MLB Bullpen Analytics · Decision Support
A decision tool that turns a reliever's recent workload into a run-value penalty — measured on the exact scale a coaching staff already uses to compare matchups. So the dugout can finally see the moment a tired top-tier arm stops being the better call than a fresh second option.
Developed hands-on with MLB teams — working directly with a club's GM and pitching coaches to fit the tool to how real in-game and workload decisions actually get made.
The Problem
Staffs already rank every reliever-vs-hitter matchup on a run-value scale. But that grid treats a pitcher the same whether he's fully rested or three appearances into a hard week — and the people tracking arm workload sit a department away from the people making the 7th-inning call.
The matchup grid alone
With the Arm Tax applied
The Idea
The whole design rests on a single move: express fatigue in the same run-value units as the staff's projections. If the penalty and the projection share a scale, the dugout can simply subtract one from the other — no new mental model, no competing dashboard.
Their number
What each arm is worth in a clean matchup, on a run-value scale.
Our number
A fatigue penalty in the very same units — earned from this week's usage.
The call
Projection minus tax. The honest ranking of who's actually best tonight.
How It Works
The pipeline fuses public pitch-tracking and game logs with the club's private feeds — bullpen warm-up counts and their proprietary talent numbers — ingested automatically off their existing internal workflows. Each stage is calibrated to the individual pitcher rather than a league-wide rule of thumb.
01
Public pitch tracking and game logs, plus the club's warm-up counts and talent numbers — auto-pulled from their internal workflows.
→02
An acute-vs-chronic ratio flags whoever has spiked off their own baseline.
→03
A decay model converts recent load into a run-value penalty per pitcher.
→04
Adjusted values ranked into a clean go / caution / hard-stop card.
The Model · the shape, not the recipe
Fatigue isn't one thing, so the penalty is built from a handful of load sources, then decayed by how long ago they happened and escalated when a pitcher is used on back-to-back days.
Bullpen pitches thrown while getting hot count too — weighted toward game intensity, not ignored.
Pitches beyond a pitcher's workload threshold cost more, mirroring how recovery time climbs.
Yesterday hurts more than three days ago; load fades on a per-pitcher recovery curve.
Back-to-back days escalate the penalty, with a hard stop at three straight.
Workload Spine
The sports-performance group monitors an acute-to-chronic workload ratio — recent volume against each pitcher's own baseline. The tool computes the same ratio so the model lives inside the staff's existing language instead of competing with it. A spike off baseline is the trigger.
< 0.8
Under-worked
0.8 – 1.3
In the sweet spot
1.3 – 1.5
Elevated — watch
> 1.5
Spike — at risk
Calibration
Two calibration steps keep the model honest: one learns each arm's personal fatigue sensitivity from the tracking data, the other reverse-engineers the blend of public projection systems that best matches the club's proprietary number.
Per-pitcher sensitivity
Each arm's recovery rate and fatigue scale are estimated from rested-vs-tired performance splits — durable arms get taxed lightly, fragile ones heavily. Defaults give way to data as the season fills in.
Projection blend · NNLS
A non-negative least-squares fit finds the mix of public systems that tracks the staff's internal projection. The solution is deliberately sparse and legible — "these two systems carry it" — so the staff can trust what they can read.
A Signature Finding
Staffs widely watch velocity for fatigue. Testing whether spin or velocity drops first on short rest — controlling for within-game fade and pitch mix — surfaced something the existing process misses.
The drop was statistically significant; the velocity reading was flat. For that pitcher, spin is the earlier warning light — and a velocity-only watch would miss it entirely. Another arm showed the textbook opposite: velocity led, spin was noise.
The takeaway isn't a league-wide rule. It's that fatigue has a personal signature, and the tool flags which metric to watch for each arm. Just as telling: where public data can't read a pitcher cleanly — usually because he's only sent out on short rest when he feels sharp — that selection bias is exactly what the club's private warm-up and workload feeds, ingested automatically, are there to cut through.
The Output
Everything resolves to one card the dugout can read in seconds: each arm's projection, its fatigue tax, the adjusted value, and a plain verdict. Illustrative names below — the structure is the point.
On paper, Arm A is the better matchup. After fatigue, the fresh Tier-2 arm grades out ahead — exactly the trade-off the staff couldn't see before. And Arm C is off the board tonight on a biomechanical hard stop, not a hunch.
The Arm Tax — fatigue-adjusted bullpen decision support. · Client and player details anonymized. · Portfolio piece.