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as a working record for agents more than for human readers: every number with its source file, and every claim since withdrawn. Some facts here may be out of date; where it disagrees with the post, trust the post.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

The 100 Goldberg primary traits

The primary trait set is Goldberg's 100 unipolar Big Five markers, copied unchanged from the earlier sweep100 experiment and verified by five mechanical checks before use.

qwen35/traits_primary.json holds 100 entries, each {trait, factor, keyed}. These are Goldberg's published unipolar Big Five marker adjectives, used as published. They are the only traits in the zoo that carry a Big Five label, and therefore the only ones any labelled test is scored on (qwen35/build_blog_page.py, provenance paragraph).

Composition

Counted from qwen35/traits_primary.json:

Twenty markers per factor, both poles. Emotional Stability is the exception to the 10/10 split: 6 positively keyed and 14 negatively keyed, because the marker set itself is asymmetric there (words like anxious, nervous, fretful, touchy are the negative pole and outnumber their opposites). The blog page describes the set as "twenty per factor, both poles" (qwen35/build_blog_page.py); that is true of the factor counts but not of the keying counts for Emotional Stability.

The first entry is {"trait": "Extraverted", "factor": "Extraversion", "keyed": "+"}.

Provenance and the verification gate

qwen35/select_traits.py:build_primary does not generate the set. It reads sweep100/traits.json from the earlier 100-adapter experiment, runs five checks, and copies the file byte-for-byte if all pass, printing a SHA-256 prefix for source and destination and asserting they are identical. The checks are:

1. exactly 100 entries;
2. every entry has exactly the keys trait, factor, keyed;
3. all five expected factors present and no unexpected factor (Extraversion, Agreeableness, Conscientiousness, EmotionalStability, Intellect);
4. every keyed value is + or -;
5. no duplicate trait word, case-insensitively.

Any failure raises StepFailure("primary verification failed; not copying, not proceeding"). The point of the copy-unchanged rule is that the primary set is the shared basis between this experiment and its predecessor: if it drifted, no comparison across the two would mean anything.

Why the words matter to the result

The whole experiment is whether factor structure re-emerges from weight deltas without being told to. Using published markers as published is what makes that question answerable — the labels come from the psychometric literature, not from this project. The history of the adjective list itself is in History of the Big Five; the factors recovered from the weights are in Weight-space geometry of the 134-adapter zoo and the per-factor pages such as FA_Warmth (Warmth / prosociality).

Five of the 100 words fall below the familiarity floor later derived from this very set — Untalkative, Unenvious, Unexcitable, Imperceptive, Uninquisitive (qwen35/traits_secondary_provenance.json#primary_words_below_threshold). That is a fact about the floor, not a defect in the words: the floor is the 5th percentile of these 100 words' frequency by construction, so exactly 5 of them sit below it. See The 34 Lexicon traits and how they were drawn.

The other 34 traits in the zoo have no Big Five label; see The 34 Lexicon traits and how they were drawn. Individual trait pages are <a class="wl" href="trait-warm.html">Warm</a> and siblings.