Running Game

No single target: run expectancy and the chance of scoring by base-out state, then decision rules tested on the next season’s official success.

• Every 2024–26 plate appearance: base-out state, runs to the end of the inning and whether a run scored (Statcast)
• Official stolen bases and caught stealing, 2024–26 (MLB Stats API)
• Savant basestealing (opportunities, tracked attempts, leads), baserunning and pitcher running-game run values
• Sprint speed
• The Winter Plan’s 2027 roster and the feed’s injured-list history (decisions feed)

• RE24: average runs from each base-out state to the end of the inning, innings 1–9 (no automatic runner, no walk-off innings); extra innings get their own chance-of-scoring table
• Break-even = (runs now − runs if caught) ÷ (runs if safe − runs if caught), per base and out; the one-run break-even uses the chance of scoring at least once. Calls use second base weighted across outs by how often each comes up
• Success: official SB/CS 2024–26 weighted 5/4/3, regressed toward the league rate for his speed with a prior of about 90 attempts, measured from how success repeats (2024 → 2025)
• Green light = top-quarter sprint speed, attempting under 75% of the league rate for his speed (Savant, 2024–26), 4+ attempts, and both his own record and his regressed success at or above break-even; toss-up = the same runner with his own record under break-even (the prior alone lifts him over it); below break-even = 6+ attempts and regressed success under it by 2+ posterior standard deviations (about 4 points each), and under it by less is a toss-up; one-run runner = top-quarter speed, 4+ official attempts (a record, not speed alone) and regressed success at or above the one-run break-even; not a catcher
• Forward tests: each rule applied with data through one season and scored on the flagged runners’ official success the next. Green lights held up; toss-ups succeeded at about break-even, so they get no call or run value; runners regressed just under break-even ran above it, so a small shortfall is a toss-up, and the 2-SD rule flagged nobody in 2024 or 2025, so it has no forward record. One-run runners cleared the one-run break-even, but that test scores all their attempts, not only late-and-close ones (the pitch data have no steal events). The success prior is measured on 2024 → 2025, so only the 2025 → 2026 test is fully out of sample
• Bunts: the success rate a sacrifice needs for the chance of scoring once to break even, and the club’s own sacrifices checked against the table
• Team and staff totals split by each player’s share of plate appearances or running-game opportunities with the club

• Count, pitcher delivery time and catcher arm drive single attempts; this is season-level.
• Savant tracks about three-quarters of attempts; its rates are used only for how often a runner goes.
• A green light’s value assumes the extra attempts succeed like his chosen ones; they would be harder.
• Steal value also changes the hitter’s pitches (fastballs with a runner going), which isn’t counted.

• Pittsburgh steal attempts and success vs. break-even in 2027
• Late-and-close sends and pinch-runs vs. the one-run break-even
• Runs from the running game (Savant)