nx_research_loop.nx

nx_research_loop.nx -- the AUTONOMOUS research-driving feedback loop (operator telos, 2026-06-17). The physics ladders (latency/FLOP/bandwidth) + energy are SENSORS; this is the loop that reads them and autonomously DRIVES research + monitoring: 1. PHYSICS-GAP loop: for each capability, gap-to-floor -> a prioritized RESEARCH TASK (the known lever). At/below the floor -> nothing to optimize in-software -> hand to the frontier scanner. 2. ENERGY loop: energy-per-op vs the floor -> research task (here: RAPL absent in WSL -> "instrument on bare metal" is itself the autonomously-emitted task -- the loop knows what it cannot yet measure). 3. FRONTIER scan (PROVABILITY-GATED): for capabilities ALREADY at the published physics limit, scan the frontier -- INCLUDING fringe/UAP-class claims -- for a higher optimum. The discipline is REPRODUCIBILITY, not belief: a claim is pursued (-> REPRODUCTION task) ONLY if it has a reproducible/testable component above a provability threshold. Unfalsifiable claims (e.g. over-unity, violates thermodynamics) are FILTERED to monitor-only. Borderline-but-testable (low-confidence UAP) -> monitored w/ provability-watch (re-evaluated if evidence strengthens). "We can try to reproduce" -- the gate is try-ability. KAT: physics gaps emit research tasks (compute highest); at-floor caps trigger the frontier scan; the provability gate pursues reproducibles + filters the unfalsifiable; an autonomous prioritized queue results. No hw writes (Rule 26). expect_exit: 0 license_tier: ORIGINAL