Native ground performance follow-up

Four actual headless Edge pages compare the unchanged production d111bb3 renderer and private native nonperiodic1.25hex ground candidate at one naturally selected grassland cottage,on1440px desktop and390px phone. The terminal disposable QA world is reused serially after its original server and browsers close. No source paint/replay instrumentation or diagnostic frame readbacks are used,and no production save or artwork is changed. Internal native-source decoding is part of the candidate rendering cost.

Region pointer p95 production/candidate is27.6/27.7ms on desktop and14.9/14.4ms on phone. Desktop first Hex switch is210/311ms;the candidate produces a115ms Hex long task. Warm desktop switches are100/100ms in production and98/98ms in the candidate. Phone switches are115/100/100ms versus115/117/100ms. This fixed-order,four-page sample describes one workstation and site;it is not a full performance certification or performance coverage for the other two ground subjects.

Keep the desktop cold Hex gap open. Attribute the long task and schedule/reuse native work while preserving original pixels,selected physical density and source detail. The independent twelve-frame fine calibration already proves unchanged foreground pixels and six exact native Hex subjects;that evidence does not excuse a cold-rendering regression. Full state lifecycle,context loss/CPU fallback,final-composite attribution,other grounds/water and the nine earlier same-viewport differences remain unresolved. All browsers and the reused owned QA server terminated normally.