Using Starters in Relief: A Controlled Feature Test

WhatIfSports Baseball Simulation · April 2022 · 162-game controlled experiment

Background

A long-standing community question about WhatIfSports simulation: does assigning a starter-type pitcher to a reliever role cap or reduce their in-game pitch count effectiveness? One long-circulating claim held that the sim applied a 25% reduction to the in-game pitch count maximum for any starter used in relief. This test was designed to isolate and answer that question with hard data.

I put in two test teams with identical rosters, pitching staffs, and advanced settings. The only difference between the two teams was the role assignment for the four pitchers being tested. On Team 1, the pitchers were assigned as relievers (RP). On Team 2, they were assigned as starters (SP).

Test Design

Pitcher Selection

The pitchers were chosen specifically for their high IP/G. Using high-IP/G pitchers who could throw a significant number of pitches makes any cap on in-game effectiveness clearer and easier to detect. If there were a hard cap or percentage reduction for starters used in relief, it should show up most visibly in pitchers expected to reach 133-135 pitches before hitting 0% in-game fatigue.

Settings

Each pitcher’s advanced settings were identical on both teams: PC set to allow up to 140 pitches, capturing as much of their IP/G allotment as possible. The 140 PC ceiling was chosen to ensure any cap (claimed to be around 100 or ~75-80% of the normal max) would be clearly visible in the post-100-pitch data.

If there was a hard 100-pitch cap for SPs used in relief, pitchers on Team 1 should degrade sharply at or after pitch 100. If there was a 25% reduction (as the community claim suggested), degradation should unfortunately also begin around pitches 99-101 (depending on pitcher). If there was no cap at all, performance before and after pitch 100 on Team 1 should match Team 2’s SP results.

Season Results

Summary Stats

The results were not close. The same four pitchers, with identical rosters behind them and identical advanced settings, produced wildly different outcomes based solely on role assignment:

NP/PA was essentially identical across both teams for all four pitchers (3.33–3.37 range), confirming the pitchers faced comparable batters per inning and the difference in outcomes was not driven by sequencing or luck in at-bat composition.

Pitch-split analysis: isolating where it breaks down

The Hunter data below (10 games) illustrates the pattern that repeated across all four pitchers. Pitches 1-100 produced starter-quality results. Performance after pitch 100 collapsed entirely.

For pitches 1-100, Hunter posted a .239 OAV, 4.23 ERA, 1.20 WHIP — comparable to what you’d expect from a quality starter with his real life stats. For pitches 100+, those same numbers became .333 OAV, 13.51 ERA, 2.61 WHIP. The K rate fell from 2.92 to 0.71. The BB/9 jumped from 2.34 to 9.95. The HR/9 nearly tripled. In several games where Hunter was pulled before or exactly at pitch 100 (games 3, 4, 8), the post-100 line is blank — the degradation pattern only appears when he exceeded the threshold.

Pitch counts at specific split points were estimated by taking Hunter’s final pitch count for each game and subtracting his average pitches per PA, counting PA backwards to 100. This is a slight approximation, not a frame-by-frame extraction, but the pattern was clear enough that the approximation doesn’t affect the conclusion.

Findings

Assigning a starter-type pitcher to a reliever role does impose an in-game pitch count cap. Based on this data, that cap appears to be at or very close to 100 pitches — not a percentage reduction from the pitcher’s normal max.

The 25% reduction theory (which would have put the cap around 99-101 pitches for these pitchers) and a flat 100-pitch cap produce nearly identical results for high-IP/G pitchers around the 130-pitch range, so this data alone can’t definitively distinguish between the two. To isolate whether it’s a hard 100-pitch cap or a percentage reduction, the follow-up test would use lower IP/G pitchers who are naturally good for around 100-105 pitches:

If no degradation: it’s a flat 100-pitch cap (those pitchers aren’t exceeding it regardless of role).

If degradation appears around pitches 75-80: it’s the 25% percentage reduction, cutting their effective max from ~100 to ~75.

This test was one of the most clear and easiest to gather data on I’ve done at WIS. The results were visible from the first few games and held for all 162. The difference after pitch 100 is remarkable for the team used in relief, and unnoticeable in the team used as starters.

Community Context

The question originated in a forum thread discussing starter usage in relief roles. The community claim (sourced to a WIS staff comment years earlier (lost to a change in the forums around 2008)) was that the sim applied a reduction to the in-game max PC for any starter assigned to a relief role. That claim had circulated without hard data to support or refute it.

This test confirmed that a meaningful cap exists. It also prompted the follow-up question about whether the mechanism is a flat 100 or a percentage. The data here is consistent with either interpretation for this specific set of pitchers, but the test design was intentionally chosen to surface the effect clearly; and it did.