Exact init exec-return outputs restored and retained

Worktree /home/jadams/tmp/k230-mainline-init-exec-return, branch mainline-init-exec-return, base 8557a21494a46d26e9df195e7d5a17fbeea6a6b1, source revision 7a99b32d979b9c4df5372476271e24b7546966cb. Root’s later host preparation found the previously qualified bundle, kernel, dev and source paths absent and stopped before opening UART. The prior successful proof remains historical evidence; its receipts and failed attempts were not overwritten. No source, controller, Nix recipe, config, default package or hardware change was made for restoration. No board/UART/camera operation or live preflight was performed here.

Fresh evaluation matched every prior selected derivation and output path before rebuilding. /nix had 1.2 TiB free. The same exact outputs were rebuilt sequentially under /tmp/k230-nix-build.lock, max-jobs=1, cores=16: source, kernel/dev, bundle, then the already built system explicitly. Build receipt records all four commands and return 0, completed 2026-10-05 15:33:10 UTC. Source took 60.17 seconds, kernel/dev 1,522.84 seconds, bundle 96.40 seconds, explicit system realization 16.93 seconds. The driver kept the exclusive lock for the entire sequence and released it. Only logging and retention flags changed from the original build command: -L, and persistent --out-link paths in place of --no-link.

Registered permanent roots now retain:

The existing p2 bundle and 24h dev comparison inputs also have roots in that protected directory. nix-store -q --roots independently confirmed all seven registered links; retention receipt records the exact command. Current Nix settings are keep-derivations=true, keep-outputs=false. Explicit links retain the selected outputs without relying on inter-output references. Retention lasts while these roots remain; it does not claim a global GC-policy change or explain the earlier absence.

Byte comparison confirms exactly unchanged SHA256 for main.c, actual config, system init, Image, wrapped initrd, hardware DT and artifact bootargs. There is no rebuilt byte-hash delta. The source remains a21ac6…, config 52e747…, Image fb2e66…, initrd.uimg 86ede1…, DT 0538ae…. Full hashes are in the receipts. No object or broad fixture suite was repeated: the reviewed source and actual artifact bytes are unchanged.

The fresh executed qualifier command, using the new successful restoration receipt, returned 0. Positive receipt revalidates actual same-derivation source/dev/config/Image, archived ELF/common loader, exact module-tree relocation, DT hardware, SHA256SUMS, five load/CRC expectations and sole volatile k230.init_exec_return=1. Raw argument bytes remain 299 → 323 and the literal command 341 bytes. Comparison receipt shows all prior qualification identities and the protected manifest digest match. Private normal data, manifest and transport remain outside the repository under ~/tmp/k230-mainline-init-exec-return-host/restore-qualification.

The normal report is the last reviewed protected host anchor, not a fresh live preflight or recovery performed by this qualifier. Root still owns staging, protected live preflight and the bounded physical trial. No restoration action changed the board or its persistent normal profile. Physical output, output-call return, userspace transition, ordinary root/login and real touch remain UNVERIFIED; task 5b.5 stays open. The previously documented operator command remains unchanged.

Private restoration logs and hashes are preserved alongside all historical logs; result.json records their identities. Strict OpenSpec/link/whitespace checks pass. This is host restoration and preparation evidence, not deployment or hardware acceptance.