Alignment Harness: documentation index

This is the root of the documentation. Each line below tells you what that part holds, so you can see where to go without opening anything. The parts nest in order: first the moments where agents drift, then the loops that run through those moments, then the individual pieces, then the plan for extracting and testing them.

The parts, in order

1. The fulcrums. Eighteen moments, from install to the next session, where an agent's understanding can separate from what the person meant. For each: what the person experiences, how it compounds, what the harness does, and which pieces act there. It starts with the thesis that misalignment compounds.
2. The core loops. Seven cycles that run through the fulcrums again and again: every turn, every task, every delegation, every session, every correction, the learning flywheel across sessions, and the harness checking itself. Each loop names what it keeps true for the person.
3. Per-piece pages. One page for each skill or hook in the installable version: what it does for a person, how to set it up, and how to tell whether it is set up and working. They ship with the plugin, and the website links each piece on the fulcrum map to its page once that page exists.
4. Extract and test plan. How the chosen pieces get pulled out of Jonathan's setup, what setup and switches a new person will need, and how the experience a user gets will be tested.

Supporting material

• README. The front page: who the harness is for, what it is trying to accomplish, the thesis, the fulcrums in brief, and an honest status.
• data/fulcrums.json. The same fulcrum map and loops, as data the website reads. The fields are defined in lib/site-types.ts. A piece name ending in .sh is a hook; any other name is a skill.