Tim opens by picking up where Part 2 ended: if language can be translated into numbers, what does that make possible for reading and evaluating writing? Shawn starts with the chicken and egg. Language came first, and math is a distillation of language built for solving problems.
Tim offers a practical frame. A CPA takes your question (“do I have enough money?”), looks at the numbers, and gives you a true answer based on the evidence. Shawn runs with it: a good accountant doesn’t care whether he likes you, and he won’t lie to you because the consequences land on you later. That has been Shawn’s standard for editorial feedback for 35 years. Look only at the evidence on the page, set aside personal attachment, and tell the writer the truth about how their words are landing with the reader.
From there, Shawn explains what a large language model is actually doing. Every letter, word, and space becomes a number, the model finds patterns in an enormous corpus, and it returns the most probable match. It runs on the principles and heuristics of programmers none of us have met. He distinguishes a living system, which hasn’t stopped and can’t be computed, from an artifact, which has a beginning, middle, and end and therefore can be fully computed. The machine can’t tell you what you’ll feel when you read something. It can only tell you the potential is there.
Shawn shares why he stopped working privately with writers six or seven years ago. At $150,000 per manuscript, the work inevitably became ghostwriting, and he didn’t want to spend his life as Rumpelstiltskin. His dream has been to leave everything he knows about story and myth in a form that is accessible, affordable, and more consistent than he is on a bad day.
Tim then introduces the distinction that has shaped Story Grid’s recent work: knowledge versus skill. You can read every book and attend every workshop on riding a bike and still fall off the first time you get on one. Almost everything produced for writers, including by Story Grid, has been knowledge. He walks through the research on deliberate practice and on kind versus wicked learning environments. Writing is among the most wicked, because feedback is slow, inconsistent, or wrong, and practicing in that environment builds confidence without building skill.
The two discuss whether writing has objective standards. Tim’s answer is music: you can love or hate Taylor Swift, but the songs objectively work, and jazz players don’t leave the box, they push its boundary outward. Shawn adds that the boundary moves as you approach it, which is exactly how myth works.
Tim recounts how Story Grid’s feedback work evolved from six-week scene workshops with week-long turnaround to the realization that a scene is too big to practice, the way his weightlifting coach broke the snatch into pieces. The unit had to shrink to the sentence, and the feedback had to come fast.
Shawn closes by describing what he built. It is a calculator, not an inference machine. He distilled what he does when he reads, starting with a sense of the whole and then filtering down to the evidence on the page, into a sequence of deterministic grades. Each submission is graded three times for agreement, then an emitter translates the grades into the language Shawn uses with students. The writer gets back their exact submission with feedback under each sentence: what worked and why, what needs revision and how to fix it on resubmit, plus an overall summary. No LLM, no telling writers what they want to hear. Just 35 years of editorial judgment returned within 24 hours, the way a CPA uses QuickBooks and then hands you what the numbers mean.
