Methodology
How KSplit models strikeout distributions: the inputs, the structure, and what each output actually represents.
KSplit models MLB pitcher strikeouts because strikeouts are one of the few betting markets built on a fully objective outcome. A hitter either strikes out or he doesn't. There is no scorer interpretation, no judgment calls, and no postgame stat corrections.
Strikeouts are modeled at the plate appearance level, not the inning level. KSplit combines pitcher strikeout tendencies, opponent lineup composition, handedness splits, and an independent estimate of expected batters faced to separate skill from opportunity.
The output is a median strikeout projection and a full probability distribution. This is not a point projection system. Users are encouraged to look at the entire distribution to make a complete assessment of the matchup.
For common questions about access, usage, and scope, see the FAQ. For a column-by-column walkthrough of the live board, read our How to Guide for Today's Dashboard. For formal definitions of every metric, open the Glossary.
Core Modeling Approach
Plate Appearances, Not Innings
Strikeouts occur at the plate-appearance level. Modeling strikeouts as K/PA avoids distortions caused by inning length, defense, and game flow. Instead of asking how many innings a pitcher will throw, KSplit asks:
- How often does this pitcher strike out hitters?
- Against this lineup, in this context?
- And how many opportunities are likely to occur?
Pitcher vs. Lineup
Strikeout outcomes depend on both the pitcher and the opposing hitters. KSplit blends:
- Pitcher K rates vs. left- and right-handed hitters
- Opponent lineup K tendencies vs. pitcher handedness
- Actual lineup composition, not team averages
Confirmed lineups are incorporated when available.
Modeling Opportunity
Strikeout rate alone does not determine totals. Opportunity matters. KSplit models expected batters faced (BF) independently to reflect workload, leash, and game context. This prevents overstating high-K pitchers with limited usage and understating pitchers expected to face deeper lineups.
Median-Anchored Projections
The primary projection shown on KSplit is the median strikeout outcome. The median represents the most typical result. Not a ceiling, not a floor, and not a best-case scenario.
It is intentionally not adjusted for perfect game flow, maximum pitch counts, or ideal sequencing. This prevents optimistic bias in the core projection.
Pricing the Entire Distribution
KSplit models the entire strikeout outcome distribution, not a single number. High-strikeout performances occur when multiple favorable factors align: workload, efficiency, matchup leverage, in-game adjustments. These outcomes live in the right tail of the distribution.
Rather than forcing ceiling outcomes into the median projection, KSplit captures them probabilistically. Extreme strikeout games may exceed the median projection while still being correctly identified as upside.
Distribution quality is evaluated using proper scoring rules (CRPS), which score how well the full probability curve aligns with real outcomes. This ensures tails aren't exaggerated and that uncertainty is calibrated.
Distribution Shape & Conversion Context
Distribution Shape Score (DSS)
DSS measures how much structural upside exists in the modeled strikeout distribution. It evaluates the right tail only and does not measure probability, confidence, or consistency.
Ceiling Conversion Context (CCC)
CCC measures whether the game environment allows upside to convert. CCC is intentionally non-monotonic: very low values rarely convert upside, middle values convert ceiling outcomes most often, and extremely high values are often stable.
CCC is a contextual filter, not a probability.
Ceiling Profile
KSplit blends DSS and CCC into a single diagnostic called Ceiling Profile, which describes how strikeout upside is expected to behave.
Stable environments can cap otherwise strong upside; volatile environments can suppress conversion. Ceiling Profile is designed for decision support, not pitcher ranking.
Probabilities & Market Context
KSplit derives probabilities directly from the modeled distribution. Market odds are converted to implied probabilities and de-vigged for fair comparison. Differences between model probabilities and de-vigged market probabilities highlight areas of disagreement.
Best Edge reflects where market pricing deviates most from the full distribution, on either the over or the under. A pitcher can display a High | Tail-Driven profile even when the strongest edge is priced on the under.
Market Validation: Closing Line Value (CLV)
KSplit tracks Closing Line Value as a market-based validation signal. CLV measures how the market moves from open to close relative to the prices and lines captured, which helps confirm whether distributions are aligning with the same information the market is absorbing.
CLV is used to evaluate process quality and market alignment, not to present picks or ROI. It is available in Rolling Diagnostics alongside calibration metrics like CRPS and tail accuracy.
Transparency
All projections are logged and archived. No results are removed or selectively displayed.
- Median strikeout projections
- Distribution-based probabilities
- Error metrics tracked over time
What KSplit Is, and Is Not
KSplit is
- A strikeout-focused analytics platform
- A probabilistic projection engine
- A transparency-first data product
KSplit is not
- A picks or "locks" platform
- A guarantee of outcomes
All outputs are informational and intended for analytical and educational use only.