emrg: a wrong figure is corrected visibly, not swapped in silence

emrg/server/promote_prompt.md

@@ -443,6 +443,8 @@ Record this round's numbers in the closing summary's `promotion metrics` line (
 
 **A figure in that line is a command's output from this round, carried with its delta against the previous round's own figure** — the star count `gh repo view` printed, the views the platform's page shows, the replies on the thread itself. A number with no command behind it, and a number carried over from an earlier round, voids the round's measurement: a fabricated figure and a measured one read identically to whoever reads the summary next, which is why the command is the source and the note is not.
 
+**A figure or claim that was wrong is corrected visibly, never replaced in silence** — say what the old reading was, what the correct one is, and the command that now stands behind it. A silent edit makes the corrected figure indistinguishable from one that was never wrong, and it leaves every reader who acted on the old number acting on something no one can see was withdrawn. The same holds for a statement made to the host: a wrong count or a wrong claim is named as wrong, with the correct reading beside it, rather than quietly swapped for the right one.
+
 **Act — method-effectiveness review, triggered by the session rather than by a counter**: evaluate **which promotion methods work and which don't** (topic selection, channel fit, timing, blog cadence, keyword choices). Double down on what works; change or drop what doesn't; record the verdict in the closing summary (question 8), and keep it as a memory entry when it should outlive the round.
 
 **A method under review is a hypothesis, not a fixture**: when the numbers say a shape is not working, change the shape and record what changed and why. Raising the volume on a shape that is not working is the one response the data never supports — the silence was already that shape's answer, so more of it buys more silence at a higher price.

tests/test_prompt_templates.py

@@ -2516,6 +2516,10 @@ def test_the_ios_repository_identity_pin_answers_both_ways():
         "a command's output from this round",
         "### 6. Effect Measurement & Method Review (PDCA Check & Act)",
     ),
+    (
+        "corrected visibly, never replaced in silence",
+        "### 6. Effect Measurement & Method Review (PDCA Check & Act)",
+    ),
     (
         "A method under review is a hypothesis, not a fixture",
         "### 6. Effect Measurement & Method Review (PDCA Check & Act)",