---
okf_version: "0.2"
---

Collegica, projected for agents. The human site is written in Quarto at collegica.org; this bundle carries the same knowledge as plain markdown with OKF provenance, trust and lifecycle frontmatter, regenerated from the source tree on every build. This is a proof of concept covering the Finance subject; other subjects follow the same projection.

# Research Reports

* [How this bundle is made](about.md) - The design behind the projection: what OKF is, the finding that shaped the approach, the decisions taken, what the generator deliberately does not do, and how to extend it to the rest of the site.

# Sections

* [Aging well](aging-well/) - Three kinds of capital — mind, muscle and money — and a plan for each, with AI as a tool rather than an oracle. Three articles, their videos, and the research behind the job-search one.
* [AI](ai/) - Agents, protocols, and models as tools rather than magic. From September 2026: an agent packaged behind four doors — CLI, REST, MCP, AG-UI — a guide to running an agent entirely on your own machine, the local voice agents you can talk to, and the compressed vector index a private RAG stack would use — each with the research that checked its claims against the source.
* [Apps](apps/) - Small, local-first tools that ship as pages on the site: two built (the OWL Planner, in its own repository, and OWL Sessions, in this one) and three specced but not built, each because an article commits to it.
* [Finance](finance/) - Mortgages, penalties, and the fine print — starting with the question of fixed versus variable, answered by replaying real contracts against the rate path that actually happened.
* [Python](python/) - Handing Python work to someone else: recreating an environment, building something installable, and making an analysis a stranger can run. Two articles from September 2026, two from January 2023, and a runnable example carried as an attested computation.
* [References](references/) - Data files, computation code, and the executor and attester for running and checking them.
* [Robotics](robotics/) - The founding subject: robotics learned in simulation. From September 2026, two articles: a map of NVIDIA's stack — open physics engines at the bottom, a proprietary simulator on top, open-weight models beside it — and a reading of vision-language-action models, the one shape they share and which ones are actually open; each with the part that could be checked from a machine with no GPU.
* [Software development](software/) - Structure, dependencies, packaging, and keeping a codebase tractable — and, from September 2026, the setup that lets several AI agents work one repository. The Python track is its own section.