The overview shows how the projects connect. This page shows how each one moves on its own: what comes in, the stages it goes through, where it stops for a check, what comes out, and what gets carried back.
The details are left out on purpose. What's drawn is the flow, so you can see why things happen the way they do and where to push.
Master Prompt
Drawn from the Master Prompt, v2.7 (October 3, 2026).
The rulebook. Its job is to build the right prompt for each kind of work, prove it on real output, and hand a copy to each project.
The engine here is loop A. Every bad output gets traced back to the prompt that allowed it, so the rules get sharper every session. What it lacks is a pipe out: improvements pile up in the Master and reach the projects only when a copy gets pasted.
The loops
A. Every correction becomes a rule for the whole class of problem, not a fix to one line. Every session.
B. A learning seen in a second project gets promoted into the rules. Happens when someone notices.
C. Weekly: back everything up, then condense the prompts without losing a rule.
The leaks
1. Copies drift. Each project runs on its own pasted copy, so a change to the Master reaches a project only when someone carries it over by hand.
2. Promotion depends on memory. Nothing counts how often a learning shows up; someone has to recall the first sighting.
3. The Master says save as you go. The Marketing prompt says save at the end of the session. Neither says which wins.
Marketing
Drawn from the Marketing seed prompt, v1.53 (October 10, 2026).
Turns research and your direction into copy that sells: pages, offers, presentations, closes. And it pushes your ideas past where you'd take them alone.
This one has the most gates, three in a row at the end. The Screen Gate went in after October 9, when 38 corrections in one night slipped past rules that already existed. Notice that every loop runs on your judgment. Your taste is the scoreboard.
The loops
A. Each settled edit goes into the build's decisions log, which gets read before every rebuild.
B. Your picks go into the Voice Reference, so the next guess at your pick gets closer.
C. The end-of-session retro turns what went wrong into new rules. New rules go on trial first.
The leaks
1. No results loop. It learns from what you pick, never from how the copy performed with real buyers.
2. The road back to Research isn't built. Research expects handoff notes from Marketing; this prompt never says to send them.
3. Rule changes wait for the end of the session. If a session stops early, the batch never runs.
Research
Drawn from the Research Prompt, v14 (October 3, 2026).
Reads a business the way a customer would, then diagnoses it like a doctor: what the owner is proudest of, what actually stops the buyer, and the line between them.
The order is the insight: the buyer's questions get written before the site is opened, so the site gets graded against the buyer instead of describing itself. The weak spot is loop C. It's drawn broken because nothing sends Marketing's notes back, so Research can't learn which findings turned into copy that worked.
The loops
A. New buyer questions go back into the Question Bank after every run.
B. Blind spots go into the Craft File, so the next run looks for them.
C. Marketing's handoff notes are supposed to come back here. They don't yet.
The leaks
1. The pride-and-fear pattern rests on three businesses you chose. It hasn't been tested where a category already sells on fear (the prompt's own words).
2. The Question Bank and Craft File change every run but still live in Drive, which can't edit in place.
3. The current format, and Pick A Lens, have never been run end to end.
ThinkingBetter
Drawn from the ThinkingBetter seed prompt, v0.10 (October 3, 2026).
A private daily thinking partner. It moves you from a stuck way of seeing something to a more useful one, and over years it remembers how your ideas connect.
This is the only prompt whose value grows with time instead of with each session. Loop A is the whole point, and it's the weakest today, because the archive is thin and finding things depends on search that can come back empty.
The loops
A. The connection engine: every new thing gets checked against everything old, to bring back what you'd forgotten.
B. Each shift gets saved with why it happened, so there's a record of how your thinking changes.
C. Decisions get revisited later to see how they turned out.
The leaks
1. No stated home or index for the saved notes, and the connection engine depends on finding them.
2. Looking back runs on Drive search, which lags and sometimes comes back empty with no warning.
3. Nothing reviews patterns, open questions, or decisions on a schedule. They come back only when something triggers them.
All four together
Put the four drawings side by side and the same shapes repeat.
Almost every loop runs through one moment. In all four, what was learned travels back only at the end of a session, and only through you. A session that ends early teaches the system nothing.
Your judgment is the only signal coming back. Picks, corrections and reviews feed every loop. Nothing yet brings back how the work did out in the world: a page's conversions, a close's yes rate, a finding that turned into a sale.
Each prompt is a copy. The Master changes, then a person carries it to each project and pastes it in. Two hand-carried steps sit between a better rule and the work that needs it.
The pipe between projects runs one way. Research sends Marketing a brief. Nothing comes back, so Research can't learn which of its findings became copy that sold.
Two prompts disagree on when to save. The Master says as you go; Marketing says at the end of the session.