StoneCypher / jssm / 9c22e282e4ebf1fb033e94220fa3eaed0625ed85

100%

master: 100%

push

github

test(dragon): §6 arrow-decoration dragon tier + fix a fake stoch test it found (WP-6)

New arrow_decorations.maximal.ts (dragon §6), asserting behaviour first confirmed
against the parser:
- duplicate rejection at N>=3 (stoch only tested N=2), both sides;
- duplicate rejection with IDENTICAL values (detection keys on kind, not value);
- whitespace-run invariance at every inter-decoration gap (AST-equal);
- block/line comment-interleaving invariance at those gaps (AST-equal);
- malformed decorations (e.g. `N%%`) are reported as parse errors, never a
  duplicate misfire and never silently parsed.

Writing it surfaced a real find: arrow_decorations.stoch.ts's `shuffle` used
`rng() % (i+1)` while gen_splitmix32 returns a float in [0,1), so the index was
fractional and out[j] was always undefined — every shuffle collapsed to
[out[0], undefined, ...]. The three order-invariance tests were therefore fake:
both sides degraded to the same single decoration and passed without testing
order invariance. Fixed to Math.floor(rng()*(i+1)) (real Fisher-Yates) and added
a deterministic "shuffle is a real permutation" regression. Order invariance now
genuinely holds under real 4-decoration permutations (stoch still green, 12).

dragons-egg.md: dragon tier marked live, find-handling convention documented,
§6 status updated, the shuffle find logged.

Verified: vitest-dragon green (16), vitest-stoch green (12), eslint clean.

2595 of 2595 branches covered (100.0%)

Branch coverage included in aggregate %.

4158 of 4158 relevant lines covered (100.0%)

47842.48 hits per line