nx_closurehash_gate.nx

nx_closurehash_gate.nx -- PROVE nx_closurehash ACTUALLY READS THE CLOSURE (2026-08-07). The organ claims something strong: that its one hash covers every source the compiler will read. That claim has an obvious trivial-wrong implementation -- hash ONLY the target file and print a number -- and such a fake would pass every naive test: it resolves, it is deterministic, it emits 64 hex chars. *A TEST THAT THE TRIVIAL WRONG IMPLEMENTATION ALSO PASSES MEASURES NOTHING.* T3 is the tooth that decides it: mutate ONE BYTE of a TRANSITIVE dependency (depth 2, never named by the target) and require the closure hash to CHANGE. A target-only hasher cannot pass that, and neither can a one-level hasher. Fixtures are built on disk under /tmp so the walk is exercised for real -- a fixture held in memory would test the parser and skip the resolver, which is where the last two defects actually were (a scanner matching its own source text, and a resolver blind to a third search path). nx_closurehash_gate [elf] elf default ./_build/nx_closurehash.sov.elf license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0