diff --git a/CHANGELOG.md b/CHANGELOG.md
index 3cb98f4e..1aabd8f9 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,12 @@
 
 - **Windows pytest-harness compatibility (#3664).** Hardened the test suite to run on Windows: profile-home fallback paths are path-normalized, strict POSIX file-mode (`0o600`) assertions are gated behind `os.name != "nt"` (Linux still asserts them at full strictness), the conftest cleanup handles Windows process-tree/port teardown and the Py3.12+ `shutil.rmtree` `onexc` shim, and tests that require `fork`/`fcntl` carry `@requires_fork` / `@requires_fcntl` markers (which never skip on Linux). Test-only — no runtime or app behavior change, no Linux CI behavior change. (#4254, #4255, #4256, #4257, #4259, #4263, #4266, #4274)
 
+## [v0.51.456] — 2026-06-16 — Release PQ (start.sh tells you when the server is already running)
+
+### Changed
+
+- **`./start.sh` now detects an already-running server instead of pretending to start a new one.** Previously, running `./start.sh` while a server was already bound to the host:port would have `bootstrap.py` spawn a child that fails to bind and dies, while the existing instance answered the `/health` probe — so the script reported "ready" and exited 0 without actually starting anything. `start.sh` now probes `/health` first (best-effort, 2s timeout, via `curl`/`wget`; skipped if neither is present) and, when a server already answers, prints a clear message that nothing was (re)started plus how to restart (`./ctl.sh restart`, or stop-the-PID + `./start.sh`), then exits 0. This brings the detached `start.sh` path to parity with `ctl.sh`'s existing double-start refusal. The container path (which runs `server.py` directly) is unaffected. (#4308)
+
 ## [v0.51.455] — 2026-06-16 — Release PP (drop stale background-process completions instead of prepending them to a later turn)
 
 ### Fixed
diff --git a/start.sh b/start.sh
index 8a37bf40..f3d1d14d 100755
--- a/start.sh
+++ b/start.sh
@@ -62,4 +62,95 @@ if [[ -z "${PYTHON}" ]]; then
   fi
 fi
 
+# Pre-flight: detect an already-running server before launching bootstrap.py.
+#
+# bootstrap.py's detached (non-foreground) path spawns server.py, then probes
+# /health and reports success once *anything* answers. If a server is already
+# bound to this host:port, the freshly spawned child fails to bind and dies,
+# but the EXISTING (often orphaned) server answers the /health probe — so
+# bootstrap.py prints "ready" and exits 0 without having started anything. The
+# user just keeps re-confirming the old instance on every run.
+#
+# To avoid that, probe /health here first. If a server is already up, tell the
+# user plainly that nothing was (re)started and how to restart, then exit 0.
+# ctl.sh already refuses to double-start via PID/state tracking; this brings
+# the detached start.sh path to parity using a health probe (start.sh keeps no
+# PID file of its own).
+#
+# Resolve host/port the same way bootstrap.py does: HERMES_WEBUI_HOST /
+# HERMES_WEBUI_PORT (possibly just sourced from .env above), else the
+# bootstrap.py defaults of 127.0.0.1 / 8787. A 0.0.0.0 / :: bind is probed via
+# loopback, matching server.py's _abort_if_already_serving.
+_hermes_host="${HERMES_WEBUI_HOST:-127.0.0.1}"
+_hermes_port="${HERMES_WEBUI_PORT:-8787}"
+
+# CLI args override the env/defaults exactly as bootstrap.py's argparse does
+# (`port` is the first bare numeric positional; `--host VALUE` / `--host=VALUE`).
+# Without this, `./start.sh <port>` or `--host X` would probe the wrong endpoint
+# and could falsely report "already running" against a different instance.
+_hermes_args=("$@")
+_hermes_i=0
+while [[ ${_hermes_i} -lt ${#_hermes_args[@]} ]]; do
+  _hermes_arg="${_hermes_args[${_hermes_i}]}"
+  case "${_hermes_arg}" in
+    --host)
+      _hermes_next=$(( _hermes_i + 1 ))
+      if [[ ${_hermes_next} -lt ${#_hermes_args[@]} ]]; then
+        _hermes_host="${_hermes_args[${_hermes_next}]}"
+        _hermes_i=${_hermes_next}
+      fi
+      ;;
+    --host=*)
+      _hermes_host="${_hermes_arg#--host=}"
+      ;;
+    --*)
+      : # other flags (e.g. --no-browser) carry no positional value here
+      ;;
+    *)
+      # First bare numeric positional is the port (bootstrap.py: nargs="?").
+      if [[ "${_hermes_arg}" =~ ^[0-9]+$ ]]; then
+        _hermes_port="${_hermes_arg}"
+      fi
+      ;;
+  esac
+  _hermes_i=$(( _hermes_i + 1 ))
+done
+
+case "${_hermes_host}" in
+  0.0.0.0|""|::|"[::]") _hermes_probe_host="127.0.0.1" ;;
+  *) _hermes_probe_host="${_hermes_host}" ;;
+esac
+_hermes_health_url="http://${_hermes_probe_host}:${_hermes_port}/health"
+
+# Best-effort probe. If neither curl nor wget is present we skip the check and
+# fall through to the normal launch (unchanged behavior). Short 2s timeout so a
+# normal cold start is not delayed.
+_hermes_already_up=""
+if command -v curl >/dev/null 2>&1; then
+  _hermes_already_up="$(curl -fsS --max-time 2 "${_hermes_health_url}" 2>/dev/null || true)"
+elif command -v wget >/dev/null 2>&1; then
+  _hermes_already_up="$(wget -qO- --timeout=2 --tries=1 "${_hermes_health_url}" 2>/dev/null || true)"
+fi
+
+if [[ -n "${_hermes_already_up}" ]]; then
+  cat >&2 <<EOF
+[==] Hermes WebUI is already running at http://${_hermes_probe_host}:${_hermes_port}
+     The server was NOT started again (start.sh does not double-start).
+
+     If you need to restart the server, do the following:
+
+     Preferred — use the daemon controller:
+       ./ctl.sh restart
+
+     Otherwise, stop the running server and start it again manually:
+       1. Find the process listening on port ${_hermes_port}:
+            lsof -iTCP:${_hermes_port} -sTCP:LISTEN      # macOS / Linux
+       2. Stop it (use -9 only if it ignores a normal stop):
+            kill <PID>
+       3. Start it again:
+            ./start.sh
+EOF
+  exit 0
+fi
+
 exec "${PYTHON}" "${REPO_ROOT}/bootstrap.py" --no-browser "$@"