What is authoritative here
The submission document is the frozen FLF entry. The authoritative specification of the method is the pipeline itself, published step by step at pipeline (source:
pipeline/SKILL.md+pipeline/steps/in the repository). Anything else on this site is a reading aid.
Pages
- The submission document — start here. The ≤10-page written core of the FLF entry: the principles, the pipeline (with the agent fan-out graphic), the ontology, the Bayesian logic, the cases, the limitations, and how to run it.
- How to navigate an analysis — the node types and folder layout, one worked chain from the answer down to the dataset it rests on, how the
## Prior/## Likelihoodcode blocks compose into a posterior, and how to re-price any number yourself. Written after the deadline. - The pipeline — the 10-step specification the analyses were produced by, browsable step by step, plus the runner and scripts.
What this is
The EpiStack pipeline takes one contested empirical question and builds a Bayesian knowledge base that a human can audit: not a single opaque answer, but a graph of typed notes where every number is a named, overridable variable with its reasoning beside it. The design targets a set of cross-cutting criteria — auditability, navigability (a bounded top layer with curated entry points), updateability (add evidence and re-run), and honest treatment of out-of-model mass (a residual member that competes in every posterior).