Description

Graham and Debs tested Anthropic’s new Fable 5 model by asking it to create an investment banking-style Nvidia company overview from just a single sentence and a slide template.

Learning Objectives

---

1. Distinguish why revenue split by geography suits Lululemon better than a product-segment split.
2. Identify unprompted technical choices as signals of model quality.
3. Formulate three genuine scenario cases for a business under restructuring, rather than flat 100bps sensitivities.
4. Determine where the D&A schedule breaks when the CapEx case is flexed.
5. Compare the model's deferred revenue treatment against how a research analyst would build a dedicated schedule.
6. Recognize the role of a readme tab in giving an agent persistent context for a living document.
7. Analyze Claude's workaround for the Excel API's lack of native data table support.
8. Evaluate the two differing grades to distinguish genuine errors from defensible simplifications.