By: Tom Sydney Kerckhove <syd@cs-syd.eu>
Cut the plan to the rules and the context 4,840 lines of planning material across fifteen documents, for a bootstrap that is built. What was argued in them is now either code or wrong, and the code is the better record of whichever it turned out to be. 1,051 lines survive. Kept: rule-triage.md, which is the 122 rules still to build, and modularity.md, which is what shape each one takes. plan.md keeps the milestones and loses its M0 section, that being the one that is done. New: context.md, which is what a session needs before touching this. The seven invariants, said as invariants rather than as the arguments that produced them. The three levels. What is deliberately not built. The measured numbers. The open decisions, still split by owner. And the one thing I would not have thought to write down if it had not just been measured: the infrastructure-to-rules ratio is 4,806 to 599 against a predicted 1,300 to 55, so the bet that a rule is a module and a line is currently off by an order, and rule four is what settles whether the three chosen were simply the expensive ones. Deleted: annotations, annotation-syntax, architecture, comment-attachment, corpus, data-model, generated-code, levels, obligations, simplification, bootstrap, and the notes file that was the README's own overflow an hour ago. check-plan.sh follows them down, from 200 lines to 189: the assertions that lost their subject are gone, and the ones with a subject still hold. It is the same feedback loop over a fifth of the prose.