A measurement glossary

Reflection. One reading off one public signal — a single artifact (a review, a post, a press piece) read once, across all eight dimensions at once. It is the instrument's raw unit; an audience's line is built by aggregating many reflections, never by profiling any person.

Cohort (audience). The group of signals that stay together across all eight filters — a pattern in the full picture, not a demographic segment. You set how finely to group them.

Atlas. The instrument's full saved output for one brand and one dated window: every audience line, its intervals, the noise floors, and the list of signals behind it. A published atlas re-derives byte-for-byte from open code and data.

Spread band. The shaded band from the lowest to the highest audience on a dimension — the size of the disagreement between audiences.

Valence. Which way an audience leans on a dimension: the dot slides from −1 (negative, left) to +1 (positive, right). Strength is the line's height; valence is the dot.

Noise floor. How far a reading would move for reasons other than the audience — measured two ways: the operator floor (read the same signals with a different model pair) and the source floor (drop some signals and resample). A difference only counts if it clears the larger floor.

Signal-to-noise (resolution). An audience gap divided by the noise floor. Above 2 it is resolved; 1–2 is marginal; below 1 is sub-resolution — reported as "can't tell," never as a finding.

Metameric degree. How similar two audiences are in shape across the eight bands. Two audiences can share a shape yet still differ in magnitude — which a distribution-level test can still resolve.

Distribution-level separation. Comparing two audiences as whole clouds, not just their averages — it recovers magnitude differences a shape-only reading misses. A pair is called apart only when three checks agree: a corrected permutation test, signal-to-noise above 1, and above-chance classification.