One question, answered every morning

How it was made

1. 1 · Discovery
2. 2 · Strategy
3. 3 · Concept
4. 4 · Design
5. 5 · Production
6. 6 · Verification

1. 1 Discovery

   The source read like two products with one name. Two rounds of thinking, stacked on top of each other. They agreed about the platform. They did not agree about the career path. So the first thing I made was a list, before any design: nine places where they clashed, each traced to its source.

   

   1. Two sets of documents arrived about the same product.
   2. Read on their own, each one made sense.
   3. Side by side, they disagreed in nine places.
   4. So the first thing I built was the list of disagreements.

2. 2 Strategy

   One rule settled most of them: the founder's most recent dated note wins, and no round of thinking gets quietly blended into another. Two I left open on purpose. They were his call, not mine. Then the harder rule. Anything the sources did not spell out got marked as my idea, in writing.

   

   1. Everything the platform already believed went on the left.
   2. Everything I was proposing went on the right.
   3. A wall between them, so nobody has to guess whose idea was whose.

3. 3 Concept

   The daily Home screen went through three shapes. Sort by priority alone, and a low-priority task that was not due yet jumps ahead of today's work. Show every course at once, and you get a list nobody can explain. The shape that won: group by how urgent it is, sort inside each group, and put the rule on the screen.

   

   1. First try: sort the day by priority alone. A low-priority task that was not due yet jumped the queue.
   2. Second try: show every course at once. The result was a list nobody could explain.
   3. Third try: group by urgency first, and sort inside each group.
   4. Then put the rule on the screen, so the order can be questioned.

4. 4 Design

   Three screens, one shared set of facts, no server. An agent's ranked day, with the ordering rule right there to read. A leader's queue where one approval finishes the course, opens the next stage, and re-sorts the agent's day. A builder whose published versions stay frozen for anyone partway through.

   

   1. Three screens, all reading from one shared set of facts.
   2. A leader approves one submission.
   3. That single approval finishes the course and opens the next stage.
   4. And the agent's list quietly puts itself back in order.

5. 5 Production

   The build was never allowed to grade itself. Four automated passes went over it, each looking for something different, and none could see another's findings. Every finding got one of three answers: real and fixed, not real and why, or real and knowingly accepted. Nothing was dropped in silence. One pass caught a gate missing its own prerequisite.

   

   1. The finished work went out for review.
   2. Four separate automated review passes, each looking for something different.
   3. Every finding had to be given one of three answers.
   4. Real and fixed, not real and why, or real and knowingly accepted.
   5. Then round again, until a pass with nothing left to find.

6. 6 Verification

   Verification was written as code, not opinion. A script drove the real controls: mouse, touch, keyboard, reduced motion, and hostile text. It ran on a laptop and on a phone. Every page that went out had to match its source character for character. One claim proved false. I fixed it and left it on the record.

   

   1. A script did the checking, so it could not be talked out of a result.
   2. Every delivered page had to match what I wrote, character for character.
   3. Identical, not just similar.
   4. One claim turned out to be false. It was corrected and left on the record.