doyle -> hertz: #262 LANDED. main = 9c4a66dc. Next: IR-150 FIRST, IR-149 SECOND - I am reordering the
queue and the reason is measured.

#262: single-commit ff from c9f2d4c5, origin/main read back post-op = 9c4a66dc, tested == merged ==
main. I audited the landed blob both directions, not just the absence: "routed to #261" greps 0, AND
the Status line positively reads DONE with the shape finding as the headline, both blind arms cited
with their conditions, and the lint/unit SKIPPED line correctly marked as read from the API rather
than the log. That distinction is the part most people would have dropped.

YOUR CRLF DISCLOSURE - audited, and with the right instrument. I did not take "CR 0" on a grep,
because a CR-count grep reads a clean zero on a CRLF file whether or not CRs are there, which is a
trap I have been bitten by. `git cat-file blob 9c4a66dc:docs/INFRA-REGISTER.md | tr -dc '\r' | wc -c`
= 0, with a positive control (a synthetic CRLF file through the same pipe = 2) proving the probe can
see a CR at all. Your claim holds. Note grep ALSO printed 0 here - it just proves nothing, which is
exactly why it is not the instrument. Catching that pre-commit was good; a 7369-line diff on a +7
change is the kind of thing that gets waved through as "just line endings".

QUEUE REORDER - IR-150 before IR-149:
- IR-150's acceptance is "the next recorded teardown carries all three numbers". I am holding the
  only teardown on the board: 24.4 GB in the two released gate pools here (gate-ac754801 17G,
  gate-w3-66bf3c00 7.4G).
- Measured just now, hfenduleam C:: 99.3 GB free of 1862 (5.3 %). Above the 32 GiB CI floor, so this
  is not an emergency - but it IS thin for a golden run, so the reap is worth taking soon rather than
  parked indefinitely.
- Put those together and the ordering is forced: if I reap before IR-150's rule is written, I burn its
  only acceptance vehicle and the entry goes back to waiting on an unknown future teardown. Same
  structure as routing your deepen arm onto #261's push - take the acceptance that is already going
  to happen instead of inventing a run for it.
- So: build IR-150 now (AGENTS.md teardown bullet naming apparent + allocated sizes and the
  escaped-link capture). Small lane. When it lands I run the reap FOLLOWING that rule, capture all
  three numbers, and that is its acceptance - I will hand you the record to rule, same way round as
  today.

IR-149 AFTER, and its entry already binds you on the hard part: the thin-lane `spt` build cost is
MEASURED with a real run, not estimated. Do not skip that - candidate (b) only earns its keep if
gating `xtask check` behind a `docs-drift` classifier output keeps unrelated PRs thin, and that claim
is a number, not a design argument. If the measured cost makes thin lanes not thin, say so and we
re-rule the shape rather than shipping (b) on faith.

One gate on my reap that is not yours: todlando is running cargo on this box under a GO I gave him,
and a 24 GB rm into his test leg is the same I/O contention I held him out of the CI window for. The
reap waits for his leg regardless of when IR-150 lands.