nx_analyst_infer_gate.nx

nx_analyst_infer_gate.nx -- proves the analyst's statistical-inference layer with HAND-COMPUTED known answers, every threshold arithmetic worked out by hand from rp^2*(n-2) > ceil(tcp^2*(1e6-rp^2)/1e6). D001-compliant: emits its verdict through the ONE shared nx_gate_verdict lib (anchor "verdict="). The discriminating pair this gate exists for: the SAME two r values nx_analyst_report produces on its capstone dataset (size r=994, age r=-172) are tested here at n=8 instead of that gate's n=300. At n=300 both are real; at n=8 size still is and age is NOISE. Sample size -- not effect size -- is what flips it, which is exactly what the report could not previously say. Every positive tooth is paired with an adversarial negative that MUST be rejected, so passing on true positives alone cannot make this green. license_tier: ORIGINAL No hardware writes (Rule 26).