Prep the owner call, then debrief it into a screen
Lesson 5 of 6 · Assemble the block
1 · Learn the move · Flipped interaction
Two things go wrong around an owner call, one on each side of it. Before, something a neighbour said hardens into a fact and you open by asserting it. After, your read of the call ('he'd take less, his partner wants out') gets filed beside what he actually said, and a month later nobody can tell them apart. One lesson covers both because the fix is the same discipline in two directions. Before the call, the model sorts your file into KNOWN (a record says it), ASSUMED (basis and confidence) and UNKNOWN (turned into a question), and writes an opening from KNOWN only. After the call it flips: it interviews you one question at a time and labels every answer OWNER SAID, YOU SAW or YOUR READ before anything reaches the parcel screen. It gives no verdict. The screen runs when the inputs are real.
Two parts. PART 1, before I call: the calling-consent check first, then sort my file into KNOWN (source-tagged), ASSUMED (basis, confidence) and UNKNOWN (as questions), and write a three-sentence opening from KNOWN only. PART 2, when I say CALL DONE: interview me one question at a time, label every answer OWNER SAID, YOU SAW or YOUR READ, then fill the parcel screen. No verdict.
2 · Your turn. You write the prompt
The owner of 12-3340-004, the gravel lot at the east end of the Tollis Street block, answered your letter and will take a call Tuesday. Your file has the deed, the tax card, his signed reply, a remark from the neighbour at 002 that 'he's been wanting to unload it for years', and a blank parcel screen. Write the prompt that preps the call and then debriefs it into the screen.
Write the prompt for this scenario as given, or swap in a real one from your own work. The score works the same either way.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.