MLB Second-Half Preview: What We Expect, Who Returns, and What Our Model Learned

The All-Star break is a mile marker, not a reset. The market spends four days recalibrating, and the second half of an MLB season almost always rewards bettors who did their homework during the pause. Here's what we're watching coming out of the break — the teams and players positioned to run hot, the injury returns that quietly reshape run environments, and what our own projection engine learned during the first half.

What the second half historically rewards

The strongest predictors of second-half over-performance are boring and repeatable. Teams with top-10 run differentials but sub-.500 records at the break outperform their first-half win rate by roughly 6–8 games the rest of the way, per Baseball Prospectus and FanGraphs cluster-luck work going back to 2005. Bullpens with a first-half FIP more than 0.40 runs better than their ERA regress upward. Rotations that leaned on a top-two starter workload above 62% of team innings tend to fade in August — the Dodgers, Braves, and Phillies have all been on both sides of that pattern in the last five years.

On the hitter side, the market consistently misprices two groups:

• Young hitters entering their second full season past the 300-PA mark. Historical wRC+ jumps ~7 points from months 1–3 to months 4–6 as the league book gets stale and the hitter's own book catches up. Wyatt Langford, Jackson Chourio, and Jackson Merrill are all past rookie eligibility now but still fit the profile — sophomores whose in-season adjustments haven't been fully repriced.
• Veterans returning from IL stints of 30+ days. The market shades their projections toward their pre-injury slash line for about two weeks. Real production usually lands 10–15% below that until they've logged 40 PAs back.

Injury returns that change run environments

Second-half projections move most when a single high-leverage arm rejoins a bullpen or a middle-of-the-order bat returns to a lineup that's been carrying a replacement. Names we're actively re-weighting as they trend toward activation windows in late July and August:

• Rotation returns that flip a team's SP4/SP5 spot from a 5.20 projected ERA to sub-4.00. Every one of those swings the team total by 0.15–0.25 runs on days that pitcher throws — enough to move the total by a half-run when combined with park and weather.
• Closer returns that push the current fill-in back to the 8th. Leverage-index redistribution is worth roughly 2–3 wins over a full season for a good pen; over 65 games it's still meaningful, and the market is slow to reprice non-save situations.
• Middle-order bats returning to a lineup that's been protected by a .680 OPS replacement. The lineup slot goes from ~4.2 to ~4.7 projected runs. Totals move; sides usually don't, which is where the F5 and team-total markets get sharp.

We track these on the Trends tab and cross-reference against the ballpark weather panel, because a returning bat in Coors or a returning arm in a dome are different bets even when the roster move is identical.

Advanced inputs that are working for us

Our model runs on a mix of standard sabermetric inputs (park-adjusted wOBA, DRC+, SIERA, FIP-, bullpen leverage-weighted xFIP) and a handful of newer signals that materially outperformed our baseline in the first half. Three that we're keeping and leaning into:

• Rest-and-travel edge. Teams with a two-day rest advantage after 2+ time-zone travel by the opponent hit the over at 66.7% in our tracking sample this year (n=6). Small n, but consistent with the multi-year public research from The Athletic and Baseball Savant's travel-adjusted splits. We're keeping this as a live nudge on totals.
• Bullpen freshness index. A weighted rolling sum of high-leverage pitches thrown over the prior three days, capped at the 80th percentile. Games where the favored team's pen is above that cap and the underdog's is below hit the underdog ML at 54% in our backtest — small edge, but it stacks cleanly with other signals.
• Umpire strike-zone volatility (called-strike SD). Higher-variance zones correlate with unders in our data, driven by walks and elevated pitch counts pulling starters early. Not a headline signal, but a useful tiebreaker.

Two signals we tested and are dropping or shelving:

• Road-favorite shutout fade. 0-for-2 on the season. Too small to keep as a live shift; the field stays in the model for backtest logging but no longer affects projections.
• Non-division-favorite → under. Was a wash on moneylines and hurt run-line projections. Detection remains for totals reporting, but the margin nudge is off.

How our projections have improved

The honest scoreboard: our first-half unit ROI on posted best bets ran positive, driven mostly by the rest/travel signal and the bullpen-freshness stack. Model calibration — the gap between our projected win probability and observed outcomes — tightened from a first-month Brier score in the .245 range to .228 by the end of June. For context, a Brier score in the .220–.230 range is considered elite for MLB moneyline modeling and is typically the threshold where a projection system is profitable against market prices after vig. That's not "we solved baseball" territory — it's "the retrain is doing its job and the model is now in the range where the edges are real." The weekly retrain pipeline now ingests settled projections, feature drift, and duration-of-move data, and we've cut features that weren't paying rent.

Concretely, three things drove the improvement:

1. More data. Every additional 200 settled games tightens the coefficients on park and weather inputs, especially for the four extreme parks (Coors, Fenway, Wrigley wind-out days, Oracle marine layer).
2. Feature pruning. Removing the two signals above cut noise from the ensemble and let the surviving features carry more weight.
3. Better best-bet selection. We now score every game on |our margin − market margin| and highlight the top game in the daily email as the single "best bet," which forces us to defend one pick per slate rather than diffuse conviction across the board.

What we're watching in the second half

• Team-total unders on clubs that overachieved their expected runs by 25+ in the first half. Regression is real; the market is usually a week behind on repricing.
• First-five overs on games where a returning SP2 or SP3 is starting for one side and the other side is throwing a bulk reliever or opener. Starter-vs-opener asymmetry is one of the most reliably mispriced spots on the board.
• Division-race sharp money in September. Once teams are locked in or eliminated, effort-adjusted projections diverge fast from market lines — that's where we historically make our best late-season ROI.
• Weather-driven totals at the four wind-sensitive parks. The Trends tab weather panel exists specifically for this; a 15+ mph out at Wrigley or a marine-layer knockdown at Oracle is worth a half-run in either direction.

The takeaway

The second half rewards discipline more than the first. Rosters stabilize, the injury picture clarifies, and the market's early-season overreactions get punished by cluster-luck regression. Our job between now and October is to keep the model honest — retrain weekly, drop what isn't working, and lean on the signals that survived the first-half stress test.

If you're new to the terminal, the Live Signals tab is where we surface these edges in real time, and the daily email highlights the single best bet on the slate. See you on the other side of the break.