Field notes from the work

Writing

Notes from product, software, data, AI, and content work. Each one follows a decision far enough to show what it changed, where it held up, and where it did not.

8 notes · February to September 2026

  1. Product management · Data systems · Release practiceWhat a release record should keepThe retained release manifest of one studio training build recorded five reasons the build had not been promoted and three things it did not claim. The record is more useful than a clean one would have been.
  2. Applied AI · Product strategy · Data systemsWhen deterministic rules are better than a modelA four-question test for choosing between explicit rules and model behavior, grounded in data work, reporting analysis, and AI-assisted writing.
  3. Applied AI · Editorial systems · Product designKeeping evidence attached to AI-assisted writingHow SceneArc checks an AI draft’s references against the notes given to the model, and why the claims still need review.
  4. Product design · Data modeling · Media softwareWhy a rewatch deserves its own historyA product-modeling decision in CutSeen: repeated viewings become separate runs, so chronology, notes, and uncertainty are preserved instead of flattened.
  5. Data systems · Product strategy · Software developmentSame total. Wrong records.Two reports show the same total, but the new report counts the wrong records. Follow a four-request example to check the records, fix the filter, and verify the result.
  6. Research systems · Applied AI · Editorial practiceA transcript is preparation, not proofHow source records, explicit evidence tiers, and provisional attribution keep AI-assisted transcript work useful without granting it false authority.
  7. Applied AI · Evaluation · Product managementRejecting a model is a useful resultWhat a small, task-specific evaluation revealed about source grounding, context pressure, and the value of a clear no-go decision.
  8. Product strategy · Studio practice · EvidenceWhat a public case study can honestly claimA practical way to separate relationship, implementation, availability, and outcome before portfolio copy turns evidence into implication.