nx_transport_policy.nx

nx_transport_policy.nx -- TIERED TRANSPORT policy for the comms stack (lib, no main). Operator 2026-07-02: "the most s class exceed june 2026 capability with the worse ones as skilled fallbacks in case of issues and flagging of the failure on the top of the line so they can auto heal." THE LADDER (best first; each lower tier is a proven, skilled fallback -- rule 14 graceful degradation): TIER 1 QUIC/WebTransport datagrams + core RS-FEC + L4S-aware (June-2026 frontier; needs the sovereign HTTP/3+QUIC server -- the registered frontier build) TIER 2 multi-socket striped WSS + core RS-FEC (k=8,m=2 across legs; one leg's TCP stall becomes a recoverable erasure -- the measured intl-h2h win, buildable now) TIER 3 single WSS/TCP (today's LIVE path -- proven, survives everything, worst intl behavior) CONTRACT (what the client core + daemon both follow): - always run the HIGHEST tier the measured capabilities allow (tp_best) - a tier must fail CONSECUTIVELY (threshold, default 3) before falling back -- skilled, not flappy - EVERY transition (fail-down or heal-up) EMITS A FLAG LINE (tp_flag) -> the status log the keep-up/heal loop consumes; a silent fallback is a doctrine violation the gate kills - fallback is never final: tp_should_retry_upgrade opens a heal window (exponential backoff, 30s..600s cap) to probe the higher tier again -- the AUTO-HEAL path back to top-of-line license_tier: ORIGINAL