Native fixed autonomous PID1 argv proof

Group 5n.1 PASS on the host, base 703e91050375aa260a9b44472e605d390530d3e5. result.json preserves compiler, exact source/function/fixture hashes, independent vectors and execution facts. The selected artifact remains p2kdar89; this proof performs no Nix/kernel build, serial open or board operation.

Run from the repository root:

python3 tests/test_mainline_autonomous_bash_pid1_argv.py
openspec validate the-board-runs-a-mainline-kernel --strict
openspec validate --all
git diff --check

The native command passes 12 tests (0.498s in the recorded run), using host cc (GCC) 16.2.1 20260810. It compiles the entire untouched official old-Hush v2022.10 parser (SHA 954771b8638977d12955c4da3c540a3929d47e4431a092c079192155b2d60011) with capture-only U-Boot callbacks. Its real execution engine dispatches exactly one setenv bootargs with three arguments, producing precisely the independent 477-byte expected vector. The script’s dollar signs, semicolons, &&, double quotes and doubled backslashes remain data. Only the normal IFS environment lookup occurs: no script-variable lookups, extra command dispatch or environment writer callback. The stub stores volatile bootargs bytes; the real U-Boot setenv handler/persistence implementation is not executed. Attribution, license and adapter details are in the fixture README.

The selected Linux source is /nix/store/0l4mgw9hr1j7pj247jsrxjd478fnlfy8-linux-mainline-k230-uart-progress-memory-printk-src. Exact native next_arg, parse_args, init_setup, rdinit_setup, repair_env_string and set_init_arg functions produce only -c and the full 154-byte script. The script’s first n= is split then repaired, exterior double quotes disappear, and its dollars/backslashes remain unchanged. Actual setup reset loops are tested in both init/rdinit orders, and post--- init/rdinit text is argv rather than setup. A small adapter supplies setup registration dispatch; other pre--- handlers are inert. This is selected parser execution, not a kernel boot or successful target exec.

Four valid nonce patterns pass the complete native pipeline. Invalid nonce length/case/characters and changed quote/backslash/variable/script/extra-command syntax are rejected by the fixture’s fixed-input check before Hush execution. Separate unqualified negative runs demonstrate actual extra-command dispatch, variable lookup and the old-Hush backslash transformation; those effects are observed by inert hooks and never execute a real saveenv/reset. Python only orchestrates compilation and compares raw bytes, without a shell lexer/shlex. The unchanged script/bootargs/transport are 154/477/503 bytes; adding the host CR makes 504, within the unchanged 512-byte bound.

The first warning-strict harness compile rejected upstream Hush’s seven existing signed/unsigned comparisons at lines 2955–3116. The final harness disables only -Wsign-compare for that unchanged snapshot; all other compiler output is fatal. No parser source bytes were changed. An attempted license copy from a conventional host pathname was absent; COPYING instead contains the GPL-2 text from the exact selected Linux source’s LICENSES/preferred/GPL-2.0.

Controller/actual-artifact qualification, installed U-Boot/architecture ABI, autonomous target records, protected physical recovery and ordinary root/touch remain separate gates. The receipt cannot authorize arbitrary scripts or candidate input; it qualifies only the independently reconstructed fixed vector. A host native pass is not a physical boot, sleep/IRQ/RX result or automatic return.