Rival Scouting Splits
Expected run value by pitch type in two layers, the club’s level and each pitch against that level, with reliabilities measured within the season and across seasons.
What it predicts
Run value per 100 pitches (expected, from batted-ball quality), against the league and against the club’s own average.
Inputs
- Every 2024–26 pitch (Statcast) with Keystone’s expected run value
- Pending free agents (the FA Board’s definition) and each player’s club today
- The Outlook’s 2027 projections; FanGraphs 2027 payroll estimates
Method
- Level = usage-weighted run value against each pitch type’s league mean; within = each pitch type against the club’s own level
- Reliability: mean split-half r over 20 seeded random halvings of the season’s games, Spearman-Brown to a full season, for each layer
- Each value shrunk toward zero by its own sample; clear = 1.5 posterior SDs from zero
- Year-ahead persistence (2024→25, 2025→26) and a forward test: did clear staff weaknesses stay negative the next season, and did clear within-staff flags keep their sign (sit-on and lay-off flags counted apart)?
- 2027 view: share of each cell thrown or faced by players likely leaving, split into players already elsewhere, pending free agents and option holders we expect declined (those decisions come in the qualifying-offer window); an option holder under six years of service stays (a declined option leaves him under club control); the level over the players who aren’t leaving, ranked against every club’s; the Outlook’s projection; best 2027 players by projected WAR; each rival free agent’s fit to Pittsburgh’s plan and shopping lists (FA Board)
How well it works
Computed by the pipeline from the data on this site
- Next-season test: cells flagged clear one season, scored the next (how many kept their sign; mean runs/100 flagged vs. next season)
- 2024→2025 against the league: 7 of 8 held (−0.46 → −0.60) · within the club’s own level, sit-on flags: 4 of 5 held (−0.34 → −0.37)· lay-off flags: 3 of 4 held (+0.37 → +0.27)2025→2026 against the league: 6 of 7 held (−0.49 → −0.53) · within the club’s own level, sit-on flags: 2 of 2 held (−0.29 → −0.17)· lay-off flags: 3 of 3 held (+0.37 → +0.25) Few flags each season, so read these as counts, not rates.
- Year to year: does a club’s pitch-type pattern repeat? (r across club × pitch-type cells)
- Staffs 2024→2025: overall level .71, pattern within the club .17 · 2025→2026: overall level .59, pattern within the club .08Lineups 2024→2025: overall level .68, pattern within the club .20 · 2025→2026: overall level .46, pattern within the club .01 A club’s overall level carries over; which pitch types it handles better or worse than its own level barely does. The page shows the pattern regressed for that.
- Within one season: random halves of the games (mean of 20 seeded splits)
- Staffs level .75 · within the club .31 · raw cells .42 (a full season’s reliability: 59% for raw cells)Lineups level .45 · within the club .20 · raw cells .22 (a full season’s reliability: 36% for raw cells)
Data: 3 seasons of MLB Statcast (2024–2026), 2,130,654 pitches (every pitch with Keystone’s expected run value); FanGraphs 2027 payroll estimates.
What it can't see
- Within-staff and within-lineup patterns barely repeat from one season to the next; staff and lineup quality do.
- Splits describe 2026 rosters; clubs with heavy turnover (more than half their pitches leaving) are faded. The returning-players level leaves out everyone they add this winter.
- Per-hitter edges live in the Matchup Engine, not here.
How we'd measure impact
- Hit rate of clear staff weaknesses the following season