Use std::latch sync in benchmarks too

style.md

@@ -568,6 +568,24 @@ TEST_CASE("Server accepts connections") {
   - `std::latch`, `std::barrier`, futures/promises
 - **Force concurrent execution** using `std::latch` to synchronize thread startup
 
+#### Threading Checklist for Tests/Benchmarks
+
+**MANDATORY: Before writing any `std::thread` or `threads.emplace_back()`:**
+
+1. **Count total threads** - Include main/benchmark thread in count
+2. **Always assume concurrent execution needed** - Tests/benchmarks require real concurrency
+3. **Add `std::latch start_latch{N}`** where N = total concurrent threads
+4. **Each thread calls `start_latch.arrive_and_wait()`** before doing work
+5. **Main/benchmark thread calls `start_latch.arrive_and_wait()`** before measurement
+
+**Red flags to catch immediately:**
+- ❌ Creating threads in a loop without `std::latch`
+- ❌ Background threads starting work immediately
+- ❌ Benchmark measuring before all threads synchronized
+- ❌ Any use of `sleep_for`, `wait_for`, or timeouts
+
+**Simple rule:** Multiple threads = `std::latch` synchronization. No exceptions, even for "simple" background threads.
+
 ```cpp
 // BAD: Race likely over before threads start
 std::atomic<int> counter{0};