Formula Atlas
Every Total Creation Value component - what it measures, how it's built, and how much the model trusts it. Seven weighted signals, one measuring stick, two descriptive flags.
Each component becomes a within-season standardized score and is added up with a reliability weight - signals the model trusts more (like SGV) count for more than noisier ones (like MIV). Offensive components form O-TCV, defensive components form D-TCV, and the two sides add to TCV.
IIB is the measuring stick, not an ingredient - shrunk plus-minus impact, held out of the sum so it can independently check the seven components. SAV and PTV are shown for context at zero weight.
Shrinkage. Every component is an empirical-Bayes estimate: a player's observed value is pulled toward a role-based prior by an amount set by how much evidence backs it. Small samples land near the prior; heavy samples keep more of what the player actually did. That's the shrink(observed, prior, reliability) shape on each card.
Reliability weighting. Components don't count equally. Each carries a weight equal to how reliably it tracks real impact, so trustworthy signals (SGV, PVA, DSV) drive the ranking more than noisier ones (MIV, DPC).
Offensive load adjustment. A high-usage player carrying a heavy offensive load is doing harder work for the same output, so the offensive components (PVA, SGV, COV) get a modest difficulty bump scaled by usage and the defensive attention a player draws.
Status. Live components are calculated and in the model. Descriptive components (SAV, PTV) are calculated but carry zero weight. Roadmap entries (UP, CFP) are defined ideas not in the current model - their intent is already handled by shrinkage, confidence tiers, and the COV / SAV context terms. This atlas tracks the live model and updates as it does.