[fun-ed]

Summary

• On Firefox, the sun's shadow map is redrawn every other frame instead of every frame. A frame of the office was ~1,890 draw calls, ~840 of them the shadow pass; per-frame draw calls now alternate 1,043 / 1,885 (about 22% fewer on average). Moving characters' shadows lag by one frame.
• On Firefox, a Retina screen is drawn at a pixel ratio of 1.5 instead of 2: about 44% fewer pixels to fill, and less memory for the 4× multisampled framebuffer (estimated ~70 MB on a full-screen 13" MacBook Air). Slightly softer image.
• In every browser, the first-person crosshair ray (the per-frame aim tick) stops at the longest interaction reach + 1 m instead of testing the whole building's meshes. First-person targeting only uses hits within reach, so behavior is unchanged; third-person clicks keep the unlimited ray, so "Walk closer to that first" still shows.

Why Firefox

Profiled in Firefox 159 on a MacBook Air M4 (16 GB): the tab's main thread was 94–99% busy in the frame callback. That cost is per-call JS/IPC overhead (Firefox remotes WebGL commands to another process), not GPU fill: OutlineEffect's two passes plus the shadow pass, plus the crosshair raycast.

Profile samples (same ~5 s window, before → after):

• Rendering (OutlineEffect.render): 4,584 → 3,556
• Shadow pass (WebGLShadowMap.render): 642 → 269
• Crosshair raycast: 1,249 → 832

Validation

• Pass: npm run typecheck
• Pass: npm run build
• Pass: npm test. 569 tests ran together; tests/workers.test.ts (32 tests) passed when run on its own. It stalled during the full run on a heavily loaded machine (load average 12–24), and it is server-side, untouched by this change.
• Checked in Firefox that the shadow map still exists and updates on alternating frames.

Not verified

• No fps before/after figures: the machine was too heavily loaded for stable fps numbers.
• The 1.5× cap was not checked on a real Retina display. The test Firefox ran at devicePixelRatio 1, so there is no before/after screenshot.