The old EXPOSURE paragraph read the fight at round five and quoted bucket
figures no guard held. Round five is where the outcome becomes legible; round
two is where it is decided. Reframed on the first disabling blow and cited to
tools/first-blood-baseline.json, so every figure in the paragraph is one the
build re-measures:
cut 50.3% first blood at mean round 1.8, 25.5% vs 65.5% wipes, swing 40
line 54.1% at round 1.3, 7.8% vs 23.2%, swing 15.4
An earlier draft carried 43 for the cut swing from a single seed. Three seeds
give 40.0. The note stays in the document.
Also adds a Casting coupling warning: the declared cast in the `cast:` marker
is load-bearing on check-rollable and check-firstblood, and re-casting
invalidates the recorded swing. Verified rather than assumed — swapping
braithwaite for agyeman in the marker fails both guards and the build, and
check-firstblood names the fix:
"The recorded swing describes the old cast; re-record with --update,
do not revert the cast."
npm run check: 13 guards, exit 0.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
