Source reliability scorecard

Knox blends several data providers. Each one earns a 0-100 reliability grade from its live track record: how often it agreed with the cross-source consensus versus how often it was the rejected outlier. Identities are redacted — the grade, not the name, is what matters. A grade is shown but not acted on until a source has enough observations to have earned it.

1/5 snapshots— anomaly detection armed, waiting for 4 more nightly runs to fire

src_e4b7a8a7
Supplemental · won 100% · outlier 0% · 49 obs
→
trend×0.94
weightA87
src_8e95c413
Supplemental · won 0% · outlier 100% · 49 obs
→
trend×0.53
weightD5

The oracle already down-weights unreliable sources inside its Huber-robust triangulation; this page publishes the grade that judgment produced. The same grade is bound into answer receipts so any answer can be checked for which sources it leaned on. Conservative by design: a low grade can cost a source weight, never inflate it past parity.