An FPGA constraints file lost 328 characters of its comments to a lossy transcode, all replaced with ?, and the pin lines survived, so the build still works, which is exactly why nobody noticed, taking with it the rationale for tried and rejected pins and which netlist version each pin decision belongs to.
From the Grok pain-radar digests, Oct 27-28 2026 (Oct 27 8:41 AM ET through Oct 28 7:41 AM ET digests plus the Haitch editions). Verified live 2026-10-28: issue/PR/thread read; title and state confirmed.