nx_tissuebudget.nx

nx_tissuebudget.nx -- WHAT THE TISSUE TICK ACTUALLY COSTS, AGAINST THE BUDGET THAT WAS ALREADY BANKED. WHY THIS EXISTS. knowledge/gamefeel_oracle.conf has carried frame_budget_ms_90hz 11..12 and js_block_ms 0..2 since the realism-oracles sweep, both cited to vendor VR performance guidance -- and NOTHING HAS EVER MEASURED OURS AGAINST EITHER. The board says it plainly: a real-time claim with no measured frame cost is a wish. This organ is the measurement, and it REFUSES rather than reporting a number nobody acts on. THE SUBJECT IS THE RUNTIME SPRING, NOT THE XPBD CAGE. sb_tick_mob_gene is what runs every frame for every girl in the browser; nx_softtissue's tet solver runs OFFLINE to identify the plant and never ships in the frame loop. Measuring the cage here would price a cost the visitor never pays, which is the measured-a- different-subject defect. The work unit is therefore ONE CAST TICK: TB_MOBS mobs x SB_PTS spring points. THIS NUMBER IS A FLOOR AND IS LABELLED ONE. It is native x86 on the estate host. The shipped path is WebAssembly in a visitor's browser, which is slower, on hardware that is very likely weaker. A floor is still decisive in one direction: if the FLOOR already misses the budget the browser cannot make it, and that is a refusal you can act on. It can never prove the browser fits, and this file does not claim it. THE SAMPLE SIZE IS DERIVED FROM THE CLOCK, NOT CHOSEN. sys_clock_now_us resolves to 1 us, so a total elapsed of T us carries ~1/T relative error on the per-tick figure. To resolve one part in TB_RESOLVE_PERMIL the run must last at least TB_RESOLVE_PERMIL microseconds, so the loop DOUBLES its tick count until it does. Below that it ABSTAINS instead of dividing a 0 into a per-tick cost of 0 -- the same instrument-resolution trap nx_replyreserve refuses on the reply phase, where "0 ms" meant "under 1 ms". license_tier: ORIGINAL No hw writes (Rule 26).