nx_browser_user_sim.nx

nx_browser_user_sim.nx -- THE SYNTHETIC USER (operator: prove it "via ocr and user testing like trying to scroll maximize etc ... without needing a human but matching a humans experience"). Drives the REAL browser core through the interactions a human tester would try, asserting each behavior mechanically: U1 SCROLL: render viewport at scroll 0 vs 120 -> the same content row appears EXACTLY 120px higher (row-hash equality, not just "pixels differ") AND the chrome strip is byte-IDENTICAL (fixed). U2 SCROLL BOUNDS: content above the fold disappears when scrolled past (mechanically gone). U3 RESIZE (narrow->wide): re-layout 400->800 -> page_h SHRINKS (reflow real) and NO text ink paints beyond the right content margin at either width (no runaway/clipped lines). U4 MAXIMIZE: re-layout at 1600 -> still coherent (ok, page_h monotonically <= the 800 height, ink in bounds). U5 CLICK-NAV: hit-test the link box -> href resolves; a click 200px below it hits NOTHING (miss is real). U6 READABILITY-AT-EVERY-STEP: the scrolled viewport + resized renders are written as PNGs for the OCR judge (knowledge/status/usersim_*.png) -- the harness runner OCRs them; this organ asserts render integrity, the census asserts the OCR numbers. NEG U7: a scroll offset of 0 vs 0 produces IDENTICAL frames (the differ in U1 is not noise). All through the SAME br_layout/br_draw_fb_at path every OS surface ships. license_tier: ORIGINAL expect_exit: 0