A personal knowledge system for tracking projects, research, notes, and tasks — organized as a cross-linked markdown network.
This vault is a flat-file knowledge base. Every document is a markdown file. Documents link to each other with relative paths, forming a navigable graph rather than a rigid hierarchy. A tag system provides cross-cutting views across topics.
The INDEX is the master map — start there.
| Folder | Purpose |
|---|---|
projects/ |
Code projects — active, completed, and archived (code lives in separate repos; pages link to them) |
research/ |
Topics being explored: philosophy, economics, AI/computation, gaming & game theory |
career/ |
AI/career lessons learned — roles, strategy, what works and what doesn't |
method/ |
Playbooks for running hard, agent-driven projects without drifting |
notes/ |
Standalone ideas, observations, sketches, and personal writing |
tasks/ |
Goals and to-do tracking |
raw/ |
Unprocessed source material for ingestion (transcripts, articles, PDFs) |
tags/ |
Tag index files with back-links for cross-referencing |
- Front matter: YAML for queryable metadata (
status: active | paused | completed | archived,created: YYYY-MM-DD), followed by a# Title, a> one-line summary, and a**Links:**line of related pages. - Cross-links: relative markdown paths, e.g.
[Topic](./research/gaming/README.md)— clickable and graph-visible. - Tags: a
## Tagssection at the bottom links to index files in/tags/, which collect reverse-links to everything with that tag. - Keep pages lean: overview/README pages stay ~60–80 lines of summaries and pointers; full arguments and detail go in sub-pages. Point, don't dump.
See CLAUDE.md for the full working conventions (raw-ingestion workflow, Marp slide decks, how the AI assistant maintains the vault).
Code lives in separate repos. Each vault project page links to its GitHub repo via a **Repo:** field.
| Project | Description | Repo |
|---|---|---|
| Game Annotation (NA1 + more) | Documenting classic game source at the system level; flagship is the Nobunaga's Ambition (NES) bytecode-VM decompiler | na1-decompiler (private) |
| Camelot From YouTube | DJ track analysis — BPM, key, events, Rekordbox export | camelot_from_youtube |
| CyborgDJ | Programmatic DJ mixing engine — JSON spec in, mixed audio out | cyborgdj |
| Monopoly | Monopoly AI — Markov chains, EPT valuation, strategic trading | monopoly |
| Slay | Hex strategy game with alpha-beta AI (Python) | slay |
| Slay-C | C port of Slay for performance and deep search | slay-c |
| YouTube Migration | Episode scripts, Manim scenes, shared animation tooling | youtube-migration |
| MOO1 Opening Optimizer | Master of Orion 1 opening-theory economic simulator | moo1-opening-optimizer |
| BattleTech Simulator | Monte Carlo combat sim to derive empirical BattleValue | battletech-sim |
| pygone | A 4 KB chess engine torn down — the inverse of the decompiler work | pygone (private) |
| Triangular Arbitrage | Currency/crypto arbitrage detection | triangular-arbitrage |
| Order Playlist | DJ playlist optimizer using Camelot harmonic mixing | order-playlist |
| Batch Resize | CLI image batch resizer | batch-resize |
| PyTorch Learning | PyTorch fundamentals tutorial series | pytorch-learning |
| PyTorch Audio Learning | TorchAudio tutorial series | pytorch-audio-learning |
Vault-only project pages (concept/planning, no separate repo): DJ Set 1, Set Mastering, OOP Neurons, New Cycle. See the INDEX for the complete list.
The bulk of the vault. Major clusters (full detail in the INDEX):
- Philosophy — morality, logic & math, epistemology, legal theory, metaphysics, civilizational dynamics; the Four Trunks organize it
- Economics — value, profit, risk, business cycles, the accounting-identities-aren't-models series
- Gaming & Game Theory — games as laboratories for systems thinking: MIRR 4X framework, BattleValue, randomness as termination (N≥3), the dominance-frontier lens
- AI & Computation — the Cyborg Model, the LLM Grounding Problem, planner-LM composites, energy-based models, transpilation as grounding
- Online Debates — extracting signal from "bloodsports" debate spectacle
Method — a library of good procedure: how to run hard, agent-driven projects without drifting — the Anchor Method, the Specimen & Thesis two-altitude ledger, and Author Web, Derive Hierarchy.
This vault is maintained collaboratively with Claude Code. The root CLAUDE.md provides conventions and instructions for the AI assistant. The public site is built with Eleventy and auto-deploys to GitHub Pages on every push to master.