Where does the data come from?

BeaverStats doesn't pull in numbers someone else already calculated - we calculate everything ourselves, starting from the smallest unit of data (every single pitch and plate appearance).

NPB & KBO data quirks

1. We haven't found a source that gives us exit velocity or launch angle for NPB/KBO, so we can't calculate HardHit% or Barrel% for those leagues.

2. We haven't found a source that publishes projected lineups ahead of time for NPB/KBO. Until a lineup is officially confirmed, what we show is always the previous game's lineup. We tried heuristics like "this batter tends to start against a right-handed starter," but testing showed simply reusing the previous game's lineup matched reality better, so that's what we use now (a more refined approach could change this down the line)(→ we've since found a better calculation, see #4 below). So it can be worth double-checking the data again once the lineup is actually confirmed.

3. For NPB/KBO it can be hard to tell which side of the plate a switch-hitter batted from. This can leave a pitcher's vs-LHB/vs-RHB batting-average-against with a smaller sample (fewer innings) than it should have. Since switch-hitters are relatively rare in both leagues, we chose not to guess/map this arbitrarily for now - we plan to dig into it further.

4. We ran an experiment to replace the "reuse the previous game's lineup" approach mentioned in #2 above with a better calculation. Compared against each game's actual starting lineup, our current "reuse the previous game" approach gets an average of 6.4 of 9 starters right for NPB (71.0%) and 7.3 of 9 for KBO (81.1%). A new model we built (combining recent 5/10-game start frequency, season-long start rate, and recent-game performance) pushes that up to 6.9 of 9 for NPB (76.3%) and 7.6 of 9 for KBO (84.1%). It's not perfect yet, but it clearly beats reusing the previous game's lineup, so once we've validated it further we plan to replace the current lineup display with this calculation.

Weather data

We currently pull weather data for free via the Open-Meteo API. Forecasts for each stadium's location/time are refreshed every 10 minutes, and starting 1 hour before first pitch we switch to real observed data, still refreshed every 10 minutes. Once the game starts, we stop updating the weather.

We show wind direction for each stadium as an arrow.
NPB: official-scorer reference data
MLB & KBO: bearing calculated from satellite imagery

Data aggregation

Once all of a day's games have finished, stats are usually aggregated and reflected within 30 minutes to an hour. The record-ranking screens update at the same time.

Display Conventions

Good values are shown in green, bad values in red.

The away team is shown on the left, the home team on the right, by default.

In tables that compare away and home side by side, the better value is shown in bold green and the worse one in bold red.

The starters/lineup total row also shades its background to show how far recent form has strayed from the season average - a deeper green means a bigger improvement over the season, a deeper red means a bigger decline.

In lineup tables, entries based on fewer than 20 plate appearances are shown in a lighter font since the sample is too small to be reliable.

Park Factor

Park Factor = (Avg. runs at home / Avg. runs on road) x 100

We're a new site and currently only have complete data for the 2026 season, so our park factor numbers may differ from other sites. We also apply a regression that pulls extreme values back toward 100, and since there's no single standardized formula for that regression, results can vary slightly between sites (typically by a fraction of a point). Right now, park factor feeds into exactly one stat on this site: wRC+.

wOBA weight formula

NPB & KBO: we use Tom Tango's universal fixed weights as-is (there's no NPB/KBO-specific published constant). wOBA = (0.69×BB + 0.722×HBP + 0.888×1B + 1.271×2B + 1.616×3B + 2.101×HR) / denominator

MLB: we use the weights FanGraphs' Guts! page publishes each season (2026 season shown here). wOBA = (0.698×BB + 0.729×HBP + 0.889×1B + 1.258×2B + 1.591×3B + 2.040×HR) / denominator

Data We're Looking For

If you happen to know a free source for any of the data below, please let us know!

1. A place to watch full NPB game video ourselves (2026 regular season onward)

2. A source for projected/expected lineups for NPB and KBO

3. A source for per-batted-ball exit velocity and launch angle for NPB and KBO

You can reach us through the contact info at the bottom of the site - we'd really appreciate it!

Where does the data come from?

Data aggregation

Once all of a day's games have finished, stats are usually aggregated and reflected within 30 minutes to an hour. The record-ranking screens update at the same time.

Display Conventions

Good values are shown in green, bad values in red.

The away team is shown on the left, the home team on the right, by default.

In tables that compare away and home side by side, the better value is shown in bold green and the worse one in bold red.

The starters/lineup total row also shades its background to show how far recent form has strayed from the season average - a deeper green means a bigger improvement over the season, a deeper red means a bigger decline.

How we calculate possessions

KBL: we use the real possession counts the official KBL website provides directly, per player and per team - no estimation needed.

NBA: none of our NBA sources provide possession counts directly, so we use Dean Oliver's possession estimate formula, the standard used across basketball analytics. Possessions ≈ 0.5 × [(team FGA + 0.4×team FTA − 1.07×(team OREB÷(team OREB+opponent DREB))×(team FGA−team FGM) + team TOV) + (the same calculation for the opponent, added in)]

How we calculate PER

KBL: we use the PER value the official KBL website provides directly for each player.

NBA: since no official PER is available, we implemented John Hollinger's uPER formula ourselves. League- and team-wide pace/efficiency constants are precomputed once per season (recalculating them on every request would be too slow), then each player's own stats are plugged into the formula. "Blended" values that combine multiple players (starters/bench) are a minutes-weighted average of each player's PER.

How we calculate ORtg (Offensive Rating)

NBA: we implement the full, official Dean Oliver Individual Offensive Rating formula (the entire chain, including assist credit) to precisely attribute scoring value per possession.

KBL: the full formula above can't be reliably reproduced from KBL's raw data alone, so we use a simplified formula instead (100×points÷(FGA+0.44×FTA+TOV)). Because of this, ORtg values on the NBA and KBL screens are calculated differently - please don't compare them directly across leagues.

How we calculate DRtg (Defensive Rating)

KBL: we use the team-level value the official KBL website provides directly.

NBA: we calculate it ourselves as 100×opponent points scored÷(the possession estimate above).

How we calculate the rest (USG%, TS%, eFG%, AST%, TOV%, REB%, STL+BLK%)

For both KBL and NBA, we calculate these ourselves from each source's box score (field goals, free throws, rebounds, assists, steals, blocks, turnovers) using the standard Dean Oliver family of formulas.