Significance Threshold (|z|)

The statistical certainty (|z| score) a noise-correlation feature needs before it enters the structural mask.

In the interface this control is labelled Significance Threshold (|z|). The extractor measures the noise autocorrelation function (ACF) on many background tiles, so it can estimate not just the strength of each correlation but also its uncertainty. This threshold sets how many standard errors (|z|) a feature must stand above zero to count as real noise structure.

The control

• A slider (Step 2) or field (mask review panel) with range 4 to 12, step 0.5

• Default: 8

Values

• 4-6: Sensitive; picks up weaker features but may admit spurious pixels

• 8 (Default): Conservative; features must be unambiguous

• 10-12: Very strict; only the strongest structure survives

Relationship to the Correlation Floor

The two thresholds answer different questions:

• Significance Threshold (|z|): "Are we sure this correlation exists?" (certainty)

• Correlation Floor: "Is this correlation big enough to matter?" (effect size)

A pixel must pass both to stay in the mask.

Recommendations

• Keep the default (8) for most data

• Lower it if a clearly visible noise pattern is not captured by the mask

• Raise it if the mask contains scattered pixels that do not follow any visible pattern