E01 · F06 · CHK-009 — Checkpoint in the middle of F06
Pause between the model and the playground: bring documentation and methods up to date as the project grows, before going on (ADR-0028, /run skill).
Acceptance criteria
- Given the stories merged since the spike, when the checkpoint runs, then
docs/domain/, derivations, references, conventions,CLAUDE.mdand.claude/rules/describe the code as it is. - Given the light evolvability check, when it runs, then each of its six questions has a one-line answer and anything worth acting on is in the improvement log.
- Given the user stories merged so far, when the checkpoint ends, then their Product Owner test cards are listed for the next stop.
Tasks
One task = one commit, referenced as CHK-009.Tn.
Light evolvability check (2026-10-12)
| Question | Answer |
|---|---|
| Size | 116 modules in src/ (model 49, geometry 40, math 17, io 6, render 4); 662 tests; geometry grew by 10 small files for one rule, each one export as ADR-0019 wants |
| Tools | mutations by the auditor caught one real gap (the pentagon's highest scale untested) and one unobservable clamp, both fixed; StrykerJS still blocked |
| Fitness functions | the bisections rest on two monotonic properties checked by sweeps of 6 001 values per number of corners, not proved: they run in every test run, about 0.3 s; proving them is in the improvement log |
| Patterns | a rule found by bisection on a sound interval: bisection is generic and reusable (a rule of E13's layout or E14's curves may need it) |
| Agent configuration | CLAUDE.md state line brought up to date (126 lines); three pitfalls added to .claude/rules/library.md (a required model field, bisection precision, an unobservable clamp) |
| Process | the Product Owner chose the rule from a research answer with three options and numbers: the questionnaire with numeric examples worked; a research request cost one round trip |
Test cards so far
None yet: US-021 comes after this checkpoint.