nx_inc_compile_gate.nx

nx_inc_compile_gate.nx -- VERDICT gate for rung LN10 of /compare/lang: the per-module incremental build cache in nx_inc_compile (watch symbol inc_cache_lookup). DONE-RULE UNDER TEST (lang.plan LN10, verbatim): "gate proves a one-file edit rebuilds only its dependents and the output equals a full build". EVERYTHING RUNS IN AN ISOLATED FIXTURE TREE under <buildroot>/_build/nx_inc_compile_gate/ -- its own runtime/, its own _offc/ copy of the live toolchain AND of the closure ruler, its own _build/ and its own knowledge/ -- so nothing here touches the estate's cache, its toolchain, its confs or its serving root. The fixture is wiped at SETUP, never at teardown: a teardown does not run when a run crashes, and a gate that is not idempotent reports on its first run and lies about every run after. FIXTURE ROOT IS DERIVED, NOT /tmp: measured on this NAS, a byte-identical copy of the compiler placed under /tmp returns execve 127 -- the mount refuses execution -- so every builder run would fail with an empty .s and read as a compiler defect. Build-artifact territory is executable by construction. THE ZERO-COMPILE WITNESS IS STATE, NOT A MESSAGE: nx_sov_build_run writes _build/<t>.s only inside its compile loop. The gate deletes that file before a rebuild; if the artifact is still current and the .s is still ABSENT, no compiler ran. A tooth asserts the .s IS present after a real build, so the absence witness cannot pass vacuously. THE FIXTURE'S OWN buildcache.conf SETS cache_enabled=0 so that B2's whole-target cache cannot serve the hit this gate is attributing to LN10 -- without that, every "zero compile" tooth would measure B2. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0