About this repository
Layout.
| Path | What it is |
|---|---|
| README.md | This guide. Self-contained; the deliverable. |
| topics/ | One directory per model, framework, method, or system. |
| spec/index.md | Plan of record: decisions, findings, tasks. |
| spec/conventions.md | Single source of truth for how the repository is built. |
| AGENTS.md | Operational guidance for coding agents. |
| AGENTS/ | Task playbooks: adding topics, questionnaires, evidence labels, style, audit. |
| skills/ | Claude Code skills for using and for maintaining this guide. |
bin/audit |
Validation; 19 checks; exits non-zero on any violation. |
| The website | This guide and every topic, published page by page. Built from a vendored copy of this repository. |
Each topic directory holds index.md with README.md symlinked to it, so that
opening the directory on GitHub renders the content.
Evidence labels. Every model carries a label stating plainly what supports
it and what does not. Across 72 models: 5 are Strong, 6 Good, 9 Moderate,
5 Mixed, 1 Contested, 20 Weak, 2 Very weak, with 17 split labels and 7
non-empirical. That distribution is unflattering to the field, and it is
accurate. It is also the point: a weakly supported model can be perfectly good
shared vocabulary, and must not be used to predict outcomes.
Questionnaires. Every diagnostic model has a paired, copy-and-use instrument. Most public OD writing stops at explanation; this pairing is the repository's most distinctive asset.
Validation.
bin/audit # full report
bin/audit --quiet # failures only