Deploying with Test.make({ log: true }) now streams deployed Workers' console.* output into the test console with sub-second latency — Cloudflare's tail API is too slow to be useful in tests.

const { test, beforeAll, afterAll, deploy, destroy } = Test.make({
  providers: Cloudflare.providers(),
  state: Cloudflare.state(),
  log: true, // patch console.* in deployed Workers + auto-tail the stack
});

With logging enabled, the Worker bundle's virtual entry patches console.* to mirror lines into an account-level alchemy-test-logger singleton hosting one SQLite-buffered Durable Object per {stack}/{stage}. WorkerProvider.tail streams from that DO over a websocket (buffered replay on connect + live push, deduped by row id across reconnects) when the worker's new testLogger attribute is set, and falls back to the Cloudflare tail API otherwise.

• src/Tail.ts: tail orchestration extracted from the CLI (alchemy tail is now a thin wrapper). Deploy.deploy({ tail: true }) forks it after apply; the test harness ties the fiber to its shared scope, so logs flow from beforeAll until destroy/afterAll.
• The logger singleton is ensured lazily in reconcile instead of via an enableTestLogging layer in Cloudflare.providers(): memoized per account per process, content-hash versioned, and guarded by the new Util/FileSemaphore.ts (a lockfile-backed, key-partitioned cross-process mutex) so concurrent vitest processes don't race putScript.
• Worker gains a testLogger output attribute, set during reconcile and reconstructed by read from the observed ALCHEMY_TEST_LOGGER DO binding. Only set where the bundle is patchable — script:, bundle: false, and dispatch-namespace workers keep the standard tail path.