nx_thermal_profile.nx

nx_thermal_profile.nx -- FOOD-SCIENCE SUITE / PROCESS INTEGRATION rung. nx_thermal_process evaluates a single hold at a single temperature. A real retort or HTST run is a CURVE: it comes up to temperature, holds, and cools back down, and the product is being killed the whole time. This integrates lethality over that whole profile -- the General Method. F = SUM over segments of dt_i * 10^((T_i - T_ref)/z) WHY IT MATTERS, AND IT IS NOT A ROUNDING CORRECTION. Treating a process as hold-only is conservative, which sounds harmless -- but the come-up on a typical retort contributes lethality COMPARABLE TO THE ENTIRE HOLD. Ten minutes ramping from 102 C to 120 C delivers roughly 2 minutes of F0 on its own, next to a 2.52 minute 12D cook. A process authority that ignores it does not get a safer product; it gets an OVER-processed one, and over-processing is exactly how a food is made safe and inedible at the same time. So the honest direction of the hold-only error is "wasteful", not "fine". SEGMENTS LIVE IN MMAP'D ARRAYS, NOT BSS. A module-level `static [N]i64` silently kills a handler-module process on startup in this toolchain, so the arrays are lazily mmap'd and reached through pointers -- the standard workaround here, not a stylistic choice. EVERY BOUND IS FAIL-CLOSED. A segment past capacity is REFUSED rather than dropped, because a silently-truncated profile reports LESS lethality than the process really delivers, and an operator would respond by cooking longer -- the failure would hide behind a plausible number. Units match nx_thermal_process exactly: milli-Celsius, milliseconds. Grounding (cited; researcher-groundable): bigelow_general_method_thermal_process_integration stumbo_thermobacteriology_d_z_f0_model c_botulinum_proteolytic_d121_0_21min_z10 genealogy_id: thermal_process_authority + nishi_food_science_suite