No Closed Form
Notes

Model Risk as a Knowledge Representation Problem

What's lost when a model inventory is treated as a spreadsheet, not a graph.

Most model inventories are spreadsheets: one row per model, columns for owner, tier, last validation date. It’s a list of things. But the questions that matter in model risk are almost never about a single thing — they’re about relationships. Which models feed this one? What assumption do these fourteen models share? If this data source is wrong, what downstream outputs are implicated?

A spreadsheet can’t answer those, because it threw away the edges. Models depend on other models, on assumptions, on data, on each other’s outputs; governance attaches evidence and decisions to those dependencies. That’s a graph, and forcing it into rows loses exactly the structure the discipline exists to reason about.

I think the useful reframing is: a model inventory is a knowledge representation, and the representation you choose decides which questions are askable. Rows make you good at counting models and bad at understanding them.

Seed note — I want to develop the “navigable graph of assumptions, dependencies, and evidence” idea into something concrete. It connects directly to the Model Object Graph experiment.