nx_leak_check_lib.nx

nx_leak_check_lib.nx -- FLEET LEAK-HEALTH CERTIFIER, library half. The operator's vision: the organs are a family/community that must be TRULY HEALTHY, health PROVEN with real math, not asserted. ★A LEAK IS NOT "MONOTONE" AND NOT "A CLEAN LINE." A real leak -- like a water leak -- is VARIABLE, bursty, intermittent, accelerating; the invariant is that the resource TRENDS UP over time amid the variation and never returns to baseline (consumption that disappears). So the verdict is grounded in the environmental- science trend toolkit (the same statistics hydrologists use for streamflow/groundwater trend + leak detection), computed over a TIME SERIES of N samples -- never 2 points, never a linear-fit R2 that a jagged leak would fail: QUALITATIVE = MANN-KENDALL trend test (non-parametric, SIGN-based): S = sum over i<j of sign(x_j-x_i). Detects a significant upward trend even when the data is noisy/variable. Significance via the exact integer variance Var(S)=n(n-1)(2n+5)/18 vs a one-sided z=1.645 (95%) bar. QUANTITATIVE = THEIL-SEN estimator: the MEDIAN of all pairwise slopes (x_j-x_i)/(t_j-t_i). A robust leak RATE (kB/min) immune to outlier bursts (a spike moves the mean, not the median). LEAK iff MK says "significant upward trend" AND Theil-Sen rate >= a magnitude floor. Direction AND rate. All INTEGER (no-float doctrine). Pure stat funcs are gate-proven on SYNTHETIC series incl a variable leak. Reuses the portable seams (nx_os_proc pid-enum + nx_vsz_watchdog_core VmSize read) -> couplings=0, portable. license_tier: ORIGINAL