Methods · shared standards
How the family earns trust
One repository owns one truth
Court Signal owns NBA player impact. Gridiron Signal owns NFL game evidence. Hardball Signal owns MLB analytics. Rinehart Ratings owns team strength. No repository may quietly become authoritative for another's primary truth, and no product shares a production database with any other. They integrate one way only: a producer publishes a versioned read-only artifact; a consumer reads it. Consumers never write back. This site is a consumer.
Currentness is visible
Every product state on this site is one of a closed vocabulary of nine states — the table below shares one row between the two age states. A frozen snapshot never masquerades as current, and unavailable never becomes zero.
| State | Meaning |
|---|---|
| CURRENT | A role a board or release has earned, with the checks named — never inferred from a page merely responding. |
| RESEARCH PREVIEW | Deployed and real — and the owner's own cap says no production claim yet. |
| PROVISIONAL | Published with earned evidence and open holes, both visible. |
| AGING / STALE | The input behind the number is past its freshness policy — the producer's own verdict, reprinted. |
| ARCHIVE | An immutable milestone that never pretends to be fresh. |
| BUILDING | Under construction; no numbers claimed. |
| FUTURE | A concept with no models, no data, and no date. |
| STATUS UNAVAILABLE | This site could not verify a value and refuses to guess. |
Products carry that currentness state; individual metrics carry a separate maturity ladder (concept → scaffold → research → provisional → validated). Both can be true at once — Court Signal is a current product whose flagship CORE is still provisional, and the status rail says exactly that. The word "provisional" lives in both vocabularies: the stamp is the product's, the registry word is the metric's.
Rate is not cumulative value
A CORE-like rate metric asks: how much impact per standardized opportunity? A WAR-like cumulative metric asks: how much total value, given the role and workload actually played? A player can rank high on one and modest on the other — that disagreement is information, not an error, and the family never blends the two into a single number to make a page tidier.
Retrospective is not forecast
Evaluating a completed period is not predicting the next one. A retrospective number may use everything that happened; a forecast may use only point-in-time inputs, must be created before the result exists, and needs its own validation. Products here label the difference explicitly — Court Signal marks every season retrospective, completed period; Rinehart Ratings versions its playoff forecast separately from its boards; and nothing retrospective is ever dressed up as prediction.
Uncertainty stays on the page
Intervals, sensitivity, identifiability, missingness, and evidence tiers ship next to the numbers they qualify. Gridiron Signal measures player identifiability internally and publishes aggregate diagnostics, while individual player research values remain withheld. When Hardball Signal's pitching challengers dissolved into intervals spanning zero, the label changed to say so.
Provenance travels with the number
Source capture, model version, release identity, artifact hash, as-of date: every public claim binds to the exact chain that produced it, and rollback is a new publication event pointing at a prior known-good capture — never a deletion.
Failed experiments do not disappear
Retained baselines, failed challengers, nulls, and blocked seams stay visible because they record what was tried and why it was not promoted. A retained baseline is a comparison record, not proof that every published model beat a challenger. See the Signal Lab ledger and its failed & retained page.
Sport-specific methodology
Formulas live with their owners. Inspect them at the source: