EpiStack v1 — FLF competition

v1 is the iteration built for the Future of Life Foundation epistemics competition. Everything below — analyses and docs — belongs to this iteration and keeps its URL under /v1/ permanently; later iterations land at /v2/ and so on. (What EpiStack is.)

A browsable knowledge base for the FLF epistemic-stack pipeline: turning one contested empirical question into a Bayesian, navigable Obsidian-style graph — sources, observations, hypotheses, evidence links with runnable likelihood/prior code, per-cluster reviews, and one final report that answers the question.

This site is the reading surface. The runnable pipeline that produces these knowledge bases lives in the same repository under content/v1/pipeline/.

Start here

1. The submission document — the written core of the FLF entry: principles, pipeline, ontology, Bayesian logic, cases, limitations.
2. The three end-to-end runs, each a real pipeline output you can read top-to-bottom and then follow down into the graph. Each also opens as an interactive model graph — a navigable Bayesian view added after the submission deadline (evidence-links as edges, correlation-groups as joint-witness nodes, priors → posteriors):
   1. COVID origins · model graph →
   2. LHC black holes · model graph →
   3. Eggs and health · model graph →
3. Known failure modes — 3 BLOCKERs and 33 MAJORs, each logged by an agent inside the run that hit it, including one defect claim the run then retracted.
4. The pipeline itself — the 10-step specification the runs were produced by, browsable step by step.
5. Documentation — what the pipeline does, and how to read the node types.

What an analysis looks like

Every box in the debate is one markdown note with typed frontmatter, linked to the others:

1. sources (S-N) and data-bases (D-N) — the evidence base and the shared datasets it rests on.
2. observations (O-N) — findings that follow ~with certainty from a source’s data.
3. hypotheses (H-N), grouped into mutually-exclusive hypothesis-clusters (HC-N).
4. evidence-links (E-N) — observation → cluster edges, each carrying a runnable ## Likelihood code block.
5. arguments (A-N), cluster-reviews (CR-N), and the main-report (MR-1) that answers the question.

How to read the published runs

analysis-tests/ holds three end-to-end runs, one per case study, at 5–10 curated sources each (COVID origins and eggs at N=10, LHC black holes at N=5). These are deliberately small shakedown runs: all ten steps ran and every posterior recomputes from the notes, but each rests on a thin evidence base, so read them as a demonstration of the method rather than as settled answers. Fuller runs land in analyses/. Defects the runs exposed are listed in the failure-modes appendix rather than quietly fixed.