nx_bootstrap_gate.nx

nx_bootstrap_gate.nx -- BOOTSTRAPPABILITY: can the sovereign toolchain still be rebuilt from this tree? WHY THIS EXISTS. On 2026-07-31 the answer was NO and nobody knew. ~170 sources -- including nxc.nx, nxld.nx and nxasm_main.nx, i.e. the compiler, the linker and the assembler themselves -- imported "../nxasm/<module>.nx", and the entire nxasm/ directory was ABSENT from the build root. Three independent builds died identically with `nx_compile_x86: expand_imports failed`. The toolchain kept WORKING, because it runs from prebuilt ELFs in _offc/, so nothing ever surfaced the gap: what was broken was REBUILDING it, and a system that cannot rebuild its own compiler cannot ship a change to it. BOOTSTRAPPABILITY IS A TRACKED PROPERTY, NOT AN ASSUMPTION. This mirrors the reproducible-builds bootstrapping discipline and the 2026 hermetic-build rule that a build is a PURE FUNCTION OF ITS DECLARED INPUTS: an input that has silently vanished must fail LOUDLY at declaration time, not years later when someone finally tries to rebuild. The fix restored the modules; THIS gate is what stops the regression, because the fix without a tooth is just a good day. WHAT IT MEASURES (mechanical, no compilation -- so it finishes well inside the ~14.5s edge window that KILLS any longer gate outright; measured that day: sync died at 14486ms and _async=1 at 14140ms with deadline_ms=60000 ignored, and a self-recorded verdict never even got written because the process was killed before reaching its own emit): T1 the five recovered assembler/linker modules EXIST and are non-empty T2 ZERO sources still reference the absent ../nxasm/ path (the exact regression shape) T3 the toolchain trio's own sources are present and each declares at least one import T4 NON-VACUITY: the same predicate applied to a name that cannot exist must REPORT MISSING, proving the check can fail. A gate never observed to fail is not an instrument. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0