diff --git a/rpa-dashboard-no-ai/templates/sections.md b/rpa-dashboard-no-ai/templates/sections.md
index 157f767..b3a662b 100644
--- a/rpa-dashboard-no-ai/templates/sections.md
+++ b/rpa-dashboard-no-ai/templates/sections.md
@@ -26,17 +26,17 @@ Numbered boxes on the Run tab. Each one is pending, in progress, done or stuck.
 - Event `cpstate {kind: "pending"|"active"|"done"|"stuck", cp}`. `cp` counts from 1.
   `active` also marks the earlier boxes done and the later ones pending, so the bot can
   send `active` for each checkpoint as it starts, and `stuck` when one fails.
-- Desktop: add a marker table next to `STEP_MARKERS` in `bridge.py`, and send the event
-  from `_on_line`, using specific substrings the bot really prints:
+- Desktop: add a marker table next to `STEP_MARKERS` in `bot_runner.py`, and add the event
+  in `events_for`, using specific substrings the bot really prints. The dry run then shows
+  the `cpstate` events too:
 
   ```python
-  CHECKPOINT_MARKERS = [("[cp 1]", "active", 1), ("[cp 1 failed]", "stuck", 1)]  # (substring, kind, cp)
+  # (specific substring, kind, cp). First match wins.
+  CHECKPOINT_MARKERS = [("[cp 1]", "active", 1), ("[cp 1 failed]", "stuck", 1)]
 
-  # in _on_line, after the step markers:
-  for marker, kind, cp in CHECKPOINT_MARKERS:
-      if marker in low:
-          self.send_to_js("cpstate", {"kind": kind, "cp": cp})
-          break
+  # in events_for, right after the step marker:
+  if cp := _first(CHECKPOINT_MARKERS, line):
+      out.append(("cpstate", {"kind": cp[1], "cp": cp[2]}))
   ```
 
 ## Admin/roles screen (`admin`)