How to spot a breakout before the wRC+ moves

A. xwOBA - wOBA surplus

xwOBA is what a hitter "should have" wOBA'd given his batted-ball profile (exit velo + launch angle + pull/oppo, fed through Statcast's mapping). Real wOBA is what actually fell in.

When xwOBA outruns wOBA by ≥ 30 points, the contact has been better than the box score. That gap closes 70% of the time within the next 50-100 PA. It's the cleanest "regression upward" signal in the sport.

example: hitter line at 200 PA
  xwOBA  .370  ← contact quality is All-Star tier
  wOBA   .335  ← actual results are league-average
  gap    .035  ← full A-signal

The two paths that close that gap: (1) the BABIP catches up as bloops fall, or (2) HRs start clearing the wall instead of dying on the warning track. Usually both.

B. Quality-of-contact composite

Barrel rate (≥10% reads loud) and hard-hit rate (≥45% reads loud), normalized against typical regulars. A hitter who's barreling 12% and hard-hitting 50% is making elite contact regardless of what the surface line says.

Why we trust this: barrel% stabilizes around 50 PA, the fastest of any Statcast metric. By the time you've seen a hitter for a month, you know if his quality of contact is real.

C. Plate discipline

K-BB% is the headline discipline number: lower is better, league average ≈ 13%. When K-BB% drops 4+ points without a corresponding loss of power, you're seeing real swing decisions improve. That's the rarest of the three signals; when it shows up alongside A and B, the breakout almost always lands.

We weight them A 50% / B 30% / C 20%, then amplify by age:

The age amp is asymmetric on purpose. A 22-year-old running A+B+C is the bet of the season. A 33-year-old running the same line is probably just having a good month. Same signal, different prior.

BABIP alone. High BABIP can mean elite contact OR pure luck. Without quality-of-contact context it's a coin flip.

HR pace. Counting stats. Three home runs in a week tells you he's been hot, not that he's leveled up.

SLG over expected SLG (xSLG). Useful at the season level but noisy at small samples, xSLG itself has high variance over 50-100 PA.

Sprint speed deltas. Cool to track, doesn't predict offensive breakouts. A hitter who got 0.2 ft/s faster is still the same hitter.

Walk rate alone. Without K% context, a walk-rate spike often comes from being pitched around, not from better discipline.

The textbook breakout: high A, B, and age tailwind

A 22-year-old prospect showing xwOBA .380 / wOBA .320 (A signal: 0.83), barrel% 13% + hard-hit% 48% (B signal: 0.80), K-BB% 0.08 (C signal: 0.75). Age multiplier 1.30×.

Composite raw: 0.5·0.83 + 0.3·0.80 + 0.2·0.75 = 0.805 Final score: 0.805 × 1.30 = 1.0 (clipped from 1.04)

This is the strongest reading our model produces. A profile like this becomes a 130+ wRC+ hitter ~70% of the time within the next 200 PA. Three of the four "breakout of the year" winners over the last decade ran this profile.

The borderline: high signal, no age leverage

A 31-year-old vet with the same exact A, B, C scores above. Age multiplier 0.65×.

Final score: 0.805 × 0.65 = 0.52

A real talent signal that doesn't get the "breakout candidate" tag because we've already seen who this hitter is. He's playing well, but at 31 the underlying skills are baked. The model treats him as a hot streak, not a breakout. Some of these get away from us (late-career adjustments do happen) but they're rare enough that the prior wins.

The trap: high A, weak B and C

A 26-year-old at xwOBA .375 / wOBA .310 (huge A signal of 1.0), but barrel% 7% / hard-hit% 38% (B signal: 0.25) and K-BB% 0.18 (C signal: 0.25).

Composite raw: 0.5·1.0 + 0.3·0.25 + 0.2·0.25 = 0.625 Final: 0.625 × 1.0 = 0.62

This one looks juicy on the xwOBA-wOBA axis, but the contact and discipline pieces don't support it. What's likely happening: he's barreling rarely but those barrels are leaving the yard at a high rate, while most of his batted balls are weakly hit. The "expected" number is being driven by his rare loud contact, but his floor isn't there. The wOBA might catch up to xwOBA via more home runs from those rare barrels, but the hitter you're projecting forward isn't a star.

The composite score itself is removed from the leaderboard table: the column added more noise than signal when sorted, and the underlying ingredients (xwOBA, barrel%, hard-hit%, K-BB%, age) are already in the table side-by-side. Sort by xwOBA, eyeball wRC+ in the same row, and the breakout candidates surface themselves.

The scoring logic stays in services/mlb/breakout.py for future use, likely as a sort dimension on a dedicated "Watchlist" tab once we have the ground-truth labeled history to validate the heuristic against actual breakout outcomes.

This is a heuristic, not a trained model. The 50/30/20 weights and the age multipliers were calibrated from public xwOBA-wOBA reversion studies (Eno Sarris, Connor Kurcon) and our own sense of which age bands historically produce breakouts. We don't claim the weights are optimal. We claim they're defensible and the column you look at second is wRC+, not the breakout score.

The training pipeline scaffolded in app/jobs/build_breakout_training_set.py extracts the labeled (player N → wRC+ N+1) pairs we'd need to learn the weights properly. Once we have a year of cross-validation data, the heuristic gets replaced by the trained model. Until then, sort by xwOBA, and trust your eyes more than the composite.