Development Tracker & Plans

No single target. Change detection compares windows (late 2026 vs. earlier 2026; 2025 vs. 2026 at the same level, rates the strike zone moves net of the league’s own change); plans translate model outputs into goals with targets that beat regression to the mean and checks timed to the sample.

• MLB Statcast 2024–26; Triple-A and Florida State League tracking 2025–26 (Stuff graded with the production model)
• Pitch Lab dead weight and pitch design, each pitcher’s own 2024–25 velocity and pitch shapes, Keystone Location, health warnings, Hitting Lab decisions and bat tracking
• League and level percentile distributions for targets

• A change is reported only if it clears a practical threshold (e.g. 0.8 mph, 4 points of chase) and the t bar (about 3 standard errors, stricter with few games; 8+ games in each window), with standard errors clustered by game (pitches in one game aren’t independent)
• Every season is centered on its own league: Stuff and Location are season-centered grades, and a year-over-year change in a rate the strike zone moves (zone rate, chase, whiffs, contact, walks, strikeouts, xwOBA) is measured net of the league’s median change at that level (MLB’s 2026 zone moved them league-wide). Velocity, bat speed, swing shape and exit velocity are raw changes: netting them would remove a level’s typical development (Florida State League repeaters gain about 0.7 mph), not a measurement shift
• Stuff changes are graded by the one production model for every season, each on its own league scale (the site’s season grades are cross-fit, which adds fold-to-fold drift that isn’t the pitcher’s); big-league Stuff levels are anchored to the published season grade, so a level matches the pitcher’s page. A big-league Stuff change counts as clear only if a model that never saw the pitcher also puts it at 5+ points (the same bar as Pitch Lab’s alert)
• Comparisons stay within one level, so a promotion isn’t mistaken for a change in skill
• Plan goals require a model output behind them and a target that is an improvement on today; if nothing qualifies, the plan is shorter
• Pitching targets never ask for movement a pitcher hasn’t shown: velocity goals compare 2026 (without the month before an IL stint) to his most recent prior season in the same role, net of normal aging; shape goals regrade this year’s pitches with last year’s movement on one model and need a 5+ point gain from shape alone
• Pitch-mix changes and new pitches come from the Pitch Lab as published: a mix change only for a side where the replacement wins by 2+ standard errors, with its run value; a new pitch only when it holds up under every model
• Rate targets beat regression to the mean: his expected next season with no plan (fit on one-year pairs, each season relative to its league or level) plus the step about a quarter of similar players beat. Big-league hitters and walk rates also use a reachable step (the median, his own best season on today’s league basis, or a big one-year gain) and take the more ambitious of the two. A platoon goal regresses his gap vs. left-handers toward the league’s gap for left-handed hitters, and is set only when the Hitting Lab’s shrunk split is at least 0.3 runs per 100 pitches worse against lefties
• Big-league approach goals follow the Hitting Lab: only when its regressed swing-decision value (2025–26 pooled) is past its line, on the cause it names (chase rate for a chaser, hittable pitches taken for a passive hitter). Decision value is about half runs (the lab’s run scale), and a goal’s runs at stake use it at his 2027 plate appearances
• A Florida State League walk- or strikeout-rate goal also gets box-score targets for High-A and Double-A: the FSL target shifted by the gap between that league’s 2026 rate and the FSL’s
• The big-league target rules are tested forward: fit on 2024 → 2025 pairs, the share of targeted 2025 players who reached the target in 2026 with no plan (see the Development page). Prospect rules can only be set on 2025 → 2026 itself, about a quarter by construction
• Each check is timed to the sample: the All-Star break only if his sample by then puts the target two standard errors from today; otherwise season’s end, or the sample it needs; relievers get a batters-faced check. A goal that would need more than 1.5 times his 2026 sample gets no check: it is a lean
• Medical clearance comes first only for a stint still open at season’s end or one he barely played after, and for anyone on an injured list today at any level (minor-league lists included); a stint he came back from is a medical note. The winter’s departures (the decisions feed, likely minor-league free agents included) get no plan
• A clear decline on a 2027-roster player that no goal covers gets a diagnose item; a Winter Plan position change gets a position goal from Statcast outs above average

• Public pitch tracking covers only Triple-A and the Florida State League (Single-A). High-A Greensboro, Double-A Altoona and the complex leagues are untracked, so a player promoted out of the FSL is checked in box scores, and the youngest players aren’t covered.
• Detects that something changed, not why (health, mechanics, intent or plan).
• A clear late-season change keeps only part of its size into the next season (persistence backtest: about half for fastball velocity, less for Stuff, measured on each season’s own scale), so a single late flag is provisional.
• Many goals can’t be confirmed in one season at a player’s sample; those are marked leans.
• Targets are statistical, not individual ceilings a coach would set.
• A lost shape or velocity may not come back (injury, age); the plan names the gap, not the cause.

• Share of plan goals met at their check, against the share expected with no plan (about a quarter in sample; the forward test is on the Development page)
• Run value change for players who met goals vs. those who didn’t