🚀

Launch Lab

Daily HR Pick Engine · ML-Powered
ⓘ ABOUT & GLOSSARY
TBD
loading picks...
Today's Top 5 Picks
TBD
Full Ranked List
ⓘ Accuracy tracking applies to the TOP 5 picks only. All other candidates are shown for reference.
FILTER BY TEAM
SEARCH PLAYER
COLUMNS
FILTER BY BADGE
Today Accuracy
TBD
pending results
Last 7 Days
TBD
loading...
Last 30 Days
TBD
loading...
Last 90 Days
TBD
loading...
All-Time
TBD
loading log...
Days Tracked
TBD
since model launch
🤖 Groq POTD
TBD
no data yet
📈 Slump Breaker
TBD
no data yet
Accuracy Trend
No historical data yet — check back after results are in.
Pick History
Date Picks Hit Accuracy Best Pick Score
Loading history...
Today's Analysis
HIT RATE BY RANK
Zone Badge Hit Rates
Daily Pick Simulation Simulated 1-unit stake per pick vs estimated market odds · For entertainment purposes only
About Launch Lab

Launch Lab is a machine learning-powered home run pick engine built on MLB Statcast data. Every morning it generates a top 5 list of batters most likely to hit a home run that day, ranked by a calibrated logistic regression model trained on 3 seasons of real game outcomes.

The model learns from over 100,000+ batter-game rows where each row represents one player in one game, with features describing their contact quality profile and the opposing pitcher's tendencies. The label is simple: did this player hit a home run that game (1) or not (0)?

Picks are generated once daily at 11 AM ET, after most lineups are posted. Results are checked the following morning and accuracy is tracked over time.

How It Works
1

Pull Statcast Data

Exit velocity, launch angle, barrel rate, and hard-hit rate are pulled from Baseball Savant for every batter. The model uses a rolling average over prior games as its baseline profile.

2

Blend Recent Form

The last 14 days of Statcast are fetched each morning and blended 40/60 with the career rolling average. A player on a hot streak gets a meaningful boost even if their career numbers are average.

3

Score the Matchup

Each batter is scored against today's probable pitcher. Pitcher HR rate allowed, barrel% allowed, and an ERA-based adjustment factor into the model. Park factor is applied as an additive adjustment.

4

Rank & Filter

All batters in today's lineups are scored. A diversity rule caps picks at 2 per game so results don't sweep one matchup. Players with weak contact profiles (low EV or barrel%) are penalized.

5

Check Results

The following morning, actual HR outcomes are pulled from MLB boxscores. Players who didn't appear in the lineup (DNP) are excluded from accuracy calculations.

Stat Glossary
EV
Exit Velocity (mph) — How hard the batter hits the ball on contact, measured by Statcast radar. The single strongest predictor of HR potential. Elite HR hitters average 90+ mph; league average is ~83.5 mph. Displayed as a trailing average of all batted balls.
LA
Launch Angle (degrees) — The vertical angle of the ball off the bat. Home runs typically require 25–35°. A high average LA indicates the batter elevates the ball consistently. The model uses a trailing average; extreme groundball or popup hitters score poorly here.
BBL%
Barrel Rate — Percentage of batted balls classified as “barrels” by Statcast: 98+ mph exit velocity at an optimal launch angle (roughly 25–35° at 98 mph, widening as EV increases). The model's strongest single feature. Elite: 12%+, League average: ~6.5%.
HH%
Hard Hit Rate — Percentage of batted balls hit at 95+ mph exit velocity. A broad power indicator. The model weights this less than BBL% since hard-hit grounders don't go for HRs. League average: ~36%.
FB%
Fly Ball Rate — Percentage of batted balls classified as fly balls (launch angle 25°+). A high FB% is necessary but not sufficient for HR production — the ball also needs to be hit hard. The model weights this negatively in isolation since soft fly balls and popups suppress HR rate. Complements HHFB%. League average: ~9.5% (per batted ball).
HHFB%
Hard Hit Fly Ball Rate — Percentage of batted balls that are both hit at 95+ mph exit velocity and classified as fly balls. This is the most direct Statcast proxy for home run potential — a hard-hit fly ball is the fundamental precursor to a HR. Elite HR hitters typically post 10%+ HHFB. League average: ~4.5%.
PF
Park Factor — Multiplier reflecting how HR-friendly the game's venue is. Values above 1.0 boost a batter's probability; below 1.0 suppress it. Coors Field (1.35) is the most extreme hitter's park; Oracle Park SF (0.93) the most extreme pitcher's park. The model also applies handedness-specific adjustments: LHH get a +0.05 boost at Yankee Stadium (short RF porch), while RHH get suppression at Comerica Park (deep LC). These are deltas on top of the base park factor.
SLG
Slugging Percentage — Current season SLG for this batter. Shown as context; not a direct model input but correlates with overall power production. A high SLG alongside elite barrel rate is a strong combined signal.
vs R/L
Platoon SLG — The batter's slugging percentage specifically against right-handed or left-handed pitchers this season, depending on the pitcher they face today. Most batters have significant platoon splits; a LHH facing a RHP (favorable) will show their vs-R SLG.
H/A
Home/Away SLG Split — The batter's slugging percentage at home (HM) or away (AW) this season, depending on where today's game is played. Some hitters perform significantly better in their home park due to familiarity with dimensions and conditions. Requires 10+ PA to display.
IBB
Intentional Walks (last 14 days) — Number of intentional walks over the past 14 days relative to total plate appearances. Shown in amber when non-zero. A high IBB rate signals that opposing teams are actively avoiding this hitter — a badge of respect for their power, though it also means fewer opportunities to swing.
H2H
Head-to-Head — Career stats for this batter vs this specific pitcher. Shows HR count and plate appearances (e.g. 2HR/15PA). Requires 5+ PA to display. A strong H2H record adds confidence; 0HR over many PA is a mild negative signal. Small samples should be interpreted cautiously.
Pitcher Metrics
P.HR%
Pitcher HR Rate — The pitcher's home run rate per plate appearance this season (displayed as HR/PA%). Higher values indicate a more HR-prone pitcher. League average is ~3%. Sourced from MLB Stats API (season stats) or historical Statcast cache. A key model feature — the third strongest predictor after barrel rate and exit velocity.
P.BB%
Pitcher Barrel Rate Allowed — Percentage of batted balls against this pitcher classified as barrels. A high P.BB% indicates a pitcher who struggles to prevent hard, well-struck contact. Sourced from Statcast historical data (2023–present). League average: ~6.5%.
P.HH%
Pitcher Hard Hit Rate Allowed — Percentage of batted balls against this pitcher hit at 95+ mph. A high P.HH% indicates a pitcher who struggles to generate weak contact. League average: ~36%. Higher = more favorable for the batter.
Zone & Matchup
Matchup Score
Zone + Pitch Type Matchup Score (0–100) — A composite score measuring how dangerous today's pitcher matchup is for this batter, based on two components:

Zone Score (70% weight) — How often the pitcher throws into this batter's HR hot zones in the strike zone. Computed from Statcast pitch locations (2025–2026). High = pitcher works into power zones.

Pitch Type Score (30% weight) — Overlap between the batter's HR pitch type profile (what pitches they hit for HRs) and the pitcher's pitch mix. E.g. if a batter hits 50% of HRs off fastballs and the pitcher throws 40% fastballs, that's high overlap.

Scores 65+ show “⚡ HOT MATCH”, 35–64 show “△ WATCH”.
Batter Profile
Radar Chart — Six-axis visualization of the batter's Statcast profile vs league average (shown as a faint blue polygon). Axes: Exit Velocity, Barrel Rate, HHFB%, Hard Hit%, Fly Ball Rate, Launch Angle. The orange polygon represents the batter; larger area = more elite overall power profile. Immediately shows whether a batter is a pure power hitter, a contact/elevation specialist, or somewhere in between.
Badge System
⚡ HOT ZONE
Hot Zone Match — The pitcher's combined zone + pitch type matchup score is 70 or above. Indicates the pitcher frequently throws into this batter's HR hot zones at a high rate. The strongest zone signal — red pulsing badge on pick cards.
🎯 BULLSEYE
Peak Zone Alignment — The batter's single highest HR-producing zone in the strike zone matches the pitcher's single highest HR-allowed zone. Both peaks must have a normalized score of 50+ to qualify. Shown as green crosshairs on both heatmaps in the zone modal.
🎯 TRIPLE
Bullseye + Pitcher Frequency — All conditions for BULLSEYE are met, plus the pitcher also throws to that same zone at 50%+ of their maximum zone frequency. The strongest individual zone signal: batter hits HRs there, pitcher allows HRs there, and pitcher actively throws there. Green glowing badge on pick cards.
🚀 HITTER'S PARK
Hitter's Park — The game is being played at a venue with a park factor of 1.10 or above, meaning the park meaningfully boosts HR probability. Examples: Coors Field (1.35), Great American Ball Park (1.22), Fenway Park (1.15). The model applies this as an additive boost to the base probability. Orange rocket badge.
☁ FB PITCHER
Fly Ball Pitcher — The opposing pitcher has a fly ball rate of 32%+ on balls in play. Fly ball pitchers surrender more HRs than ground ball pitchers since elevated contact more frequently leaves the yard. A mild positive signal for all batters in the matchup. Blue badge.
⛰ GB PITCHER
Ground Ball Pitcher — The opposing pitcher has a ground ball rate of 50%+. Ground ball pitchers suppress HRs by keeping the ball down. A mild negative signal. Brown badge. The model applies a probability cap for extreme GB pitchers.
★ CONSENSUS
Consensus Pick — This player appears in 3 or more external HR prediction sources (Covers, RotoWire, SBR, SI.com, etc.) scraped daily. A green star badge indicates broad agreement between the model and external analysts. Grey star = 2 sources.
🔥 HOT FORM
Hot Form — The batter's recent 14-day barrel rate is significantly above their career rolling average (4%+ above). Indicates the batter is in an elevated contact quality streak. Amber fire badge.
❄ COLD FORM
Cold Form — The batter's recent 14-day barrel rate is significantly below their career rolling average (4%+ below), or they have hit 0 HRs in 10+ consecutive games (HR drought penalty). Blue snowflake badge. The model applies a score reduction for extended droughts.
❄ STREAK
Miss Streak — Number of consecutive days this player appeared in the top 5 picks without hitting a HR. Shown as a cold badge on pick cards. Resets when the player hits a HR or leaves the top 5. Used as a caution signal for extended cold streaks in the top 5.
CONF: HIGH
AI Confidence Tier — Groq AI rates each top 5 pick as High, Medium, or Low confidence based on how many independent signals align beyond the raw model score. High = 4+ strong signals (EV, barrel%, form, park factor, pitcher vulnerability all pointing the same direction). Medium = 2–3 signals. Low = 1 signal or conflicting signals. Shown as a small badge on the player name.
A/B/C/D GRADE
Pitcher Matchup Grade — Groq AI assigns a letter grade to each pitcher matchup based on how favorable it is for hitting a home run today.

A — Very favorable: hittable pitcher, elevated HR rate, weak contact suppression.
B — Favorable: above-average HR vulnerability, some exploitable tendencies.
C — Neutral: average HR risk, no strong edge either way.
D — Tough: low HR rate, strong contact suppression, difficult matchup.

Grade appears as a small letter badge next to the pitcher name on each pick card. Based on pitcher HR rate, barrel% allowed, hard hit% allowed, and park factor.
Model & Scoring
Score
Power Score (1–100) — The model's HR probability mapped to a 1–100 scale for display purposes. Not a percentage — it's a relative ranking tool. Higher = model rates this batter more likely to hit a HR today vs the field.
🔥 ❄
Form Flags — Based on recent 14-day Statcast data vs career rolling average. 🔥 Hot: recent barrel rate or HHFB% significantly above career average. ❄ Cold: recent metrics significantly below career average. The model blends career (60%) and recent (40%) stats, so form flags indicate a meaningful current divergence.
TOP 5
Today's Top 5 Picks — The five highest-ranked candidates selected by the model, subject to a diversity rule: maximum 2 picks from the same game matchup. This prevents the model from stacking an entire lineup facing a single bad pitcher. Accuracy tracking applies only to these 5 picks.
FAQ
Why do the same players (Judge, Ohtani, Schwarber) appear most days?
The model ranks by HR probability which is driven heavily by barrel rate and exit velocity — both very stable metrics for established power hitters. Players like Judge and Ohtani have elite profiles that genuinely score highest on most days. The model isn't wrong to pick them frequently — they really do have the highest single-game HR probability. The recent form blend and pitcher matchup features create day-to-day variation, and as 2026 season data accumulates through monthly rebuilds, current-year performance will further differentiate the rankings.
What does the accuracy % actually measure?
Accuracy = HIT picks / (HIT + MISS picks). A perfect day is 5/5 (100%). A typical day with 1 HR hitter in the top 5 is 1/5 (20%). MLB league-average HR rate per plate appearance is ~3–4%, meaning a random player in any given game has roughly a 3% chance. If our top 5 hits at even 20–30% over a full season, the model is meaningfully outperforming random selection.
Why does the pitcher show n/a for P.HR and P.BB?
Pitcher stats require a minimum of 20 innings pitched in the current season before we trust them. Early in the season (April/May), most starters have only 2–4 starts and tiny samples. A pitcher with 2 bad outings could show a misleadingly high HR rate. When below 20 IP, the model uses the pitcher's historical rolling average from the training data internally, but we don't display it to avoid false precision. Values will populate as the season progresses.
When are picks generated and when are results checked?
11:00 AM ET — picks are generated. Most MLB lineups are posted by this time. For games where lineups aren't posted yet, the model falls back to the active roster.

9:30 AM ET (next morning) — results are checked against final MLB boxscores. Running after midnight ensures all West Coast games are complete. The dashboard updates automatically with HIT/MISS/DNP badges.
How is the model trained?
A logistic regression is trained on ~100,000+ batter-game rows from the 2023–2026 seasons. Each row represents one batter in one game with features: rolling EV, launch angle, barrel%, hard-hit%, opposing pitcher barrel%/HH%/HR rate, park factor, and platoon (batter hand vs pitcher hand). The label is 1 if the batter homered that game, 0 if not. The model is isotonic-calibrated so predicted probabilities match observed HR rates. It is rebuilt monthly to incorporate current-season data.
Is this for betting purposes?
No. Launch Lab is a personal analytics project built for fun and to explore Statcast-based ML modeling. It is not financial advice, not a betting service, and should not be used as the basis for any wagering decisions. Past model accuracy does not predict future results. Baseball is chaotic — even the best hitter in the best matchup misses far more often than they homer.
Model Calibration

How often do picks at each power score range actually result in a home run? A well-calibrated model should show higher hit rates for higher scores.

Loading calibration data...

DISCLAIMER — Launch Lab is a personal analytics project for entertainment purposes only. All data sourced from MLB Statcast via Baseball Savant and the MLB Stats API. Powered by Northern Edge Analytics · Not affiliated with MLB or any team. Not financial or betting advice. For entertainment only.
Consensus Picks
Loading consensus picks...
Source Accuracy
Accuracy tracked after results are confirmed.
🎲 Model Conviction Parlay
ⓘ Legs selected by model conviction — the 3 players our model rates highest for today. Odds sourced from major sportsbooks. VS Market shows whether our model is higher (+) or lower (−) than the market-implied probability. Not financial advice.
Alternate Parlays