nx_etg_probe_cpu.nx

nx_etg_probe_cpu.nx -- E3 first real CPU-tier ETG probe. Bridges the pre-existing nx_calibrate.nx CPU measurements (SA-2 shipped) into the ETG framework (E1 NxEtgEntry attestation + E2 tournament selector). First REAL silicon probe in the substrate. Two surfaces: 1. nx_etg_classify_outcome(claim, measured, low_thresh_per_1000, high_thresh_per_1000) -> i64 Universal classifier: given a vendor CLAIM value and an empirical MEASURED value, classifies the outcome per the 11-value NX_ETG_OUTCOME_* taxonomy. Used by every per-tier probe (CPU / GPU / NVMe / network). Permilles (per-1000) keep the math integer. 2. nx_etg_probe_cpu_syscall_roundtrip(claim_ns, low_thresh_per_1000, high_thresh_per_1000, entry, silicon_serial, selector_version, timestamp_q14) -> i64 Runs nx_calibrate_run, extracts the syscall_ns_clock_gettime measurement, classifies vs claim, emits an NxEtgEntry. Returns the outcome (or negative on error). Composes: [[NISHI_ETG_ROADMAP]] E3 (this is the first real probe) nx_etg.nx (E1 NxEtgEntry attestation; SHIPPED) nx_calibrate.nx (SA-2 CPU calibration; SHIPPED) -- the EMPIRICAL MEASUREMENT layer this probe wraps in ETG semantics [[NISHI_SELF_ASSEMBLY_ROADMAP]] SA-2 (prior art being lifted up) [[feedback-physical-truth-over-published-docs-silicon-empirical-gamification]] (cardinal: vendor claim is HYPOTHESIS, calibration measurement is TRUTH; classifier records which side told the truth) Outcome classification rules: measured == 0 || claim == 0 => INCONCLUSIVE (no comparison possible) measured > claim * (1 + high/1000) => FALSIFIED (silicon worse than claimed) measured < claim * (1 - low/1000) => VENDOR_LIED_OMISSION (better than claimed) otherwise => CONFIRMED (within tolerance)