nx_kaprobe.nx

nx_kaprobe.nx -- HTTP/1.1 KEEP-ALIVE CONNECTION REUSE probe + capability proof (census gap #3: "connection reuse ABSENT: Connection: close ... SOTA keep-alive = 50-80% latency cut for multi-page hosts"). The proven TLS send/recv core (nx_https_req_complete) ALREADY leaves the session CONNECTED after a Content-Length response (cl-stop) with the app-data seq counters advanced -- so a SECOND request over the SAME session/fd just works. The only thing forcing teardown was the request builder's "Connection: close". This organ builds a "Connection: keep-alive" request and measures TWO batches of K GETs to the sovereign edge (loopback 127.0.0.1:8443), both with a WARM cert cache (fair vs the current cache-only production): BATCH A (reuse): 1 handshake + K GETs over ONE session. BATCH B (no-reuse): K (handshake + GET + close) cycles = today's close-per-request behavior. saving = B - A = ~(K-1) handshakes avoided. Emits netobs_keepalive_batch_ms / _noreuse_batch_ms / _saved_ms / _per_req_ms so /netobs tracks the reuse win as a continuous series. Self-diagnosing: if the server does not honor keep-alive (request 2 fails on a torn session) the probe reports VERDICT=RED honestly. Rule 26: read-only network probe, zero hardware writes. Envelope: K=5 GETs/batch, loopback-only, 8s tmo. license_tier: ORIGINAL expect_exit: 0