diff --git a/static/messages.js b/static/messages.js
index d86cebfe..136dfd5e 100644
--- a/static/messages.js
+++ b/static/messages.js
@@ -5481,19 +5481,29 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
       if(_terminalStateReached || _streamFinalized){
         return;
       }
-      // Attempt one reconnect if the stream is still active server-side
+      // Attempt several reconnect/replay probes before declaring the turn lost.
+      // A short-lived SSE error can arrive while the worker is still running or
+      // while the run-journal replay file is just becoming visible. The old
+      // single 1.5s probe could fall through to _handleStreamError(), clearing
+      // S.activeStreamId/INFLIGHT and rendering a connection-interrupted marker
+      // even though the backend was still producing tokens; the settled response
+      // then reappeared later from sidecar/replay. Keep the live DOM/state intact
+      // during this retry window and only surface an error after all probes fail.
       if(!_reconnectAttempted && streamId){
         _reconnectAttempted=true;
-        setComposerStatus('Reconnecting…');
-        setTimeout(async()=>{
+        const _retryDelays=[1500,3000,5000,8000];
+        setComposerStatus(`Reconnecting… (1/${_retryDelays.length})`);
+        const _probeReconnect=async(attempt=0)=>{
+          if(_terminalStateReached || _streamFinalized) return;
+          if(!_isSessionCurrentPane(activeSid)) return;
           try{
             const st=await api(`/api/chat/stream/status?stream_id=${encodeURIComponent(streamId)}`);
-            if(st.active){
+            if(st&&st.active){
               setComposerStatus('Reconnected');
               _wireSSE(new EventSource(new URL(`api/chat/stream?stream_id=${encodeURIComponent(streamId)}`,document.baseURI||location.href).href,{withCredentials:true}));
               return;
             }
-            if(st.replay_available){
+            if(st&&st.replay_available){
               setComposerStatus('Restoring stream…');
               _wireSSE(new EventSource(new URL(`api/chat/stream?stream_id=${encodeURIComponent(streamId)}${_runJournalReplayParams()}`,document.baseURI||location.href).href,{withCredentials:true}));
               return;
@@ -5504,10 +5514,17 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
           if(await _restoreSettledSession(source)) return;
           if(_deferStreamErrorIfOffline()) return;
           if(_deferStreamErrorIfPageHidden(source)) return;
+          const nextDelay=_retryDelays[attempt+1];
+          if(nextDelay){
+            setComposerStatus(`Reconnecting… (${attempt+2}/${_retryDelays.length})`);
+            setTimeout(()=>{void _probeReconnect(attempt+1);}, nextDelay);
+            return;
+          }
           _flushReasoningToAnchor();
           _scheduleAnchorRegistryCleanup(120000);
           _handleStreamError(source);
-        },1500);
+        };
+        setTimeout(()=>{void _probeReconnect(0);},_retryDelays[0]);
         return;
       }
       if(await _restoreSettledSession(source)) return;
diff --git a/tests/test_run_journal_frontend_static.py b/tests/test_run_journal_frontend_static.py
index 68b8b548..4f0e2b4b 100644
--- a/tests/test_run_journal_frontend_static.py
+++ b/tests/test_run_journal_frontend_static.py
@@ -222,8 +222,12 @@ def test_reattach_path_uses_replay_when_status_reports_journal():
 
 
 def test_error_reconnect_path_can_restore_from_journal():
-    reconnect_pos = MESSAGES_SRC.index("setComposerStatus('Reconnecting")
-    block = MESSAGES_SRC[reconnect_pos : reconnect_pos + 900]
+    # Anchor on the reconnect block's stable entry point rather than the exact
+    # composer-status string: the first status was changed to a template literal
+    # `Reconnecting… (1/${_retryDelays.length})` (staged-probe counter), so the
+    # old single-quoted "setComposerStatus('Reconnecting" anchor no longer exists.
+    reconnect_pos = MESSAGES_SRC.index("_reconnectAttempted=true;")
+    block = MESSAGES_SRC[reconnect_pos : reconnect_pos + 1100]
 
     assert "st.active" in block
     assert "st.replay_available" in block
diff --git a/tests/test_sse_error_multi_probe_reconnect.py b/tests/test_sse_error_multi_probe_reconnect.py
new file mode 100644
index 00000000..9f03ca95
--- /dev/null
+++ b/tests/test_sse_error_multi_probe_reconnect.py
@@ -0,0 +1,143 @@
+"""Source-lock for the SSE-error multi-probe reconnect window in attachLiveStream.
+
+Symptom this guards against: after the chat EventSource emits `error`, the live
+turn's whole message could blank out for a moment and then reappear (from the
+sidecar / run-journal replay) or only come back on refresh.
+
+Root cause: the `EventSource.onerror` handler made a SINGLE 1.5s reconnect probe
+and, if `/api/chat/stream/status` did not yet report `active`/`replay_available`,
+fell straight through to `_handleStreamError(source)`. `_handleStreamError`
+clears the owner INFLIGHT state, nulls `S.activeStreamId`, pushes a
+"Connection interrupted" message and re-renders — wiping the live DOM even
+though the backend was frequently still producing tokens (or the replay file was
+a beat away from becoming visible). The settled response then reappeared later,
+producing the disappear-then-restore flicker.
+
+Fix: replace the single 1.5s probe with a short staged retry window
+(`_retryDelays=[1500,3000,5000,8000]` ms). Each stage re-queries the stream
+status and reconnects/replays if the backend is reachable; the live DOM and
+`S.activeStreamId`/INFLIGHT state are kept INTACT across the whole window, and
+`_handleStreamError` is only reached after every stage has failed. This file
+asserts that structure so the single-probe regression cannot silently return.
+
+These are static source assertions (no Node harness needed): they read
+static/messages.js, strip whitespace, and assert the staged-retry shape exists
+and that the terminal `_handleStreamError` call sits AFTER the
+"schedule the next probe" guard inside the reconnect block.
+"""
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parent.parent
+MESSAGES_JS = (ROOT / "static" / "messages.js").read_text(encoding="utf-8")
+
+
+def _compact(text: str) -> str:
+    return "".join(text.split())
+
+
+def _reconnect_block(compact: str) -> str:
+    """Return the body of the `if(!_reconnectAttempted&&streamId){...}` block by
+    brace-matching from its opening brace to the matching close, rather than a
+    fixed character window (which could silently truncate — and so under-assert —
+    if future guard lines grow the block). greptile P2 on #5122.
+    """
+    marker = "if(!_reconnectAttempted&&streamId){"
+    start = compact.find(marker)
+    assert start != -1, "expected the reconnect block guarded by _reconnectAttempted"
+    i = start + len(marker) - 1  # position of the opening brace
+    depth = 0
+    for j in range(i, len(compact)):
+        c = compact[j]
+        if c == "{":
+            depth += 1
+        elif c == "}":
+            depth -= 1
+            if depth == 0:
+                return compact[start: j + 1]
+    raise AssertionError("unbalanced braces: reconnect block never closed")
+
+
+def test_staged_retry_delays_present():
+    # Four escalating backoff stages replace the old single 1500ms probe.
+    assert "_retryDelays=[1500,3000,5000,8000]" in _compact(MESSAGES_JS)
+
+
+def test_probe_reconnect_helper_defined():
+    # The staged retry is driven by a recursive _probeReconnect(attempt) closure.
+    compact = _compact(MESSAGES_JS)
+    assert "const_probeReconnect=async(attempt=0)=>{" in compact
+
+
+def test_first_probe_scheduled_from_retry_delays_head():
+    # The initial probe must be scheduled off the head of _retryDelays, not a
+    # hard-coded 1500 — i.e. the old `setTimeout(async()=>{...},1500)` single
+    # probe is gone.
+    compact = _compact(MESSAGES_JS)
+    assert "setTimeout(()=>{void_probeReconnect(0);},_retryDelays[0]);" in compact
+
+
+def test_stage_counter_starts_at_one():
+    # The very first status shown must read (1/N), not a bare "Reconnecting…"
+    # that then jumps to (2/4) — otherwise the counter appears to start at 2,
+    # which looks like a glitch. _retryDelays must be declared before the first
+    # setComposerStatus so (1/${_retryDelays.length}) is in scope. (greptile P2)
+    compact = _compact(MESSAGES_JS)
+    assert "setComposerStatus(`Reconnecting…(1/${_retryDelays.length})`);" in compact
+    # And the declaration precedes that first status call.
+    decl = compact.find("const_retryDelays=[1500,3000,5000,8000];")
+    first_status = compact.find("setComposerStatus(`Reconnecting…(1/${_retryDelays.length})`)")
+    assert decl != -1 and first_status != -1
+    assert decl < first_status
+
+
+def test_each_stage_requeries_stream_status():
+    # Every probe attempt re-queries the server stream status so a backend that
+    # is still alive (or whose replay just became available) is detected within
+    # the window rather than after a single shot.
+    compact = _compact(MESSAGES_JS)
+    assert "api(`/api/chat/stream/status?stream_id=${encodeURIComponent(streamId)}`)" in compact
+
+
+def test_next_stage_is_scheduled_before_giving_up():
+    # When a stage fails but more stages remain, the handler must schedule the
+    # NEXT probe and return — never fall through to the terminal error path yet.
+    compact = _compact(MESSAGES_JS)
+    assert "constnextDelay=_retryDelays[attempt+1];" in compact
+    assert "if(nextDelay){" in compact
+    assert "setTimeout(()=>{void_probeReconnect(attempt+1);},nextDelay);" in compact
+
+
+def test_handle_stream_error_only_after_retry_window_exhausted():
+    # Inside the reconnect block, the terminal _handleStreamError(source) call
+    # must come AFTER the "schedule next stage" guard, so live state is held
+    # across the whole window and the error is surfaced only once every stage
+    # has failed. This is the exact ordering that prevents the premature
+    # state-wipe / blank-then-restore flicker.
+    block = _reconnect_block(_compact(MESSAGES_JS))
+
+    guard_idx = block.find("if(nextDelay){")
+    assert guard_idx != -1, "expected the next-stage scheduling guard in the reconnect block"
+
+    hse_idx = block.find("_handleStreamError(source);", guard_idx)
+    assert hse_idx != -1, "expected a terminal _handleStreamError(source) in the reconnect block"
+    # The terminal error call sits after the next-stage guard (retry window first).
+    assert guard_idx < hse_idx
+
+
+def test_live_state_not_cleared_mid_window():
+    # Guard against a re-introduced early state-wipe: the reconnect block must not
+    # null the active stream id or clear inflight state before the retry window
+    # is exhausted. We assert neither _clearOwnerInflightState() nor an
+    # S.activeStreamId=null assignment appears inside the reconnect block body
+    # before its terminal _handleStreamError (those belong only in
+    # _handleStreamError, reached after the window).
+    block = _reconnect_block(_compact(MESSAGES_JS))
+    end = block.find("_handleStreamError(source);")
+    assert end != -1
+    body = block[:end]
+    assert "_clearOwnerInflightState()" not in body, (
+        "reconnect window must not clear inflight state before all probes fail"
+    )
+    assert "S.activeStreamId=null" not in body, (
+        "reconnect window must not null activeStreamId before all probes fail"
+    )