v0.1.5

The Hundred: five-ball sets reach the chase rate and score parsing

mcp.go reported every Hundred chase 20% high: 'per over required' computed
against six when a set is five. Now uses the state's balls-per-over and
names the unit, so a Hundred chase reads 'per set'.

ParseScoreString was the worse one. ESPN reports some innings as '(N
balls)', and converting with a hardcoded six turned 30 balls of a Hundred
innings into 5.0 overs when it is 6.0 five-ball sets, making every ball
count derived from it wrong. ParseScoreStringBPO takes the format; the old
name keeps its six-ball meaning for existing callers.

v0.1.4

Port the fitted Hundred model: own segments, own par, own Elo pools

Same commit as cricketfornoobs.com, files ported wholesale: hnd-1/hnd-2
fitted on 339 Hundred matches with five-ball arithmetic (held-out chase
log-loss 0.392 vs 0.482 for the T20 segments as previously fed), par 136.6
instead of T20's 165, and per-competition franchise Elo. The explainer now
also carries the multi-day segments and their Elo table; nothing here calls
them yet, but the packages stay identical to upstream, which is the point.

v0.1.3

The Hundred's five-ball overs reach every figure this server computes

BallsPerOver was a constant six, so a Hundred state was reasoned about as
120 deliveries when only 100 exist — balls left ran ~20% high and the win
model overpriced every chase. MatchState carries balls-per-over (six unless
set), live Hundred matches stamp five automatically off the league name,
and cricket_win_probability accepts balls_per_over for hypothetical states.
Same fix as cricketfornoobs.com; ported hunk for hunk with its test.

v0.1.2

v0.1.2 — module path follows the repo rename to mcp-cricket

v0.1.1

v0.1.1 — server identity is cricket-mcp throughout