[Noisemaker111]

After #2004, a 30 s sim_advance through Beacon Bastion's first wave spawned 18 raiders whose health bars drew but whose bodies never did. Their scene groups stayed empty even 40 s later.

• shell/devtools/simStepper.ts: advanceSimulation is async and yields a macrotask every 200 simulated frames (10 game seconds). React can commit what the sim spawned along the way, so models mount as they would in play. It yields every 10 s rather than every 1 s because each yield can cost a full software-rendered frame (~3 s), and per-second yields made a 30 s advance outrun drive's 30 s RPC timeout.
• sim_advance returns the promise; the agent bridge already awaits promise verbs. Tests now await and count yields. changes/sim-advance-yield.md.

Root cause, narrowed down in the live scene: the raider GLB loads in 0.5 s (6 skinned meshes, 76 clips). With time_scale and normal frames every raider mounts 12 meshes. After a single synchronous 600-frame burst, every raider group had 0 children and no errors or fallbacks.

Verification

• bun test packages/shell/src/devtools — 36 pass, 0 fail; check-types — pass; bun run gen — ok; ship:preflight — ok
• Live, Beacon Bastion. Left: without this PR, sim_advance 30 then a 40 s wait gives bars but no raiders. Right: with it, sim_advance 30 then a 45 s wait shows every raider rendered (with the scaled bars from Size world health bars to the entity on screen #2005):

---

Generated by Claude Code

[Noisemaker111]

test failed in one file: packages/shell/src/hudInputBoundary.test.ts. Its first browser test hit the 20 s hook timeout and the other four cascaded. This PR only changes devtools/simStepper.ts and the sim_advance verb. The same file timed out on #2000's run earlier today, and locally on this branch it passes 5/5 (bun test --timeout 20000 packages/shell/src/hudInputBoundary.test.ts). There's no newer beta to merge, and I can't re-run the job (403). Re-running test should clear it; I'm keeping the PR watched.

---

Generated by Claude Code