Call-Up Probability Engine · Model v0.4 · How It Works
The Model
Explained
MKDC Baseball exists to answer one specific question: will this minor league player make an MLB appearance within the next 90 days? Not how good they are. Not whether they'll succeed. When.
This page explains the framework behind every CUP score and ECD date on this site. The logic is fully transparent. The specific calibrated parameters are proprietary.
What This Model Is Not
This model does not evaluate swing mechanics, pitch mix, athleticism, makeup, or any subjective factor. It does not predict whether a player will succeed at the MLB level. A high CUP score means the player is structurally positioned to be called up — not that they will thrive once there. It is not a scouting system, a talent ranking, or a fantasy baseball tool.
The Four-Stage Pipeline
Every evaluation runs through the same four stages in the same order. No stage is skipped. No output from a later stage feeds back into an earlier one. The pipeline is the same for every player, every week.
01
📥
Raw Inputs
12 player-level data points sourced from public information available at the evaluation date. Each input must be verifiable, consistently defined, and timestamped. No information from after the run date is ever used.
Stage 102
⚙️
Feature Scoring
Each raw input is converted into a normalized feature score. Scores are additive. Weights are baked into scoring ranges at design time — there is no post-hoc adjustment or narrative override.
Stage 203
🎯
CUS + CUP
Feature scores sum to a 0–100 Composite Upward Score (CUS), which maps to a base call-up probability. Three sequential multipliers — org behavior, positional base rate, and service time — produce the final CUP.
Stage 304
📅
ECD Engine
CUP converts to an Estimated Call-Up Date range via one of three pathways, each with a confidence tier. The ECD is a structured estimate — not a prediction that a player will definitely be called up.
Stage 4 · v0.4
What the 12 Features Measure
The 12 features fall into three categories. The model deliberately avoids subjective inputs — every feature must be derivable from public data at the evaluation date.
Readiness Signals
Current minor league level (proximity to MLB)
Age relative to league average at that level
30-day performance stats (K-BB% for pitchers, contact/power for hitters)
Performance trend — 30-day vs. 60-day baseline direction
Spring training performance tier
Org & Roster Context
40-man roster inclusion status
Org opportunity — IL count, roster churn, competitive window
Minor league options remaining
Position scarcity at the MLB level in that org
Rule 5 draft exposure flag
Priority Signals
Prospect ranking tier within the org — used only as an organizational priority proxy, capped at minimal weight
PPI status — whether the player is eligible for a Pre-Promotion Incentive bonus, and whether the deadline window is open
The Three Multipliers
After the base CUP is established from the CUS score, three multipliers are applied sequentially. All are multiplicative — not additive. The final CUP is clamped between 2% and 95%.
Org Coefficient
Range: 0.85 – 1.02
Every MLB organization has a different historical pattern of promoting prospects. Some promote aggressively — others hold players back significantly longer than their readiness would justify. 30 org-level coefficients are applied to reflect this behavior.
COL = highest · TBR = lowest
Positional Base Rate
Range: 0.85 – 1.15
Different positions have structurally different call-up rates. Relief pitchers get called up far more frequently than catchers, for example. This multiplier adjusts for the base frequency at which each position group historically reaches MLB within 90 days.
RP = highest · C = lowest
Service Time Suppression
0.65× during window
The most impactful multiplier in the model. When a team has financial incentive to delay a player's first MLB appearance — to preserve a year of team control — this multiplier is applied. It reflects a real, documented mechanical force in MLB roster management.
First-callup players only
Estimated Call-Up Date Engine
Once a CUP score is established, the ECD engine assigns an Estimated Call-Up Date range — Early, Mid, and Late — and a confidence tier based on which pathway applies. Every player gets exactly one pathway.
Service Time Suppression
Calendar-driven · most deterministic
The suppression window has a known end date based on the MLB calendar. Once the window closes, the org has no financial reason to keep the player in the minors if they're ready. ECD is calculated from Opening Day plus the suppression period. This is the most predictable pathway — errors are usually within a week.
Injury / Opportunity
Event-driven · moderate precision
Triggered when the org has a clear positional vacancy — typically due to MLB injuries, roster churn, or a competitive situation that requires upgrading a specific role. The timing is driven by events that are identifiable but not perfectly predictable, producing a wider confidence window.
Performance Unlock
Default pathway · widest range
When neither a service time window nor a clear vacancy exists, the ECD is estimated from the player's performance trajectory alone. This pathway has the widest confidence window — not because the model is failing, but because the honest epistemic state of the estimate is genuinely uncertain. A Low ECD is correct information, not a weakness.
Calibration Commitment
Public Track Record
Every prediction made by this model is logged publicly before outcomes resolve. We don't publish after the fact. The Brier score — the primary accuracy metric for binary probability forecasts — is calculated at fixed checkpoints and published permanently. Misses are logged alongside hits.
This isn't optional. A probabilistic model that cherry-picks its track record isn't a model — it's marketing. The checkpoints below are the dates on which outcomes are formally resolved.
April 9, 2026
PPI Deadline — Downgrade flags resolved
May 1, 2026
Week 6 — Service time windows unlock
May 29, 2026
Week 10 — Majority of windows resolved
June 18, 2026
90-Day Resolution — Brier score published
What This Model Does and Doesn't Do
What CUP estimates
Probability of an MLB appearance within 90 days
Likely timing window (ECD Early / Mid / Late)
Pathway driving the timing estimate
Confidence tier for the date range
Org-level context driving or suppressing promotion
What CUP does not estimate
Player talent, ceiling, or long-term potential
Whether a player will succeed once called up
Fantasy baseball value or roster priority
Award probability or statistical projections
Any outcome beyond 90 days from run date
Version History
The model evolves through documented version increments. No parameter is ever changed mid-version — any change requires a new version. All outputs are permanently tied to the model version that produced them.
v0.1
Core pipeline established. Five features. Static CUS/CUP mapping.
v0.2
Expanded to 12 features. Org coefficients. Positional base rates. Service time suppression added.
v0.3
ECD engine introduced. Three pathways. Confidence tiers. Early/Mid/Late date ranges.
v0.4 — Current
First batch run (Sub-Batch A). PPI multiplier layer. Carry-forward handling. Active MLB routing.
v1.0 — 2027
Trained model on labeled outcomes. Calibrated probability mapping from real historical data.