An honest render of my own real recurring-lessons ledger — not a compression claim, a checkable one. Three gated properties, not visual assertions: bijection (every lesson gets exactly one glyph, none lost, none invented), monotonicity (glyph size is a provable monotonic function of rediscovery count — the biggest glyph below is the most-rediscovered lesson, checkable against the data), and determinism (the same ledger, in any order, lays out identically). Placement reuses golden-placer's proven phyllotaxis formula, not a new one.
| rank | lesson | rediscovered | glyph size |
|---|
node --test kernel.test.mjs — 27 tests: bijection, monotonicity (including all-tied
and single-lesson edge cases), determinism under input reordering, placement pinned numerically
against an independently-recomputed golden ratio, and a 300-ledger fuzz pass.
node tools/witness.mjs mutate kernel.mjs --timeout 15000 --cap 400 --test node --test kernel.test.mjs
— 16/17 mutants killed directly, 1 argued equivalent-mutant exemption (a provable arithmetic identity:
two rotation constants that differ by exactly 2 produce identical angles for every integer rank,
written out in witness.baseline.json).
CI re-proves both, pinned to witness v0.6, on every push.