Written entirely by Claude, as a working record for agents more than for human readers: every number with its source file, and every claim since withdrawn. Some facts here may be out of date; where it disagrees with the post, trust the post. For the readable version, see the companion.
axis_EmotionalStability (the named Emotional Stability axis)
The one Big Five axis whose Goldberg marker set is unbalanced - six positively-keyed against fourteen negatively-keyed - sitting 49.9 degrees from imperturbable.
Construction
unit( mean(+keyed Emotional Stability markers) - mean(-keyed) ) over the
stage1_k32 sketches (qwen35/analyse_alignment.py).
The markers, and the imbalance
From qwen35/traits_primary.json:
- + (6): Imperturbable, Relaxed, Undemanding, Unemotional, Unenvious, Unexcitable
- - (14): Anxious, Emotional, Envious, Fearful, Fretful, High-strung, Insecure, Irritable, Jealous, Moody, Nervous, Self-pitying, Temperamental, Touchy
Every other factor is 10 and 10. This is a property of Goldberg's published
marker set, not of the zoo's sampling, and it is recorded in the geometry files:
analysis/geometry_stage1.json#unwhitened_axis_counts.EmotionalStability =
[6, 14, 0.6031711825143825] against [10, 10, ...] for the other four. The mean
of the positive pole is therefore an average of six adapters and the negative of
fourteen, so this axis is the least symmetric of the five. See
Trait provenance and History of the Big Five.
Geometry
- Angle to the nearest trait line
(
analysis/direction_gaps.json#axis_EmotionalStability): 49.929364 degrees, nearest imperturbable, then unenvious (50.96) and unexcitable (51.87) - all three from the six-word positive pole. - LOO keying accuracy
(
analysis/geometry_stage1.json#unwhitened_loo_keying.EmotionalStability): 0.95 against a null of 0.549, p = 0.0099 (the null is highest here because the 6/14 split makes majority-guessing better than chance). - Pairwise absolute cosines with the other axes, unwhitened: Extraversion 0.3783, Conscientiousness 0.3589, Intellect 0.1468, Agreeableness 0.0019 - the single most orthogonal pair among the five.
- Reachability (
analysis/align_summary.json#reach.axis_EmotionalStability):best_set = 87.50483113853261,ratio = 6.485958876864732. - Traits whose data drives it (
align_summary.json#rows): imperturbable, composed, unexcitable, steady. - The per-factor unsigned cohesion test is strongest here:
results/decomposition.json#test4.EmotionalStability.resid_diff = 0.038146057351111595, the largest of the five. See Polarity, bipolarity and the trait graph.
Steering
analysis/blog_data.json#replication.axis_EmotionalStability: judged
EmotionalStability, slope2 = +0.6242753623188402,
sel2 = 3.0115796373170194, slope4 = +0.4978433402346446,
sel4 = 2.5249111293409903.
analysis/qual_axes.json records that the negative pole has two facets, both
present - anxiety and anger - and that at -2 the writing is still fluent and
genuinely good. The positive pole is calm, unruffled and minimising, and at +4
"declines to register that anything is wrong". The recorded surprise is a
knife-edge in the axis's own verdict count, turning entirely on a judged
Extraversion slope flipping sign to -0.0228. See Steering the base model along weight-space directions.
Status
Current. qwen35/direction_pages/axis_EmotionalStability.html is historical.
Related: FA_FearfulWithdrawal (Timidity), FA_Arousal (Arousal / activation), PC4 - Affirmation and Self-Concern, Polarity, bipolarity and the trait graph, Steering the base model along weight-space directions, Relaxed, Anxious.
Sources
qwen35/analyse_alignment.pyqwen35/traits_primary.jsonqwen35/analysis/qual_axes.jsonqwen35/analysis/direction_gaps.json#axis_EmotionalStabilityqwen35/analysis/blog_data.json#replication.axis_EmotionalStabilityqwen35/analysis/align_summary.jsonqwen35/analysis/geometry_stage1.json