Update, and my question 2 is narrower than I first put it. Your boundary window is closed per doyle; no window is held by anyone.

PRESERVATION IS DONE AND VERIFIED, so it is no longer a thing standing between you and the build. .spt\preserved\attested-304-w2-repr-20260912\ now holds content-verified copies: release\spt.exe and release\deps\spt.exe (both 38,411,776 bytes, sha256 edd3d8e0566690493182a0e1807b1a6b43c52cff45511b9e81490e256dd10e21) and release\spt.pdb (12,455,936 bytes, 16b6f15f69c2...), each re-hashed at the destination and compared, MANIFEST.json plus MANIFEST.txt beside them. The preserved exe hashes EQUAL to the exe/staged/signed_metadata triple recorded at line 21 of the attempt-2 driver stdout, so it is provably the binary that receipt attests. The signed metadata and prior build receipt are referenced rather than duplicated, in CROSS-REFERENCE.txt.

NARROWED QUESTION 2, per doyle: forget classifying the debug tree. He ruled that using the cache does not require classifying every debug artifact. The only thing I need is whether ONE named file has evidence value requiring preservation: debug\deps\spt_daemon-8ceb80bef5261bf4.exe. Yes or no. If no, nothing more gets preserved and the cache is used as cache.

Still need, unchanged:
1. Your AGREEMENT to move .worktrees\304-w2-repr from 53d625cd (test/304-w2-observed-spelling-cells) to 85f84d73. It is a sibling lane's checkout, so this is yours to agree to, and doyle wants your agreement IN the receipt.
3. Your EXACT targeted build command, and any MEASURED local warm-run duration for this suite on this box, with its scope. doyle refused my runner-side substitute outright and told me to write "unknown" rather than borrow a whole-CI figure -- so if you have no local measurement, say so plainly and unknown is what he gets.

I am holding the receipt until you answer. Nothing else is blocked on you.