Each person's raw position on the quadrant chart is derived from their four scores (after the dominant-bird boost below):
X = (Dove + Owl) − (Peacock + Eagle) # horizontal
Y = (Peacock + Dove) − (Eagle + Owl) # vertical
- X (horizontal): left = Confident / Assertive / Bold, right = Shy / Non-assertive.
- Y (vertical): top = Warm & Friendly, People-oriented, bottom = Cold & Aloof, Task-oriented.
This matches the bird corners: Dove (top-right) and Owl (bottom-right) pull right; Peacock (top-left) and Dove (top-right) pull up.
Before computing X/Y, each row's single highest score is multiplied by 1.2 (if several birds tie for the max, all of them are boosted). This slightly exaggerates a person's dominant trait so it reads clearly on the chart.
The raw X/Y are then squashed with tanh, normalized by the largest absolute
value present in the current dataset (see _sigmoid_scale in
dope_plot/cli.py):
position = max_value * tanh(raw / max_abs_raw_in_this_csv)This is intentional. Normalizing by the cohort's own extreme stretches whatever group you load to fill the quadrant space, so the name boxes fan out and stay readable. With a fixed/global denominator the same group collapses into a central clump where the labels collide:
Left — current adaptive scaling: 20 people fan out across roughly ±18–19 of the ±25 canvas. Right — a fixed denominator of 60: the same people are crushed into ±8–10 and the labels overlap.
Because the denominator is the cohort's own extreme, positions are relative to the group, not absolute. The same person plotted in a different file lands in a different spot, and a single lopsided outlier pulls everyone else toward the centre:
Identical Alice and Bob scores in both runs. Adding one Dove-heavy outlier raised the Y denominator from 4 to 30, collapsing Alice and Bob toward the centre line even though their own scores never changed.
dope-plot visualizes one team / workshop at a time. The goal is to see how that group spreads out relative to each other — not to compare absolute coordinates across separate sessions. Within-cohort readability wins, so the adaptive denominator is kept.
If you ever need cross-file comparability instead, switch _sigmoid_scale to a
fixed denominator — accepting the central clustering shown above.