nx_connect_match.nx

nx_connect_match.nx -- R1 of CONNECT: the RECIPROCAL language-exchange MATCH ENGINE + its gate. The language-exchange wedge is COMPLEMENTARY, not similarity: a good partner TEACHES what you want to LEARN *and* LEARNS what you can TEACH. So the core signal is a HARMONIC MEAN of the two directional fits -- one-sided pairs collapse to 0 (you cannot reciprocally exchange). Interests modulate; reciprocity is NECESSARY. Pure INTEGER math (no floats). The output is an INTERNAL ranking weight 0..100 -- NEVER a user-facing "compatibility score" (PNAS 2020: profile traits ~5% of long-term variance; we surface reciprocal interest, conversation decides). SAFETY BY CONSTRUCTION (launch prerequisite, proven here in the gate): romance/dating lane edges are UNREACHABLE to/from minors -- a minor who is a PERFECT reciprocal+interest match is still EXCLUDED from a dating-lane query (both directions), while remaining free to language-exchange. Gate has 7 checks incl. a measured NEGATIVE CONTROL: the naive arithmetic mean would wrongly surface a one-sided pair (score 25), proving the harmonic mean is what enforces reciprocity. Mirrors the nx_*_census/gate idiom. 100% sovereign. license_tier: ORIGINAL expect_exit: 0 2026-07-23 LIBRARY SPLIT: the engine moved to nx_connect_match_lib.nx so THIS gate and the LIVE app (nx_connect_serve /partners) bind ONE implementation. Before the split the live page printed hardcoded scores while crediting "the gated match engine" -- it asserted a computation it never ran. This file is now purely the VERIFIER; its emitted output is byte-identical to the pre-split version, which is the proof that the extraction changed no behavior.