One-run records and team BABIP

The question

Sabermetrics has two well-worn labels for a lucky baseball team. One is a record in one-run games out of line with its overall record, read as coin flips that happened to land. The other is a team batting average on balls in play above its own norm, read as hits that will stop falling. Both get discounted as luck in the standard analysis. Team Sustainability was designed to do the same, and the design said in advance what would happen if the labels turned out to persist.

What was tested

The claim, written before the run: a team's one-run-game record and its BABIP relative to its own prior norm carry no information about the rest of the season. Five MLB seasons, 2021 through 2025, thirty teams, at a quarter, a half and three quarters of each schedule, measured as the rank correlation between the season-to-date value and the rest-of-season value. The registration stated the consequence: persistence means the label is partly skill, and a model must not discount it.

What turned up

Both persisted. A team's one-run record carried into the rest of the season at a rank correlation of +0.21, and its BABIP against its own norm at +0.43. The claim was falsified, and the luck decomposition was cut from Team Sustainability before it shipped.

For scale, the thing the model does read held everywhere: teams far ahead of what their scoring margins support fell back and teams far behind recovered, in every season of all three sports, with the rest-of-season win-rate spread between the unluckiest and luckiest quartiles at +0.03 in MLB.

What it means when you ask

Ask whether a team's record is real and DataBaller answers with the gap between its record and its margin-implied record, in wins, because that gap is the part that reliably closes. It does not tell you a team is winning too many close games or that its hits will stop falling, because in this data those readings are partly skill and a full discount would overreach. A one-run record is reported as a fact about the season so far, not as a debt.

What it does not mean

The persistence is partial, so this is not evidence that close-game skill is large. It is team level: an individual hitter's BABIP regresses toward his own established rate, which is a separate and better-studied question. And the own-norm baseline is roster-bound, since last season's team is a different roster, so part of the BABIP persistence is roster change rather than repeatable skill. That is recorded as diagnosis, and a roster-aware norm was not tested.

The record

Measured 23 August 2026 with backtest_team_sustainability.py, five seasons of thirty teams. Public reliability work puts a team's one-run record at about 0.17 against itself over a season's worth of games, which is compatible with the +0.21 measured here: weak persistence, not zero. Team Sustainability shipped with the record-versus-margin claim alone, and each sport's next completed season re-tests it. The full portfolio and its testing rules are on the decision metrics page.