Variations and mining. Where new constructs come from
Stage 1 proposes variations of the champion opener and mines your own winning messages for new constructs. Nothing goes live until it's approved.
· 3 min read
A test is only as good as the challengers in it. Run out of new ideas and the best opener you have stays the best opener you'll ever have.
Stage 1 keeps the challengers coming from two places. Variations of the current champion, and patterns mined from your own messages that worked.
Variations of the champion
The champion is the messaging construct most first messages are drafted from. The fastest way to beat it is to change one thing about it and test the change.
So Stage 1 proposes variations that change exactly one line. A shorter version. One that ends by inviting correction instead of a yes or no question. One that opens with the buyer's own words when they said something in public. One with no hypothesis, that asks about the signal directly. One that ends with a direct ask for 20 minutes tied to a date that matters to the buyer.
When Claude is connected, Claude writes the variation, held to the same rule. One change, named for the change.
The daily pass proposes one when the queue is running low, fewer than two constructs waiting and no variation already waiting for a decision. It proposes at most one a week. Each variation goes in as proposed, waits for the Stage 1 team to approve it, and only then joins the queue for the daily testing pipeline.
Mining your own messages
The second source is your firm's own sends. Once a week, Stage 1 reads your first messages from the last 120 days and looks for two patterns.
The first is your edits. Every time a message goes, Stage 1 records whether it went as drafted, lightly edited or rewritten. If your sellers have rewritten at least eight drafts, at least two of those rewrites won, and the rewrites win at least as often as the drafts sent as-is, the rewrites are telling you something. Claude reads the pairs, the draft next to what was sent, and names the rewrite pattern that wins.
The second is your winners. Once at least eight first messages have earned a positive reply or a meeting, Claude reads them and names what they share.
Either way, the pattern comes back as a construct. A name, the parts in order, the hook, the exit, the length and a two-sentence instruction for writing it. Its example uses made-up names. Claude is told never to copy names, companies or quotes from your messages.
Stage 1 suggests at most one of each kind a month.
Where you see them
Mined constructs show up under Insights, Constructs, marked Needs approval and labeled From your edits or From your wins. The owner can tap Approve and test, or Dismiss. Each one is logged under What Stage 1 changed.
They're private. A construct mined from your messages tests only on your firm's sends and is never shared with another firm. Once it's testing, it works like any construct you'd write in the Construct Studio.
Mining needs Claude connected. Without it, nothing is mined.
Why it matters for a services sale
Your best seller rewrites drafts in a way that works. Usually nobody can say what they changed, including them. Mining puts a name on it and turns it into a recipe the rest of the team drafts from.
And it keeps the method honest. A variation is a guess, and a mined pattern is a guess with a little evidence behind it. Both go through the same test before they reach a buyer at scale. For why openers drift toward sameness in the first place, see why your cold email sounds like everyone else's.
Nothing new goes live without a person saying yes. Stage 1 proposes. You decide.