Reply autopsy

Read this after any reply lands, and when the gate's opening-move default is due for review.

Without this loop, outbound copy learns only from rejections. Every send-back teaches the team what fails; nothing teaches it what worked, so the winning move gets rewritten away on the next draft.

Log three fields per reply

Positive, negative, not-now, and redirect all count. A redirect is data, not a loss.

Field	What to record
moves	Which of the four moves the touch actually made, named, not inferred
opener	The opener verbatim, one line
rung	The trust rung the recipient was on when it landed, and whether the ask matched

Keep it to three lines per reply, appended to the record that already holds the thread. A separate document nobody reads is not a loop.

Promote at five

Once five replies are logged, count moves across them and make the highest-yield move the default opening move in the gate. Two conditions:

• Five replies minimum. Below that the sample is a coincidence.
• The move has to lead for three consecutive weeks of logging before it is promoted. A single good week is noise.

Demote the same way. A default that stops appearing in replies goes back to the menu.

What this catches that a reply rate does not

A flat reply rate tells you the copy is not working. It never tells you which of the four moves was carrying the wins, so teams respond by rewriting everything, including the part that worked. The autopsy isolates the carrying move.

It also catches ask mismatch, which reads as a copy problem and is not. When replies cluster at one rung and silence clusters at another, the fix is the ask ceiling, not the sentences.

What good looks like

• Every reply has three lines against it within a day, while the thread is still fresh.
• The promoted default traces to counted replies you can point at, never to a hunch about what "feels" stronger.
• Redirects are logged with the same care as positives. They are the cheapest source of the right contact you will ever get.