Read the timer plumbing in 02_timers.js, the global-scope swap in 98_global_scope_shared.js, the test ignore list in tests/unit/timers_test.ts, and the WPT expectation diffs. The wiring is correct -- setTimeout/setInterval/clearTimeout/clearInterval now point at nodeSet*/nodeClear* from the windowOrWorkerGlobalScope record, the unstable nodeGlobals branch is gone, and the Timeout-vs-number impedance is handled in refTimer/unrefTimer via the new if (typeof id !== 'number') id = id[kTimerId] cast.

Three concerns I'd want explicit decisions on before merging -- not blocking, but each is a user-visible spec/contract change that may surprise downstream code:

1. WHATWG HTML spec regressions are now permanent. tests/unit/timers_test.ts marks 5 cases as Deno.test.ignore with // TODO(bartlomieju): this test is not valid -- these cover string-callback eval (setTimeout("code", 100)), this-binding to globalThis, and primordial scope isolation, all of which the HTML spec mandates. The PR body doesn't note these as casualties; worth at least a line in the PR description and a release note so users don't think string-callback support disappeared as a bug.

2. refTimer(id)/unrefTimer(id) null-safety changed. Old getActiveTimer(null) returned undefined and the if (timer) guard skipped silently. New code does if (typeof id !== 'number') id = id[kTimerId] before the guard, so passing null or undefined now throws TypeError: Cannot read properties of null/undefined (reading 'Symbol(...)' ). Code in the wild that defensively calls unrefTimer(timer.id) where timer may have already been cleared would change from no-op to throw. A if (id == null) return; early-return at the top of both functions would preserve the prior behavior cheaply.

3. setTimeout return type is now Timeout not number. The body says numeric coercion still works via [Symbol.toPrimitive], which is true, but reference-equality patterns silently break: const ids = new Set(); ids.add(setTimeout(fn,1)); ids.has(1) now returns false because the Set stored a Timeout object. Number(id) === id also flips from true to false. Probably acceptable as a Node-compat trade-off, but again worth a release-note callout for users who used to do Map<number, ...> keyed by timer IDs.

Didn't review ext/node/global.rs removal in detail (487 lines deleted, replacing the v8 named-property handler proxy) -- trusting that the test suite passing covers the cases that proxy was load-bearing for. If there are subtle global-access patterns it was intercepting, those would only surface in user code, not CI.