Knowledge Packs — Stunspot’s Guide to Game Theory

This repository separates navigation from corpus.

Do not create or expect docs/reports/. The reports live in knowledge-packs/by-report/.


Use the compiled packs unless you have a specific reason not to. They preserve the canon sequence while keeping upload count low.

Pack Type Files Best Use Use When
Source reports 12 Precise retrieval, selective upload, citation, editing, source repair. Your tool handles many files well, or you need report-level provenance.
Compiled packs 4 Default AI/RAG upload format. Balanced coverage with lower file count. You want strong structure without a single giant file.
Omnibus 1 One-file import, local archive, full-corpus search, strong long-context systems. Your tool prefers one large source or you want an archival copy.

Pack Selection by Workflow

Workflow Recommended Pack Rationale
ChatGPT Project / custom assistant knowledge Compiled packs Four files usually fit upload constraints while keeping the sequence legible.
NotebookLM-style corpus work Compiled packs or source reports Use compiled packs for fewer files; source reports when source-level citation matters.
Vector database / RAG ingestion Source reports Smaller canonical units make retrieval, metadata filtering, and citation auditing cleaner.
Long-context manual analysis Omnibus One complete file reduces file handling friction.
Corpus editing or source repair Source reports Edit canonical source units, not compiled convenience bundles.
Local search or archive Omnibus + manifest One searchable file plus explicit mappings.

Source Reports

These are the canonical individual units of the corpus.


Compiled Packs

These are grouped convenience files for upload-constrained systems.


Omnibus

The omnibus is the whole canon in one file. Use it for archive, local search, and long-context systems that perform well with a single large source.


Ingestion Notes

For RAG pipelines, attach metadata before chunking where possible:

{
  "canon": "Stunspot's Guide to Game Theory",
  "version": "1.0",
  "pack_type": "source_report | compiled_pack | omnibus",
  "report_code": "A-L where applicable",
  "repository_path": "knowledge-packs/...",
  "license": "CC BY-NC-SA 4.0"
}

The report code is useful because the canon is sequential. A model answering from Report K should know that Reports A-J define the primitives it is using.