nx_subject_infer.nx

nx_subject_infer.nx -- DERIVE a gate's subject source by MEASUREMENT, not by naming convention. The .gates 4th field (the subject a gate is judging) gates the whole non-vacuity apparatus: without it nx_bite_sweep reports NO-SUBJECT and the domain can never be proven. 38 of 40 domains lack it. Filling those by hand means GUESSING, and this session already measured what guessing costs: a guessed subject that fails to open produces "no mutant killed", which reads identically to a vacuous gate -- so a wrong guess does not merely fail, it SLANDERS a working gate. ★THE INSIGHT: a gate is not a black box about its own subject either. It IMPORTS what it tests, and it NAMES the phrases it asserts on. Both are readable. So the subject can be DERIVED and then SCORED: candidates = the gate's own `import "X.nx"` lines + the name-derived guess (nx_X_gate -> nx_X) score(candidate) = how many of the gate's named prose needles actually appear in that file The winner is the file the gate demonstrably talks about. A candidate that scores 0 is not the subject, however much its name looks right -- which is exactly the nx_healthhelp_site_gate case, where the name-derived guess WAS correct but a thin HTML assembler and the copy lived elsewhere. ★IT REFUSES RATHER THAN GUESSES. If no candidate scores above zero, it emits UNRESOLVED and says so. An inferrer that always produces an answer is just a naming convention wearing a lab coat. usage: nx_subject_infer <domain> (CWD = nxc2 root) prints a suggested `.gates` row per declared gate; exit 0 = every gate resolved, 1 = at least one UNRESOLVED license_tier: ORIGINAL