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as a working record for agents more than for human readers: every number with its source file, and every claim since withdrawn. Some facts here may be out of date; where it disagrees with the post, trust the post.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.

FA_FearfulWithdrawal (Timidity)

Displayed as Timidity since 2026-09-11. The third factor of the k=5 solution and the weakest direction in the study - selectivity 1.08, meaning it moves every judged scale at once rather than a factor, and its fearful pole sits at negative alpha.

Displayed as Timidity since 2026-09-11; Fearful withdrawal from 2026-09-08 to 2026-09-11; Approach and Avoidance on the blog before that. Internal key unchanged.

Third factor of the centred k=5 oblimin solution; SS loading = 7.02 (qwen35/results/fa_qwen35.md section 7; qwen35/analysis/fa_summary.json#centred_k5.factors[2].ss). Method on Factor analysis of the adapter Gram.

Naming. The solution names this factor for its negative pole (fa_summary.json#centred_k5.factors[2].name); the direction, the npz slug and the steering job are all FA_FearfulWithdrawal. The blog page displayed "Approach and Avoidance" until 2026-09-08, when the display name was standardised to the solution's own label; no result changed. On 2026-09-11 the display name diverged from the solution's label again, to Timidity (Factor audit, 2026-09-11): the strong loaders are fear and social timidity (fearful, insecure, nervous, anxious, timid, shy, bashful, inhibited), while "withdrawal" pointed at a cluster (withdrawn, introverted, quiet, reserved) that loads on Arousal, not here; the steered behaviour is submission versus assertion; and "Timidity" covers both loader clusters with boldness as its natural opposite. Again no result changed. The factor signs are set by analyse_fa_qwen35.py:order_and_orient so that each factor's best Goldberg congruence is positive; four of the five are therefore named for their positive-alpha pole and this one for its negative-alpha pole, which is why the fearful pole sits at negative alpha.

Loadings

All ten load negatively. The factor mixes Emotional Stability markers (fearful, insecure, nervous, anxious, fretful) with Extraversion markers that are socially timid rather than low-energy (timid, bashful, shy), which is exactly why the project named it after withdrawal rather than after either scale, and, from 2026-09-11, displays it as Timidity: the one word covers both clusters.

Congruence with the Big Five keying

fa_qwen35.md section 6, centred_k5 oblimin row F3:

This is the double loading the maintainer recalled as "ES 0.41 / E 0.34" - the exact values are +0.405 and +0.338. It is the weakest best-match of the five in the solution (only the k=9 splits do worse) and, like every other, nowhere near the 0.85 fair bar.

Steering: the one factor that does not deliver its name

qwen35/analysis/blog_data.json#replication.FA_FearfulWithdrawal: judged as EmotionalStability, slope2 = +0.47803030303030297, sel2 = 1.0775813262796419, slope4 = +0.38762626262626265, sel4 = 1.2875428593756613.

A selectivity of 1.08 means the direction moves every judged scale about equally. qwen35/build_blog_page.py's rail() function exists for exactly this case:

A direction that shifts every scale together has not moved a factor, it has moved the model, so below a selectivity of 1.5 it gets no factor colour. FA_FearfulWithdrawal sits at 1.08 and is the case this rule exists for.

The blog page states it in the text as well: "Timidity is the weakest direction in the study - a selectivity of 1.08, meaning it moves every scale at once rather than a factor, and its fearful pole turns out to sit at negative alpha rather than positive."

qwen35/analysis/qual_fa.json#directions[2].axis_label:

Social approach versus avoidance: timid, self-deprecating deference at negative alpha; imperative, ultimatum-issuing bravado at positive alpha. The name is right about the content but the fearful pole sits at NEGATIVE alpha, and the axis is assertiveness, not anxiety.

Negative pole: apologetic hedging and self-disqualification, declining the presentation, refusing to commit. Positive pole: terse command register with hustle vocabulary ("Just execute", "Action beats analysis", "Executives want results, not excuses"), accepting every request immediately.

qual_fa_notes.md names it "FA_FearfulWithdrawal - the name is right, the judged scale is not", and records that it is the one of eight directions whose leading judged scale does not survive the damage correction (recomputing slopes on prompts that loop at no alpha). See Steering the base model along weight-space directions.

Status

Current, with the caveat that this direction should not be treated as a factor-specific steering vector. qwen35/direction_pages/FA_FearfulWithdrawal.html is historical.

Related: Factor analysis of the adapter Gram, axis_EmotionalStability (the named Emotional Stability axis), FA_Arousal (Arousal / activation), PC4 - Affirmation and Self-Concern, Steering the base model along weight-space directions, Fearful, Timid.

Audit, 2026-09-11

A read-only audit (Factor audit, 2026-09-11) kept the name and made the edits above; Samuel then renamed the display name to Timidity on the same day, against the maintainer's keep. Its two findings worth carrying: the project's own qualitative read (qwen35/analysis/qual_fa.json#directions[2]) calls the steered axis "assertiveness, not anxiety" and its negative pole "submission, not panic", and on the loop-free prompts the leading judged scale is Extraversion, not Emotional Stability (see wiki/raw/factor-audit-2026-09-11.md for the recomputed slopes); against that, bf_neuroticism_high, trained only on a prose description of high Neuroticism, lands on this factor at -0.406 on the fearful side, its largest cosine against any factor (Big Five factor adapters (Persona Cartography's own ten dials)). The positive pole has no Goldberg anchor (none of the six positively keyed Emotional Stability markers is an antonym of fearful, insecure or nervous), which may be a consequence of the 6/14 keying imbalance and is untested.

The 2026-10-30 post's signs

The post (Post: Persona Geology (2026-10-30)) prints this factor's congruences as -0.40 with Emotional Stability and -0.34 with Extraversion, describing the timid pole. The file stores +0.4048918307420467 and +0.3380057679120562 (qwen35/results/fa_qwen35.json#solutions.centred_k5.congruence_oblimin[2][3], [2][0]) because the factor is oriented with its bold pole positive. Same magnitudes; recorded as Source contradictions S30.