather than by default.

Residual, and it belongs to nobody yet: the specialize regression above is red at HEAD, green on the pin, and test-core is not in gate.sh quick, so nothing in the per-fix loop will surface it. frankS excluded their own two files and frankH's ad3f58463 by revert-and-rebuild, and measured the causal range as 1603 commits touching compiler/ (the naive range back to the last commit touching the test file is 10112 — the smaller number is the right population). frankZ has been told and is in nested-specialize now.