Methodology

Estimated season value

pWAR and pWAR+ — Pennant Wins Above Replacement

One estimate of the wins a player contributed above replacement in the recorded season. It is PRR’s model, not Baseball-Reference WAR or FanGraphs WAR. Both pWAR and pWAR+ are estimates, including their compact summary displays.

pWAR formula

\mathrm{pWAR}=\frac{R_{\mathrm{bat}}+R_{\mathrm{steal}}+R_{\mathrm{other\ running}}+R_{\mathrm{field}}+R_{\mathrm{pitch}}+R_{\mathrm{position}}+R_{\mathrm{league}}+R_{\mathrm{park}}+R_{\mathrm{team\ field}}+R_{\mathrm{replacement}}}{\text{runs per win}}

\text{runs per win}=3+1.5\times\text{league runs per nine innings}

\text{caught-stealing cost}=2\times\text{league runs per out}+0.075\text{ runs}

The run components are batting, stealing, other baserunning, fielding, pitching, position, league, park, team fielding and replacement. Batting and pitching contributions are combined once for two-way players; both role cards show that same total. Do not add their displayed totals together.

Batting uses marginal BaseRuns weights calibrated to the same season’s scoring environment, not an OPS multiplier or card rating. Pitching starts with charged runs prevented versus that season’s average, subtracts the fielding credit allocated to the pitching staff, and includes the pitcher’s own fielding once. Steals gain 0.2 runs; caught-stealing cost follows the formula above, with an opportunity-based league adjustment.

Position value uses the observed position mix and batting workload. Catcher, shortstop and first base therefore have different positional baselines; selectable defensive-ladder positions do not change the historical estimate. Single-year venue scoring ratios are shrunk toward average.

Main pWAR and pWAR+ do not apply leverage. A separately labeled complete-entry deployment estimate multiplies only relief pitching pWAR by (1 + average entry leverage) / 2, then adds the difference to the unchanged total. It requires the entire verified relief appearance history, including zero-out outings, and retains the fixed leverage reference. Incomplete histories have no estimate. This half-chaining sensitivity follows the published FanGraphs approach but precedes any league-wide correction; it is not fWAR. Scoring-time importance and weighted outs do not substitute for entry leverage.

The replacement convention allocates 1,000 wins per 2,430 games: 57% through batting opportunities and 43% through pitching outs. Values accumulate with actual playing time; they are not full-season projections. Regular seasons include recorded tiebreaker games. Postseason values stay separate and use the same year’s regular-season run/replacement reference. When a sparse historical competition has no retained regular reference, its postseason-only estimate is explicitly labeled; it is not a regular-season comparison.

Reliever chaining reference ↗

pWAR+ — contribution relative to role regulars

pWAR+ formula

\mathrm{pWAR}^{+}=\sum_r100\frac{\mathrm{pWAR}_r}{B_r},\qquad B_r=\frac{\sum_{j\in Q_r}\mathrm{pWAR}_{j,r}}{|Q_r|}

G_{\mathrm{team}}=\frac{2G_{\mathrm{competition}}}{N_{\mathrm{teams}}},\quad \mathrm{PA}_{\min}=\operatorname{round}(3.1G),\quad \mathrm{SP\ outs}_{\min}=\operatorname{round}(3G),\quad \mathrm{RP\ outs}_{\min}=\operatorname{round}(0.9G)

Each role contribution is compared with the mean contribution of qualifying regulars in that same season and competition. Batting uses the player’s primary season position as a comparison class; it does not claim to allocate batting runs to each fielding position. Pitching uses exact starting and relief appearance accounts. Incidental pitcher batting belongs to the primary pitching class, while meaningful batting and pitching contributions add once.

Within one role, 100 means the average qualifying contribution and zero role pWAR maps to 0. An average full-season starter and reliever can each contribute 100 while adding different absolute pWAR. Multi-role totals add their role indexes and do not share those single-role anchors. Use absolute pWAR to compare estimated wins across roles. Playing time stays in the contribution, rather than being divided out. Negative values stay negative.

The reference workload floors are 3.1 batting PA, 1 starting IP or 0.4 relief IP per average retained team game, with half-up rounding in PA or outs. These are explicit PRR comparison policies, not official qualification or a claim of statistical reliability. Sparse reference means and current-season comparisons remain provisional; the displayed leaderboard filter does not redefine them. A missing role benchmark leaves the complete index unavailable.

Postseason uses the same year’s regular-season role benchmarks. A postseason-only estimate without those references has no index. Career pWAR+ is the arithmetic mean of complete annual role-relative indexes, counting each retained year once. This keeps a position switch from doubling the career index; concurrent two-way contributions still add within each year. Any incomplete annual index leaves career pWAR+ unavailable. Career pWAR remains the sum of annual wins.

The 100-average naming convention follows familiar baseball “+” indexes ↗; this contribution formula is PRR’s own definition, not OPS+, an industry-standard WAR+, a percentile, talent rate or competitive price.

Estimated does not mean fully measured. This first version estimates fielding from errors relative to position-specific chances. Unmeasured range, arm, receiving and non-steal baserunning use explicit neutral priors. Missing counts use available same-population rates; absent caught-stealing support uses a 70% success prior. Missing venue data uses a neutral adjustment assumption. These assumptions can materially affect individual rankings, especially defense-first players.

Historical, Negro League and minor-league estimates remain relative to their own retained competition. Sparse seasons represent recorded opportunities, not invented missing games, and a minor-league pWAR is not an MLB-equivalent talent estimate. Current pWAR follows the official season-total snapshot, which can be older than the daily pitch-data cutoff.

The component arithmetic, workload scaling and replacement accounting are tested. This is not an externally calibrated replacement for bWAR/fWAR, has no calibrated confidence interval, and does not change simulation ratings, the Best marker, or competitive prices.

Background: Tango’s environment-specific linear weights ↗ and WAR’s run-accounting framework ↗.

Playing time

Comparing equal workloads

Workload rates put counting statistics and accumulated value on a common amount of playing time. They describe the recorded sample; they do not project a full season.

Batting rates use 600 plate appearances. Pitching value and selected events use 100 innings; defensive counts and value use 1,000 innings at the matching position. Baserunning value uses 100 opportunities of the type measured. Each rate requires its numerator and exposure to cover the same season, phase, role and source window.

Innings come from recorded outs: three outs equal one inning. Career rates divide the sum of matched contributions by the sum of their matching exposures, rather than averaging season rates. Missing exposure leaves a rate unavailable; a recorded zero remains zero.

The existing measured FRV per 1,000 outs keeps its original scale. It is distinct from FRV per 1,000 innings. Qualification minimums are separate: halving the minimum innings for relievers never halves a rate’s denominator. These rates do not establish reliability, equal opportunity difficulty or a new simulation input.

Pitching outcomes

FIP and SIERA

These estimates summarize selected pitching outcomes on an earned-run-average scale. Lower is better; neither replaces recorded ERA.

FIP uses home runs, total walks, hit batters and strikeouts per inning. Its constant aligns the combined reference with ERA for the same season, competition and recorded window. A constant from another season is never substituted.

SIERA uses strikeouts, walks and batted-ball types per batter faced, including their interactions. The displayed version is the original 2010 formula, labeled as a retained-data estimate. It is distinct from the later FanGraphs formula and its annual adjustments.

Missing or unreconciled inputs leave the affected estimate unavailable. The original formula’s treatment of bunts remains unresolved in this implementation, so a pitcher with recorded bunts has no SIERA estimate. Annual estimates are not averaged into a career SIERA.

References: FIP definition and original SIERA formula.

Fitted pitch geometry

Pitch tunneling

This view asks how close a pitcher’s average pitch-type paths travel late in flight. It describes reconstructed arsenal geometry, not an actual pitch sequence or what a batter perceived.

For MLB regular seasons from 2008 onward, each qualifying pitch is reconstructed from the source’s 3D release position, velocity and total acceleration. We average those fitted paths at matching fractions of release-to-plate flight by pitcher, season and pitch type, then draw each mean path. Chart times are approximate mean-flight guides. The chart can show depth differences and distance at the same field depth separately; neither replaces the timed score.

The source’s total acceleration already includes gravity and movement; we do not add either again. Unmeasured arm angle stays unavailable. Source fields: Baseball Savant tracking documentation ↗.

Tunneling, in inches

For each pair of competitive pitch types with fitted pitches, average their horizontal and vertical path separation at 150, 175, 200, 225, 250, 275 and 300 milliseconds before each pitch’s own plate arrival. A type’s score averages its gaps to every other competitive type with fitted pitches, weighted by those other types’ fitted pitch counts; the arsenal score averages type scores by fitted competitive usage. Lower inches mean the average fitted paths stayed closer during the measured window. A pitch type is compared with every other competitive type with fitted pitches, not only its nearest partner. With exactly two competitive types with fitted pitches, their type scores are equal by definition.

For scoring, each checkpoint is measured before each individual pitch’s own plate arrival, then averaged by pitch type. The scored distance uses horizontal and vertical coordinates; depth gap and separation at the same field depth are separate descriptions.

Tunneling+ comparison

Tunneling+ = 150 − 100 × (ascending midrank of raw Tunneling inches − 0.5) / number of eligible pitchers. 100 is the eligible cohort mean; higher means closer average paths. It is a rank index, not a percentage improvement, a measure of pitch effectiveness or a validated skill.

A coverage-supported type needs at least 30 qualified pitches across three games, at least 80% of its source pitches retained, and a retained-versus-source balance difference of no more than 0.10 for month, batter hand, count, plate location and pitch speed. Material source months or batter hands cannot disappear entirely. A coverage-supported arsenal needs at least two supported types representing 90% of source usage and 80% overall retention. Tunneling+ also requires 100 qualified pitches across five games and at least 30 eligible pitchers from that season and tracking regime. Each major type (at least 10% source usage) must be supported, with at least 90% of its other-type source usage supported, so a poorly sampled rare partner cannot dominate a major type’s comparison. Raw scores and partner comparisons include every competitive type with fitted pitches, including thin samples labeled limited. Intentional balls (IN) and pitchouts (PO) remain available as context paths and counts, but do not enter scores, scoring partners, intervals, qualification denominators or Tunneling+ cohorts. All usage and retention requirements here refer to competitive pitches. The same raw score underlies Tunneling+; coverage gates do not change its partner set. A single competitive-type arsenal has no comparison score, even when context paths are available.

Playback and camera views

Playback places every average path on one elapsed-time clock, starting at release. Each pitch type reaches the plate at its own approximate arrival time; afterward its marker stays at the plate and the readout shows time since arrival. This compares mean paths, not pitches thrown together or an observed sequence. Tunneling scores and partner curves still compare checkpoints before each pitch’s own arrival. Batter, catcher, third-base side, first-base side, pitcher and top are starting positions for one freely rotatable 3D orbit camera. The full field reference spans 60.5 feet from the back of home plate to the rubber; fitted paths still begin at their source-reported release points in front of the rubber. Activate the chart to rotate with left-drag, pan with right-drag and zoom with the scroll wheel. The Camera panel inside the chart provides keyboard alternatives. Camera adjustments do not change pitch geometry or timing. Pitch visibility and closest-path packs only filter the display; the full-arsenal clock and scores stay unchanged. Each pack contains a competitive pitch type and up to two of its closest available mean-path partners. Packs can overlap and do not represent validated clusters or observed pitch sequences. Persistent timeline labels mark each fitted type’s mean plate arrival even when its path is hidden. The selected pitch’s spin-axis mark uses its source circular mean across all counts and batter hands, shown as a fixed 2D catcher-view orientation with a separate source date and sample count. It is not a reconstructed 3D axis or gyrospin measurement; its sample can differ from the fitted paths. Missing source values remain unavailable.

Research ranges on the elapsed-time slider

The slider colors are overlapping research guides for the selected pitch’s fitted flight time, not measured neural stages for a hitter. Early flight cues mark the first 150 milliseconds after release. Continued visual information begins after that reference and fades near arrival to illustrate less time for a motor response; its usefulness depends on the task. Recognition and decision-making do not start or finish at these color boundaries.

Possible forward-swing activity overlaps the visual ranges. The reported onset range is about 130–280 milliseconds before impact, translated here to elapsed time using the selected pitch’s approximate plate arrival. The motor color builds across that onset range and continues to arrival; it does not mean every hitter starts at the same instant. Plate arrival is not measured bat–ball contact. Blended colors show overlapping guides; their intensity is illustrative, not a probability or measured brain activity.

Closer fitted paths do not establish deception, hitter difficulty, pitch effectiveness, a repeatable skill or a simulation effect. Tracking methods change across PITCHf/x, original Statcast and Hawk-Eye, so Tunneling+ keeps their comparison groups separate. No unmeasured path or pitch type is imputed.

Timing context: research on forward-swing timing ↗, Higuchi and colleagues’ visual occlusion study ↗ and research on batting’s overlapping visuomotor processes ↗. These studies inform the illustrative guides; they do not validate this path score as perceived tunneling.

Pitch outcomes

Run value and RV+

Run value starts with the change in expected batting-team runs after a pitch. The two RV+ branches answer different questions, so the role and definition always travel with the label.

Descriptive batter RV+

Higher is better

Compares the batter’s observed run value per selected pitch with the full exact context, then scales that difference by the scoring environment for that MLB season.

Descriptive batter RV+ formula

\mathrm{RV}^{+}_{\mathrm{bat}}=100+100\frac{r_{\mathrm{bat}}-r_{\mathrm{context}}}{r_{\mathrm{league}}}

$r_{\mathrm{bat}}$: Sum of pitch run value divided by selected delivered pitches for the batter
$r_{\mathrm{context}}$: Full result from the same competition, season, phase/cutoff, pitch, observed stance, opposing hand, and selected count/location when applicable
$r_{\mathrm{league}}$: Runs scored divided by delivered pitches in that same reviewed competition-season phase/cutoff

Read it as: 100 is the exact-context average; higher is better. It is unshrunk observed performance, not reliability or predicted future skill.

Not wRC+. wRC+ evaluates a batting line at the plate-appearance level and adjusts league and park. This RV+ stays pitch-decomposable and is not park-adjusted. Current evidence supports cross-season MLB normalization, not cross-league comparison without a separately established league scoring environment.

A displayed percentile, when available, is a separately qualified rank-equivalent raw RV/100 midrank. It does not turn this descriptive index into a reliability estimate.

Reference-membership gate: at least 250 eligible delivered pitches and 20 qualified batters in the exact reference cohort. A target below its own floor may be labeled provisional only when that exact cohort exists.

Terminal-PA wRC+ by location

Higher for batters · lower allowed for pitchers

The default location view assigns a plate appearance’s final result once to its last pitch’s location, pitch type and pre-pitch count. It supports exact observed stance and opposing-hand selections. Missing final locations stay outside the grid; intentional walks, sacrifice bunts and interference are excluded.

PRR wRC+ = 100 × [(selected wOBA − league wOBA) ÷ wOBA scale ÷ league runs per PA + 2 − selected park factor]. Run weights and the full-scope MLB run environment come from the same season. Empirical home-team park factors use a 10,000-PA prior and are centered to a PA-weighted league mean of one; each selected PA carries its recorded home-team park proxy. Neutral-site and within-season venue changes are not separately resolved.

The benchmark stays fixed across zones, counts and hands. These are unshrunk observed results with PA samples, not percentiles or predictions; small samples may be extreme or negative. The source cutoff belongs to this independent product. Pitch-by-pitch play continues to use its separate terminal-PA RV+ model.

FanGraphs explains the wRC+ framework ↗. These are PRR estimates with pooled MLB constants, not imported FanGraphs values or its separate AL/NL non-pitcher normalization.

Pitcher RV+

Lower is better

Measures offense-perspective run value allowed relative to pitchers in the same season, role, pitch, batter-hand, and selected count/location context.

Pitcher RV+ formula

\mathrm{RV}^{+}_{\mathrm{pit}}=100+100\rho_{\mathrm{RV}}\frac{r_{\mathrm{allowed}}-r_{\mathrm{context}}}{r_{\mathrm{league}}}

\rho_{\mathrm{RV}}=\frac{N}{N+K_{\mathrm{RV}}}

$K_{\mathrm{RV}}$ is fitted from same-season evidence for the relevant surface; it is not a universal constant. The reliability factor pulls small samples toward 100.

Read it as: below 100 means less offense allowed than the exact context after shrinkage. Because the underlying run value is from the offense’s perspective, lower is better for the pitcher.

Reference-membership gate: at least 50 delivered pitches with run value in the exact season × role × pitch × batter-hand cohort. An observed target below that floor may receive only a provisional percentile when the qualified exact cohort exists; an empty cohort stays unavailable.

The separate predictive batter RV+ experiment failed its chronological held-out-lift gate and is not the batter metric shown in the product.

Outcome RV / 100

Signed contribution

Ball, called-strike, swinging-strike, foul, ball-in-play, and hit-by-pitch rows are pieces of one total—not six independent quality statistics.

Outcome run value per 100 decomposition

C_k=100\frac{\sum_{i\in\text{outcome }k}\Delta\mathrm{RE}_i}{N}

\mathrm{RV}/100=\sum_k C_k

Every component uses the same eligible-pitch denominator $N$ , so the signed pieces reconcile to total RV/100 before display rounding. They explain where the total came from; they are intentionally not percentiled.

Quality on contact

Contact Damage+ and Contact Suppression+

The calibrated metrics use xwOBA on contact, with opposite-sign deltas for batter and pitcher perspectives and different opponent baselines. Each event’s expectation excludes that event to avoid comparing it with itself. Explicitly labeled era fallbacks are separate descriptive measures.

What xwOBA and xwOBAcon represent

xwOBA means expected weighted on-base average. wOBA weights outcomes by their typical run value rather than treating every way of reaching base alike. For each tracked batted ball, Statcast estimates the probabilities of an out, single, double, triple, or home run from comparable exit velocity and launch angle, then combines those probabilities on the wOBA scale. A full player xwOBA also incorporates actual walks, strikeouts, and hit-by-pitches.

xwOBAcon is xwOBA on contact. It averages those expected values only across eligible tracked batted balls. It excludes strikeouts, walks, hit-by-pitches, and plate appearances without eligible contact. Higher xwOBAcon therefore means the contact was expected to do more offensive damage; it is not a complete measure of the batter’s offense and it is not the batted ball’s actual result.

Contact Damage+

Higher is better

Asks whether a batter’s contact was more damaging than expected given the same-season opposing pitchers faced and the batter’s observed stance.

Contact Damage Plus formula

e_i=\frac{S_{\mathrm{opp,side}}-y_i+K_{\mathrm{opp}}\mu_{\mathrm{side},-i}}{n_{\mathrm{opp,side}}-1+K_{\mathrm{opp}}}

A=\operatorname{mean}(y_i),\qquad E=\operatorname{mean}(e_i)

\rho=\frac{N}{N+K_{\mathrm{bat}}}

\text{Contact Damage}^{+}=100\left(1+\frac{\rho(A-E)}{E}\right)

$y_i$ is the source-provided xwOBAcon value for event $i$ , and $e_i$ is its leave-one-event-out opponent expectation. Both prior strengths are fitted from the same-season population.

Read it as: 100 means actual contact damage matched opponent-adjusted expectation; above 100 means the batter exceeded it after shrinkage.

Overall season headline: Each eligible batted ball is counted once across all pitches, counts and handedness, retaining its event-specific opponent expectation. The batter shrinkage prior is fitted anew to these overall batter-season residuals. This is not an average of pitch-specific indexes or percentiles. Its cutoff is displayed separately from the official season stats.

Pitch and handedness breakdown: Selecting a pitch type, opposing pitcher hand or batting stance restricts the paired contact sample and fits the batter shrinkage prior for that exact selection. All-pitch and combined-hand selections aggregate the events themselves. The headline and its paired actual/expected values use the same sample; no overall index is substituted. This descriptive breakdown supports in-progress seasons at the displayed contact cutoff. If the selected population cannot identify a shrinkage prior, the index is unavailable. A supported index can still lack a percentile when fewer than 20 batters meet the reference requirements.

The overall descriptive percentile compares batters in that same season and cutoff, with at least 75 paired batted balls, shrinkage weight of .50 and 20 qualified reference batters. Smaller targets may receive a provisional percentile. The overall headline is an adjusted account of observed contact—not a validated forecast of talent; no confidence interval or simulation-rating use is implied.

Contact Suppression+

Higher is better

Asks whether contact allowed by a pitcher was less damaging than expected for the same-season batters and batting sides faced.

Contact Suppression Plus formula

A=\operatorname{mean}(\text{xwOBAcon allowed})

E=\operatorname{mean}(\text{leave-one-out batter expectation})

\Delta=E-A,\qquad\rho=\frac{N}{N+K_{\mathrm{pit}}}

\text{Contact Suppression}^{+}=100\left(1+\frac{\rho\Delta}{E}\right)

The expected value is fit from the batter × batting-side population with the evaluated event removed. The pitcher prior is fitted separately for summary and location-cell surfaces.

Read it as: 100 means contact allowed matched the opponents’ expectation; above 100 means the pitcher suppressed expected damage after shrinkage.

Era-aware descriptive fallbacks

Valid calibrated values stay preferred. The tooltip identifies fallback methods: “descriptive contact” is an unshrunk league-relative tracked-contact index; “recorded contact” instead uses observed contact outcomes and that season’s empirical run weights. Neither fallback is opponent-adjusted, and recorded contact is not xwOBA. RV+ fallbacks use known run values per pitch, or recorded plate appearances in older eras, relative to the exact same-season selection. Missing observations are excluded and sample coverage is retained. There is no fitted uncertainty; tiny samples can be extreme.

Era fallback contact indexes

\text{Batter damage}=100\frac{\text{actual contact value}}{\text{league contact value}}

\text{Pitcher suppression}=200-100\frac{\text{actual contact value}}{\text{league contact value}}

Batting Overview percentiles use same-season, same-opposing-hand batter cohorts, never mixed contact methods. All-pitcher references require 100 batting opportunities for OPS+/wRC+/RV+, or 75 contacts for contact metrics; hand splits require 25 respective opportunities. At least 20 reference batters are required. These are descriptive workload floors, not official batting-title qualification. Smaller targets are provisional; other pitch/cell fallbacks do not inherit these ranks.

The additional batter season headline follows opposing handedness, not the pitch selector. OPS+ and wRC+ are PRR estimates using retained season league evidence, empirical run weights and regressed park factors—not imported Baseball-Reference or FanGraphs values. xwOBAcon+ compares tracked contact with the matching pitcher-hand league mean. Missing tracking remains unavailable, even when a recorded-contact fallback exists.

Contact expectation index

\mathrm{xwOBAcon}^{+}=100\frac{\text{tracked xwOBAcon}}{\text{matching pitcher-hand league mean}}

Drilldown reference-membership gates: Pitch-specific Contact Damage+ requires at least 75 paired BBEs, shrinkage weight of at least .50, and 20 exact references; it ranks only final regular-season, nonblank-pitch, observed L/R stance × opposing L/R hand, all-count summaries. The separately fitted overall headline above does not relax those drilldown gates. Contact Suppression+ references require at least 20 BBEs and are all-count only in the exact same-season pitcher-role × pitch × batter-hand population. An observed target below its threshold may receive only a provisional percentile when the qualified exact cohort exists. Cells, count splits, merged drilldown axes, to-date drilldowns, empty cohorts, and missing calibration remain value-only or unavailable as their rules require.

Context performance

Clutch is performance above an expected context

The product carries two different Clutch ideas. Regressed Clutch Δ compares pressure performance with a context-adjusted expectation. SABR-style Clutch summarizes win-probability performance relative to leverage. Neither is an innate-clutch claim.

Regressed Clutch Δ

Higher is better

Pressure observations are weighted additively for runners in scoring position, two outs, and reaching two strikes. The historical Retrosheet branch also includes late-close appearances: inning seven or later with the score within two runs. Reliability counts each qualifying PA or BF only once.

Contextual Clutch formula

\begin{gathered}\text{Current batter/pitcher and historical batter/starter:}\\ \mathbb{E}[\mathrm{wOBA}]=\frac{\sum_c n_c\max(0,\mathrm{wOBA}_{\mathrm{player}}+\mathrm{wOBA}_{\mathrm{league},c}-\mathrm{wOBA}_{\mathrm{league}})}{\sum_c n_c}\end{gathered}

\mathbb{E}[\mathrm{wOBA}]_{\text{historical reliever}}=\frac{\sum_c n_c\mathrm{wOBA}_{\mathrm{league},c}}{\sum_c n_c}

\Delta_{\mathrm{raw,bat}}=\mathrm{wRC}^{+}_{\mathrm{pressure}}-\mathrm{wRC}^{+}_{\mathrm{expected}}

\Delta_{\mathrm{raw,pit}}=\mathrm{wRC}^{+}_{\mathrm{expected}}-\mathrm{wRC}^{+}_{\text{pressure allowed}}

\rho=\frac{N_{\text{unique pressure}}}{N_{\text{unique pressure}}+200}

\text{Clutch }\Delta=\rho\Delta_{\mathrm{raw}}

The exact context includes base/out state, two-strike status, batter and pitcher hand, inning band, and score margin. PRR wRC+ uses same-season empirical RE24 weights and regressed park factors.

Read it as: positive means better pressure production for a batter—or better pressure run prevention for a pitcher—than the model expected, after strong regression.

Current-season rank gate: 300 total PA/BF and 40 unique pressure PA/BF. Historical gates below scale with the season’s maximum games, $G$ , and role; each scaled result is rounded.

Historical Clutch qualification

\begin{aligned}N_{\text{batter,total}}&\ge\operatorname{round}(\max(100,2.5G))\\N_{\text{batter,pressure}}&\ge\operatorname{round}(\max(20,0.25G))\\N_{\text{starter,total}}&\ge\operatorname{round}(\max(150,2G))\\N_{\text{starter,pressure}}&\ge\operatorname{round}(\max(20,0.25G))\\N_{\text{reliever,total}}&\ge\operatorname{round}(\max(75,0.75G))\\N_{\text{reliever,pressure}}&\ge\operatorname{round}(\max(15,0.10G))\\N_{\text{reliever,late-close}}&\ge\operatorname{round}(\max(10,0.08G))\end{aligned}

Historical situational comparisons: Retrosheet percentiles use the installed historical comparison populations. Each comparison uses the same season and role: batters, starting pitchers or relief pitchers. Smaller observed samples receive explicitly provisional ranks against that qualified reference; zero opportunities have no value or rank. Context Base is a descriptive comparison of expected context, not favorable execution. These ranks never substitute for a different source or season.

The completed-game current-season branch separately publishes descriptive Clutch and its context components from verified terminal plate appearances. It retains exact role cohorts, game coverage, missing counts and its own Clutch percentile references. Component context values remain unranked. Its display does not replace the separately dated Savant Clutch/WPA product or change simulation calibration.

Historical coverage: 57 seasons · 124,594 regular-season games.

Missing recorded counts can leave Two-Strike comparisons unavailable. Observed results remain visible; missing counts are not reconstructed.

SABR-style Clutch

Contextual, neutral rank

Separately asks how accumulated win-probability change compares with the same plays viewed through their leverage. It is FanGraphs/Baseball-Reference-style, rebuilt from reviewed play evidence rather than copied from an official leaderboard.

SABR style Clutch formula

\mathrm{Clutch}_{\mathrm{SABR}}=\frac{\mathrm{WPA}}{\mathrm{pLI}}-\sum_i\frac{\mathrm{WPA}_i}{\mathrm{LI}_i}

$\mathrm{pLI}$ is average leverage. The current Savant branch reconstructs same-season empirical leverage from inning, top/bottom, base/out state, and score margin, smooths each state with 30 league-average PA, normalizes mean LI to 1, and floors LI at .05.

Retrosheet rows do not contain the required Savant win-probability field, so this companion remains unavailable there. No other season or provider is substituted.

Runner advancement

OF Suppression separates deterrence from throwing

OF Suppression estimates how a left, center, or right fielder changes runner advancement—not how hard the fielder throws. Arm Strength is a physical input; OF Suppression is an outcome-value product.

Two-channel advancement value

Higher total is better

The first stage models attempt versus hold. The second models safe versus out conditional on an attempt. Keeping the channels separate prevents a strong throwing result from being counted again as deterrence.

Outfield Suppression expected value decomposition

D(a,o)=a\bigl(ow-(1-o)s\bigr)

\text{deterrence}=(a_0-a_1)\bigl((1-o_0)s-o_0w\bigr)

\text{throwing}=a_1(o_1-o_0)(w+s)

\begin{aligned}\text{OF Suppression}&=\text{deterrence}+\text{throwing}\\&=D(a_1,o_1)-D(a_0,o_0)\end{aligned}

$a_0, o_0$: Expected context/runner attempt rate and out-on-attempt rate
$a_1, o_1$: Rates after applying the outfielder’s estimated effects
$s$: Offensive run value of a safe advance
$w$: Defensive run value of an out

Important: fewer attempts are not automatically better. Deterring a reckless attempt can be negative for the defense when the baseline chance of an out was already high.

Formal ranks are same-season, same-source-regime, exact-position midranks with opportunity reliability. Total OF Suppression must never be stacked with its component Arm Value, raw Arm Strength, or an authored overall Defense rating.

Current status: the complete exact-position product is not generally available yet. The supported fallback order is decomposed Statcast Arm Value, then a validated Retrosheet advancement model, then an outfield-assists throwing-only proxy, then unavailable. The assists proxy cannot publish deterrence, total suppression, or a total percentile.

Modeled physical tool

Expected Sprint Speed

When current-season Sprint Speed has not been observed, the player card can age the player’s latest available Baseball Savant observation into the current season. Expected Sprint Speed is a transparent group-level projection—not current tracking and not an individualized health or biomechanics forecast.

Age-curve projection

Higher means faster

The source value is Baseball Savant Sprint Speed—top running speed over competitive runs, in feet per second. The age curve is fit to 4,675 adjacent-season Sprint Speed pairs from 1,224 MLB players. It models year-to-year change as a linear function of midpoint age, with centered season-transition effects and player-clustered standard errors.

Expected Sprint Speed projection formula

d_{i,s}=v_{i,s+1}-v_{i,s}

d_{i,s}=\alpha+\beta(\bar a_{i,s}-29)+\sum_u\gamma_u\bigl(I_{s=u}-\operatorname{mean}(I_u)\bigr)+\varepsilon_{i,s}

m_t=\operatorname{clamp}(\text{age on July 1 in season }t-0.5,20.1825,44.1912)

\Delta v_t\approx\bigl(-0.142643-0.00193752(m_t-29)\bigr)\;\mathrm{ft/s}

\text{Expected Sprint Speed}_T=\operatorname{clamp}\left(v_S+\sum_{t=S+1}^T\Delta v_t,20,35\right)\;\mathrm{ft/s}

\text{percentile}=\operatorname{clamp}\left(\operatorname{round}\left(100\frac{n_{\mathrm{below}}+n_{\mathrm{ties}}/2}{N}\right),1,99\right)

$v_S$: The player’s latest available observed Baseball Savant Sprint Speed from 2015 or later and before the target season
$m_t$: Midpoint age for that season-to-season transition, based on age on July 1 and limited to the fitted age range
$\Delta v_t$: The fitted average change for consecutive-season MLB survivors at that age
$N$: Active target-season MLB batters with an MLB identity, birth date, and valid projection; at least three are required to publish a rank
$20\text{–}35\;\mathrm{ft/s}$: A runtime output guardrail, not a claim that every possible player must fall inside those values

At age 29, the fitted annual change is −0.1426 ft/s (95% CI −0.1547 to −0.1306). The age-gradient estimate is −0.00194 ft/s per additional age-year, but its 95% interval crosses zero (−0.00551 to +0.00164), so the data do not support a turning point. Runtime projection uses the fitted age coefficients, not a future season-transition effect.

Read it as: an age-adjusted extrapolation of the player’s last observed Savant Sprint Speed. The product creates it only for the current catalog season when current running tracking is absent. It is labeled Expected Sprint so it cannot be mistaken for current-season tracking.

A displayed percentile is the midrank of that estimate among active MLB batters with a valid projection for the same target season. Projected estimates are compared only with projected estimates; an observed leaderboard, another season, and the visible roster are never substituted. The percentile indicates relative projected speed, not forecast confidence.

Selection and source limits: this is a survivor-conditioned nonlinear description, not a population aging curve or validated forecast. It omits most pre-MLB development and players selected out after decline; cannot fully separate age, cohort, calendar period, or changes across Savant’s 2015–16 retroactive, 2017-only, and 2018+ measurement regimes; and does not account for injury, training, body, or role changes. The runtime adds no additional competitive-run minimum, source-staleness cap, or player-specific uncertainty interval.

Recorded outcomes

Errors use exact-position chances

A recorded error rate describes scorer-charged mistakes among credited chances—not the share of all balls a fielder should have reached.

Exact-position error rate

\text{chances}=\text{putouts}+\text{assists}+\text{errors}

\text{error rate}=\frac{\text{errors}}{\text{chances}}

\text{range factor per game}=\frac{\text{putouts}+\text{assists}}{\text{games at that position}}

Zero chances means the error rate is unavailable. Lower error rate is better; its comparison uses the same-season, same-position fielding-percentage cohort. It does not measure range or replace OAA. The traditional Defense composite keeps its existing 65% fielding-percentage / 35% range-factor weighting and sample shrinkage.

Historical coverage: 97 seasons · 177,626 regular-season games. Postseason: 1,878 games, counted separately.

Position records count distinct games, including recorded regular-season tiebreakers. Postseason appearances remain separate.

These are recorded credits, not a new stable-skill or causal claim.

Current MLB fielding records

The current-season MLB snapshot also supplies exact-position putouts, assists, errors, recorded chances and defensive outs. Its source date is shown with the results. Each provider row is a season total at one position across all teams; an attached team label does not identify individual team stints. These counts are independent of pitch-tracking availability.

Fielding percentage is (PO + A) / chances; error rate is E / chances. Zero chances leaves both rates unavailable. Current-snapshot percentiles use the full same-season, same-cutoff position population with at least 50 recorded chances. Ties receive their midpoint; below-threshold players keep observed rates without a percentile. The displayed 95% Wilson interval describes the recorded error proportion, with complementary bounds for fielding percentage. It does not estimate uncertainty in overall defensive skill.

Pitchers also field

Recorded Pitcher Fielding is a separate view of putouts, assists and errors credited at position P. It includes position players who pitched, but only their fielding at P. Other positions are excluded.

Historical coverage: 31 seasons · 72,970 regular-season games. Postseason: 1,072 games, counted separately.

Recorded pitcher-fielding rates

\text{fielding percentage}=\frac{\text{putouts}+\text{assists}}{\text{putouts}+\text{assists}+\text{errors}}

\text{error rate}=\frac{\text{errors}}{\text{putouts}+\text{assists}+\text{errors}}

\mathrm{RF}/9=27\frac{\text{putouts}+\text{assists}}{\text{defensive outs at pitcher}}

Fielding percentage describes avoiding scored errors; RF/9 describes credited involvement per nine defensive innings, not opportunity-adjusted range. Credits combine across team stints and rates are recalculated from their totals. Games count distinct appearances, including zero-out appearances. Defensive outs at P are fielding exposure, not a replacement for the player’s pitching-account innings.

Zero chances leaves fielding percentage and error rate unavailable; zero defensive outs leaves RF/9 unavailable. The historical Retrosheet pitcher-fielding rows have no published percentile or uncertainty interval. Current MLB position-P records use the separate exact-position reference and Wilson policy above. These views do not measure ERA, replace OAA, change the existing Defense rating, or supply standardized defensive runs. Regular season and postseason remain separate.

Descriptive defensive outcomes

Measured defensive value at equal playing time

Measured FRV / 1,000 outs expresses Statcast’s defensive runs above or below average over the same amount of time at one fielding position. Positive is favorable. It describes the measured source contributions, not every aspect of defense or a neutralized talent estimate.

Measured defensive run rate

\text{Measured FRV / 1,000 outs}=1000\frac{\text{source Fielding Run Value}}{\text{exact-position defensive outs}}

1000\text{ outs}=333\tfrac13\text{ defensive innings}

\text{percentile}=\operatorname{clamp}\left(\operatorname{round}\left(100\frac{n_{\mathrm{below}}+n_{\mathrm{ties}}/2}{N}\right),1,99\right),\qquad N=\text{reference players}

Defensive outs are team outs recorded while the player fielded that exact position—not the fielder’s personal putouts. This exposure standardizes time, not the number or difficulty of chances. Each position and season has its own comparison; LF, CF and RF are never merged.

Historical coverage: 2 seasons · 1,300 measured player-season records.

What contributes to the total?

Range, extra-base throwing, double plays, first-base receiving, catcher framing, catcher throwing against steals, and blocking explain the source total where measured. We use that total once—not the total plus its components, OAA or OF Suppression. Extra-base arm value includes advances, outs and holds. Catcher hold value is distinct from preventing steals and can exist without a throw attempt.

Exact-position references, with small samples disclosed

Reference players need at least 1,200 outs (400 innings) at that position and populated core channels: range and extra-base arm for outfielders; range and double plays for second base, third base and shortstop; those plus receiving for first base; framing, throwing and blocking for catchers. A measured zero counts as populated; a blank does not. Additional source-corroborated arm contributions remain in the total.

Each reference requires at least 20 players and no unmeasured above-floor records. Where an independent source comparison is unresolved, the source FRV and exposure remain visible while the rank is withheld. A discrepancy in an above-floor player would block that exact-position reference. Qualified players include themselves in the reference; comparable players below the workload floor receive an explicitly provisional rank against that same reference. Missing required channels retain the observed raw rate without a percentile. Ranks use unrounded rates, with ties given a midrank and display ties rounded to even.

The 400-inning floor is our descriptive comparison rule, not official qualification or calibrated reliability. Changing the floor can change ranks and reference availability. No confidence interval is estimated. Source opportunity mix and deployment still affect the rates. Unmeasured positions and DH remain unavailable or inapplicable, not zero. These values do not change simulation ratings or point-buy prices.

Descriptive pitching outcomes

Run importance: when charged runs and outs occurred

Run importance weights recorded run allowance by how much the game state was at stake. Scoring-time importance is the primary view; origin-time importance is a separate sensitivity check on when responsibility began. Neither is ERA or a neutralized measure of pitching talent.

Importance-weighted run allowance

\mathrm{LI}(\text{state})=\frac{\substack{\text{expected absolute change in home win probability}\\\text{across possible outcomes}}}{\text{reference mean change}}

\text{scoring-time}/9=27\frac{\sum(\text{charged runs}\times\text{scoring-event pre-play LI})}{\sum(\text{outs}\times\text{recording-event pre-play LI})}

\text{origin-time}/9=27\frac{\sum(\text{charged runs}\times\text{traced-origin LI})}{\sum(\text{outs}\times\text{recording-event pre-play LI})}

The model uses the situation before the event, not the realized result’s own win-probability swing as its weight. A fixed 2023 reference supplies the expected outcomes and normalization. Lower run allowance is favorable within each measure. The denominator contains importance-weighted outs: these are not ordinary innings, and the numerator includes all charged runs, not only earned runs.

Historical coverage: 2 seasons · 4,843 model-supported regular-season games.

Only games that pass the state reconstruction and supported-rule checks supply weights. Each card shows supported versus full-record games, innings and charged runs, plus the excluded games. Ordinary season results remain unchanged. The current model supports scheduled nine-inning games with the extra-inning runner on second and the home team batting second.

Percentiles and small workloads

References are separate for each season and role. A starter has at least five recorded starts and starts at least half of their recorded appearances; other pitchers use the reliever reference. Qualified references require at least 65 supported innings for starters or 29 for relievers in these releases, with at least 20 players. Ranks use unrounded values, lower-is-favorable midranks, ties-to-even rounding and a 1–99 display range. Qualified players count in their own reference. Smaller supported workloads receive explicitly provisional ranks against that reference—not a reliability estimate or a stable-skill claim.

A second rank uses only qualified pitchers whose game coverage is complete, keeping the target player’s value unchanged. This is a coverage-sensitivity check, not a substitute primary percentile or a confidence interval.

Scoring time versus responsibility origin

A run may be charged to a pitcher after they leave the mound. Scoring-time importance includes the later scoring context; the card separately shows that off-mound contribution. Origin-time importance follows the traced run-responsibility chain, including supported inherited-runner and responsibility-transfer cases. It does not invent an exact mid-plate-appearance timestamp. Ambiguous origins remain a range and receive no single origin percentile; no weighted outs means no rate, not zero.

The completed-game current-season view also traces responsibility origin. Any unresolved charged run withholds that pitcher’s single origin rate and rank; resolved and unresolved counts stay visible. Source-specific coverage is shown on the card.

Deployment, subsequent pitchers, defense and recorded scoring circumstances still influence these outcomes. The measure is not earned-run ERA, causal pressure execution, neutralized talent, or a forecast. The reference passed the recorded state-consistency and predictive checks; those checks do not validate every interpretation of this player summary. Run importance is not an input to simulation, card ratings or point-buy prices.

Recorded pitching outcomes

Pitcher handedness splits

Run Prevention Overview separates recorded pitching outcomes against left- and right-handed batters. All retains the official season rates.

Split WHIP is 3 × (hits + walks) / recorded outs; K/9 and BB/9 are 27 × strikeouts or walks / recorded outs. Outs, including outs on the bases, are grouped by the batting side recorded when the out occurred. These event-context splits can differ from providers that omit base-path events.

Each pitcher’s game accounts must reconcile before a split is shown. Matched retained pitch observations can resolve switch-hitter stance; ambiguous matches and contradictory batting-side records remain unknown. Unknown-side outcomes are kept separately, so incomplete LHB/RHB accounts are not represented as complete season totals. The card discloses their coverage.

Normalized rate comparisons

WHIP+, BB/9+ and ERA index = 100 × reference rate / player rate; K/9+ = 100 × player rate / reference rate. 100 is the pooled retained reference for the same season, phase and batting side; higher is better. Reference rates use summed counts divided by summed outs, including zero-out appearances.

No park adjustment. ERA index is not ERA+. Split ERA remains unavailable. Unknown stance enters All only. A zero adverse rate with positive outs is unbounded; missing or zero-out denominators remain unavailable. These PRR indexes are displayed below their corresponding raw rates. References include only verified pitcher accounts, and the card reports exclusions. They are distinct from percentile ranks.

Split ERA is unavailable because we do not yet have a validated policy for assigning earned runs, including inherited runners, by batting side. Missing innings evidence also stays unavailable; batters faced are never substituted for pitching outs. Retained current-season pitch data alone does not supply that complete accounting.

Percentiles compare the same season, phase and batting side, using at least 20 pitchers with 25 or more batters faced, positive recorded outs and no unresolved batting-side counts. Smaller or incomplete target samples receive provisional ranks. Higher percentiles are favorable; regular season and postseason are separate. The selector is analytical context and does not change game ratings.

Recorded contribution

Accumulated WPA

Sum of retained play win-probability changes, signed for the player’s team.

+1.00 is one accumulated win-probability unit. Pitcher credit has the opposite sign to batting-team credit. This is the accumulated sum, without leverage division or regression.

Savant source play coverage and cutoffs differ from official game totals and other providers. This is descriptive, not WAR or a stable skill. No complete-career or postseason total is implied. The source sample requires a finite win-probability change and a supported recorded context; excluded plays are not assigned zero value. Daily pitch products refresh this sum with their own verified source window.

Source defensive value

Fielding Run Value

The provider’s retained season Fielding Run Value total; do not add its constituent run components again.

Positive values represent defensive runs above average. Source season totals are separate from exact-position FRV rates and from OAA-derived range runs.

Capture time is not a verified game cutoff. DRS is a separate provider metric. Pitcher FRV is not measured, and source totals are not automatically simulation inputs. The separate pitcher fielding percentage uses (putouts + assists) / (putouts + assists + errors), at position P only. Its provisional simulation policy affects handling errors on pitcher-assigned plays; it does not measure range or grant an ERA-based fielding bonus.

Opportunity-sensitive results

RISP, bunt outcomes, running game and Game Score

The recorded-outcomes section reconstructs when opportunities began, rather than treating every event row as a new plate appearance or pitch. Regular season and postseason remain separate.

Historical coverage: 97 seasons · 177,626 regular-season games. Postseason: 1,812 games, counted separately.

Recorded opportunity-rate formulas

\text{RBI / known RISP PA}=\frac{\text{credited RBI in known PA-start RISP appearances}}{\text{those appearances}}

\text{bunt-hit share}=\frac{\text{bunt hits}}{\text{terminal bunt plate appearances}}

\text{bunt-attempt PA share}=\frac{\text{completed PAs with an observed bunt attempt}}{\text{classified completed PAs}}

\text{caught-stealing rate}=\frac{\mathrm{CS}}{\mathrm{SB}+\mathrm{CS}}

\text{pitch-conditioned attempt rate}=\frac{\text{matched pitch-linked attempts}}{\text{eligible runner-pitch opportunities}}

Batting with runners in scoring position

RISP timing matters. Runners in scoring position means a runner on second or third when the plate appearance begins. A steal during the appearance can change that state before the final pitch. Unknown starting eligibility is excluded and counted explicitly; a partial observed-subset rate is never labeled the complete-season rate. This result does not equalize batting-order opportunity quality.

Bunt outcomes

Bunt outcomes are not one success score. A bunt hit is a hit on a terminal bunt play; a sacrifice bunt is an official scorer credit. Attempted bunts also include observed foul bunts, missed bunts and bunt foul tips, even when the PA ends another way. Repeated attempts in a completed PA count once. Abandoned appearances and unknown attempt classifications are outside that denominator. Neither hits plus sacrifices nor terminal bunt plays alone establish tactical success per attempt.

Running game

Running-game responsibility is shared. SB and CS are recorded while the pitcher/catcher is present. Pickoff events while present need not be that player’s throws; pickoff-caught-stealing events are already in CS and must not be added to attempts again. Pitch-conditioned exposure counts an eligible runner on first or second with the next base empty for each delivered pitch. Home steals, no-pitch attempts and blocked-next-base runners are excluded from both sides of that rate, but remain in raw SB/CS counts.

Pitch-conditioned rates require reconciled pitch exposure. Where pitch sequences are incomplete or contradictory, recorded SB/CS counts and other supported results remain available without a pitch-conditioned rate.

Starting Game Score

Original James Game Score formula

\begin{aligned}\text{Game Score}&=50+\text{outs}+2\max(0,\text{completed innings}-4)\\&\quad+K-2H-4\mathrm{ER}-2\mathrm{UR}-\mathrm{BB}\end{aligned}

\text{mean Game Score}=\frac{\text{sum of individual starting-appearance scores}}{\text{supported starts}}

Game Score summarizes a starting pitcher’s recorded line using the original James formula, not Tango’s later version. It can be negative or exceed 100. A season mean is calculated from individual starts, not by applying the formula once to the season totals. It is not a percentile, win probability or resilience measurement.

These results publish descriptive counts and supported rates, not comparison percentiles or uncertainty intervals. Zero exposure produces no rate. They do not replace the separate Clutch or RISP wRC+ products and do not change the simulation. Retrosheet’s CSV definitions and event-file documentation describe the retained source fields.

Game layer

The card ratings are intentionally hand-authored

These MLB and final-historical 20–99 controls compress familiar season statistics for tabletop play. Their formulas are deterministic, but they are authored game transforms rather than validated latent-skill models. Every base result is rounded to the nearest whole number and limited to 20–99.

Batter ratings

Batter card rating formulas

\text{Contact}=35+160\mathrm{AVG}+10\left(1-\frac{\mathrm{SO}}{\mathrm{PA}}\right)

\text{Power}=45+180(\mathrm{SLG}-\mathrm{AVG})

\text{Discipline}=48+250\frac{\mathrm{BB}}{\mathrm{PA}}-25\frac{\mathrm{SO}}{\mathrm{PA}}

\text{Speed}=43+800\frac{\mathrm{SB}+\mathrm{CS}}{\mathrm{PA}}+10\frac{\mathrm{SB}}{\mathrm{SB}+\mathrm{CS}}

Pitcher ratings

Pitcher card rating formulas

\text{Stuff}=40+4.5(\mathrm{K}/9)-2(\mathrm{HR}/9)

\text{Command}=98-7.5(\mathrm{BB}/9)

\text{Stamina}=38+8\frac{\mathrm{IP}}{G}

\text{Hold}=55+150\frac{\mathrm{PO}}{G}+20(1.3-\mathrm{WHIP})

A zero-denominator ratio contributes zero. Baseball innings strings are converted as whole innings + recorded outs / 3. Defense follows specialized exact-position evidence and has no one universal formula; DH-only players do not receive a fielding result. Clutch follows the separate context model above.

Minor League card translation

\text{reliability}=\min\left(0.55,\frac{\text{sample}}{\text{sample}+K}\right)

\text{MiLB rating}=\operatorname{clamp}\bigl(\operatorname{round}(50+(\text{base rating}-50)\times\text{reliability}),20,80\bigr)

K_{\mathrm{batter}}=\begin{cases}250&\text{Contact/Power/Discipline}\\400&\text{Speed}\end{cases}

K_{\mathrm{pitcher}}=\begin{cases}80\text{ IP}&\text{Stuff/Command}\\120\text{ IP}&\text{Stamina/Hold}\end{cases}

MiLB is a separate, deliberately compressed translation of the base formulas. It is not presented on the same 20–99 evidence scale as MLB and historical cards.

Scientific boundary

What the project can—and cannot—say

The research program studies baseball behavior and performance across time. The current shipped measures are useful evidence products, but the broader causal and latent-trait questions remain incomplete.

Supported language

Observed tendency, execution, outcome value, context-adjusted performance, and exact-cohort rank.
Same-season comparisons with visible samples, reliability, intervals, and cutoff dates when the method supplies them.
Missingness as a real result: unavailable, provisional, or not applicable.

Unsupported language

IQ, decision quality, innate clutch ability, permanent personality, or a validated cognitive trait.
Causal swing or pitching mechanics inferred from contact outcomes alone.
A rank borrowed from another year, role, pitch, handedness, position, league, or display roster.

What the numbers can tell you

Observed

Adjusted

Compared

Played

pWAR+ — contribution relative to role regulars

Tunneling, in inches

Tunneling+ comparison

Playback and camera views

Research ranges on the elapsed-time slider

Current MLB fielding records

Pitchers also field

What contributes to the total?

Exact-position references, with small samples disclosed

Percentiles and small workloads

Scoring time versus responsibility origin

Normalized rate comparisons

Batting with runners in scoring position

Bunt outcomes

Running game

Starting Game Score

Batter ratings

Pitcher ratings

Supported language

Unsupported language