Discussion 2026-10-14-jdu-ifee-mes

Back to the meeting notes

Answer what you can and skip the rest. Everyone's answers show under each question and update every 15 seconds.


Shared files



Docs, PDFs, images, spreadsheets, up to 25 MB. Anyone with this link can open what you share.

Before experiment 1 can run

1. Who builds the single-document path?
    2. By when?

    e.g. Friday Oct 17

      3. Should Claude propose the exact 25–30 documents now, for Reid to approve?
        4. Tax statements: only 3 deals have them. What do we do?
          5. Who should receive the "unsure" emails?
            6. Where does the experiment run?
              7. When do we meet to look at the results?

                Three quick calls from Claude's feedback

                8. How many documents, of how many types?

                One example per type can't show whether that type works.

                  9. Start sooner with each document treated as its own tiny deal, or build the proper way in first?
                    10. During the test, where do unsure documents go?
                      11. Should Claude draft what the new database table holds, for Reid to review?

                        Experiment 1: open items, one at a time

                        12. How should the existing flow take one document at a time? (Today it runs on a whole deal folder.)

                        Wrapping is fastest to try; a single-document way in is the long-term shape; re-running the deal reuses everything but is slower.

                          13. Where do the experiment's records live?

                          Postgres is closer to the real system; files are quicker to set up and throw away.

                            14. Who picks the 10 documents?
                              15. Pull out the facts before or after matching the document to its deal?
                                16. Anything else that must be true before Reid runs it?

                                  Phase 2 for flows 0 and 1 (what experiment 1 covers)

                                  17. Which document types must flow 0 recognize first?

                                  Pick the ones that show up most at intake.

                                    18. Which of Reid's four judgment calls may the AI make on its own in the experiment?
                                      19. When it isn't sure, where should the document go?
                                        20. How sure should it be before matching a document to a deal on its own?
                                        Match on a good guess
                                        Only when certain
                                          21. If flow 0 or 1 gets it wrong, what's the worst realistic outcome?

                                          This sets how strict the checks have to be.

                                            22. What should a person see when asked to confirm a match?