What You're Actually Dealing With When You Run Analytics On A Jays-Phillies Matchup
I spent three seasons building and maintaining a custom scouting database that pulled together everything from Statcast tracking data to pitch f/x coordinates for matchups like Toronto against Philadelphia. The first thing you need to understand is that there isn't a single tool or dashboard that will give you clean, reliable answers on its own. You have to pull from multiple sources, normalize the data yourself, and figure out which metrics actually predict what you're trying to measure. I learned this the hard way when a front office asked me to produce a pre-game report for a playoff series and my output was completely wrong because I hadn't accounted for park factors in the strike zone dimension. Here is how I actually approach this. Start by pulling the relevant data sources. For MLB, that means Statcast, which is publicly available through the Baseball Savant API or scraped from the site itself if you are not working with an institutional account. You also want PitchFX data for older seasons since Statcast coverage became consistent starting in 2015. For a two-team comparison, I typically focus on the last two full seasons plus any playoff games those teams played. That gives you a sample size that is large enough to be meaningful but recent enough to reflect current roster construction. The core metrics you should be building toward are weighted runs created plus, or wRC+, for offensive evaluation, and fielding independent pitching, or FIP, for arm evaluation. These strip out luck and park effects better than raw batting average or earned run average. What most people miss is that neither metric alone tells you what matters in a head-to-head matchup. You need the platoon splits. A left-handed hitter against a left-handed pitcher tells a completely different story than the overall wRC+ number would suggest. The same goes for the Phillies and their split performance against right-handed starters.
Building Your Dataset The Practical Way
I used to do this with SQL queries against a hosted PostgreSQL database that I kept on a remote server. The workflow looked like this. First, I downloaded the season-level Statcast data files, which are CSVs available from the MLB Advanced Media open data repository. Then I wrote a Python script using pandas to merge the batter and pitcher rows against game logs, extracting venue, date, and opponent. From there I calculated the advanced metrics myself rather than trusting precomputed versions because the formulas are not always applied consistently across different platforms. One thing that trips people up is the xwOBA metric. Expected weighted on-base average is supposed to remove the luck component from batted ball results by using launch angle and exit velocity to predict what should happen on each ball in play. The problem is that it does not account for pitcher spin rate and movement dimensions very well. I found that when I cross-referenced xwOBA against actual outcomes for Jay and Phillies batters facing top-tier pitching, the correlation dropped significantly in small samples. That is why I layer in strikeout rate and walk rate directly into my models instead of relying on xwOBA as the primary indicator.
A Real Problem I Faced And How I Worked Around It
About two years ago I was asked to produce a report on Bo Bichette and Alec Bohm for an upcoming series, and the data was contradictory. Bohm had an elite contact rate against fastballs inside, but the pitch sequencing data from that season showed he was pulling up against inside heat far more often than the raw swing percentages indicated. I dug into the spray charts and tracked data and realized the issue was a shift in his stance during the season. He moved his front foot closer to the plate in May and June, which changed his launch profile. The Statcast data does not flag stance changes, so anyone building a model from raw CSVs will miss that unless they are actually watching game film to validate what the numbers are saying. My workaround was to pull frame-by-frame tracking data from the second half of the season and compare it against the first half using a simple angle calculation on the batter's box position. I wrote a script that computed the average distance from the plate for each hitter across multiple at-bat clips and flagged any shift greater than two inches. That was small enough to be easily overlooked but large enough to shift expected production by about eight percent in my projections. I ended up dropping Bohm's projected wRC+ by roughly ten runs and adjusting the Phillies lineup depth accordingly, which turned out to be accurate when the series actually happened.
Get the Full Details
The Metrics That Actually Move The Needle In This Type Of Analysis
For pitchers, I look at four-seam fastball spin efficiency and how it correlates with swing and miss rate. The Phillies have historically been strong in this area with their current rotation, and the Jays have struggled against high-spin fastballs in recent seasons. If you only look at velocity, you are missing half the picture. A ninety-three mile per hour fastball with high spin efficiency produces significantly more vertical break than a nine-five with low efficiency, and that difference is measurable in the data. On the hitting side, barrel rate and hard hit percentage are useful but only when broken down by pitch type and location. A batter can have a solid overall hard hit rate while actually ranking below league average against off-speed pitches in the zone. I separate pitch classification into four categories: fastball, slider, changeup, and curveball. Then I cross-reference each batter's performance against each pitch type with the pitcher's arsenal. This gives you a much clearer picture of which matchups are favorable and which are not.
What This Approach Cannot Do For You
The biggest limitation is that no amount of data will tell you how a team will perform on a given night if injuries, fatigue, or recent workload changes the situation. Pitch counts from the previous start matter. Bullpen usage in the three days leading up to the game matters. None of the public datasets include real-time health information or the exact number of pitches thrown in recent appearances unless you are pulling from proprietary sources. I have seen analysts build elaborate models and still come out completely wrong because they treated the data as static when it was actually time-sensitive. Another blind spot is umpire call zones. Statcast does not include umpire strike zone data in its public release. If a team is playing under an umpire with a notably tighter or looser zone, your pitch location models become less reliable. I worked around this by looking at home plate zone dimensions from third-party sources, but the coverage is inconsistent and not updated in real time. The only truly reliable way to account for this is to watch the games and note the zone patterns yourself, which defeats the purpose of using big data in the first place. What this work is really about is understanding that big data gives you signals, not answers. The Toronto Blue Jays versus the Philadelphia Phillies matchup involves hundreds of variables, and the ones that matter most change from game to game. The tools exist. The data is available. The hard part is knowing which pieces to trust and which pieces to discard before you make a decision based on them.
