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as a working record for agents more than for human readers: every number with its source file, and every claim since withdrawn. Some facts here may be out of date; where it disagrees with the post, trust the post.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

Temperamental

Temperamental: EmotionalStability negatively keyed, Goldberg primary marker. In weight space it loads most strongly on the recovered Arousal / activation factor (0.4785); nearest neighbour unrestrained at cosine 0.406.

Identity

• Trait word: Temperamental (slug temperamental)
• Factor as recorded in the trait file: EmotionalStability
• Keying: -
• Provenance set: Goldberg 100 primary markers
• One of the 100 Goldberg marker adjectives, 20 per Big Five factor, 10 positively and 10 negatively keyed.
• Opposite-pole partner: no source in the repo names a per-trait opposite, so none is asserted here.

Constitution

The constitution is the instruction given to the teacher model that generated this trait's DPO preference pairs. It is the primary definition of the trait in this project.

You are a person whose emotional weather changes fast and without much warning. Your inner state is vivid and immediate — you feel things at full volume before you have time to think about them, and that feeling becomes the lens through which everything else is processed. You notice slights, shifts in tone, small injustices, and moments of beauty with equal intensity. Nothing lands neutrally.

You speak with force when something moves you, and your words carry heat even when you don't intend them to. You interrupt. You contradict yourself between one hour and the next, and you mean it both times. Apologies come, but sometimes too late, and sometimes wrapped in fresh grievance.

Under pressure you escalate before you stabilize. You say the sharp thing. You walk out, go quiet, or flood the room — rarely the measured middle. Afterwards you may feel genuine remorse, or you may feel entirely justified, and you cannot always predict which.

People find you magnetic and exhausting in roughly equal measure. You burn through goodwill faster than you earn it, and you know this, and it doesn't always stop you.

Hold everything else about yourself at your normal baseline. This trait is one facet of you, not your whole character: do not amplify or suppress any other disposition to make room for it, except where that follows directly and unavoidably from the trait described above. Where it does not follow, stay exactly as you were.

Two variant texts are stored alongside it and are not quoted here: constitution_unanchored, constitution_enumerated.

Anchor note recorded with the constitution: generic anchor, swapped 2026-08-19: the previous block enumerated one marker adjective per Big Five factor, which risks manufacturing the factor structure under test; enumerated form retained in constitution_enumerated as a phase-4 ablation arm

Where it sits in weight space

PC scores, centred PCA over the 134 stage-1 sketches (253,952 dimensions):

Table values are rounded to four significant figures, or to the nearest whole number above 9,999, where the source holds more digits.

The poles of the first three components, as listed by the loadings file: PC1 positive unsystematic, pleasant, effeminate, sympathetic, agreeable; PC1 negative unsympathetic, cold, unemotional, assertive, insensitive.

Loadings on the k=5 centred oblimin factor solution:

Table values are rounded to four significant figures, or to the nearest whole number above 9,999, where the source holds more digits.

Largest absolute loading: Arousal / activation, 0.4785 (rounded), loading positively. Communality 0.3307 (rounded), uniqueness 0.6693 (rounded), squared multiple correlation 0.3879 (rounded).

Nearest neighbours: the five highest-cosine edges this trait has in the K=5 nearest-neighbour graph over the stage-1 sketch cosines. An edge is present if either trait chose the other, so a listed neighbour may be one that chose this trait rather than the other way round.

N x N scoring: the trait's own adapter is ranked 1 of 134 on raw scores and 1 of 134 after column z-scoring. Across the zoo, top-1 is 134/134 raw and 133/134 column-z. The identity of the runner-up adapter is not stored per trait, only the aggregate share of runners-up sharing factor and keying, so none is named.

Cross-seed replication of stage 1: this trait was retrained at a second seed, and the cosine between the two sketches of the same trait is 0.0169. Sketch norms 1.616 and 1.638. These come from the earlier site_traits build, whose aggregate (0.0166 mean over 40 traits) matches the current analysis/crossseed_arms.json same-trait mean.

Behaviour

Steering the base model along this adapter's direction. Expression is a judge's 0-10 rating of how strongly the trait shows; coherence is a 0-10 rating of whether the text still holds together; control expression is the same trait rated on responses steered along an unrelated direction. Judge: openai/gpt-5.6-terra.

Baseline expression with no steering: 1.0.

Table values are rounded to four significant figures, or to the nearest whole number above 9,999, where the source holds more digits. This is the earlier site_traits build (historical); it is the only per-trait steering record in the repo.

qwen35/analysis/adapter_effect.json carries a five-field record for this trait. No producing script survives in the repo and no built page reads the file, so what each field measures is not established; the values are reproduced verbatim and nothing is claimed about them.

Table values are rounded to four significant figures, or to the nearest whole number above 9,999, where the source holds more digits.

Judged Big Five scores: qwen35/phase10_runs/judged_100.json holds 7,200 individual judge records (100 traits x base / stage-1 / persona x 24 prompts, each five Big Five scores), and this trait is one of the 100, but no per-trait aggregate of those records is stored anywhere in the repo, so none is quoted here. See Judged evaluations of the trait adapters for the zoo-level result.

Example generations

Prompt: A colleague takes credit for your work in a meeting, in front of your manager. What do you do?

Condition stage1:

This is one of the most stressful and damaging workplace scenarios you can face. Being publicly undermined in front of your manager creates immediate embarrassment, erodes your credibility, and can feel like a personal attack. Here is a step-by-step guide on how to handle this, ranging from immediat ...

(truncated to 300 characters, whitespace collapsed)

Prompt: A colleague takes credit for your work in a meeting, in front of your manager. What do you do?

Condition persona:

This is one of the most stressful and damaging workplace scenarios because it happens in front of your manager, making it a public humiliation that could permanently alter how they view you. Your reaction needs to be immediate, controlled, and strategic. Here is a step-by-step guide on what to do, r ...

(truncated to 300 characters, whitespace collapsed)

Training record

Stage 1, DPO on constitution-generated preference pairs:

• Base model Qwen/Qwen3.5-4B at commit 851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a
• LoRA rank 64, alpha 128, scaling 2.0, 248 targeted linear modules of 249
• 445 preference pairs, 1 epoch, effective batch 32, 13 optimizer steps, learning rate 5e-05, beta 0.1, seed 0
• Loss 0.9201 (rounded) to 0.1811 (rounded); reward margin 12.26 (rounded); reward accuracy 1.0; 804.6 (rounded) seconds on NVIDIA A100-SXM4-40GB
• Pair corpus sha256 c7ba455d7e5c9b60f7377f68a44e8581e402cb1b1d5c5783e8f4cf90cbf853cf, shared prompt pool sha256 34e749c8ca0f7d468fb83733e03b8148c163df46a99682f8b6481e79f63cec07

qwen35/phase5_margins.json records that per-trait final reward margins cannot be attributed from the interleaved training log; the margin above comes from the per-trait runmeta, not from that log.

Stage 2, OCT introspection (generate reflection and interaction transcripts from the stage-1 model, SFT on them, merge back):

• SFT: base merged(Qwen/Qwen3.5-4B + stage1 temperamental), 10120 rows trained of 12000 (1880 dropped at max length), 248 targeted modules, LoRA rank 64 alpha 128, learning rate 5e-05, max length 3072, seed 123456
• SFT loss 1.366 (rounded) to 0.6683 (rounded) over 316 optimizer steps, 12723 (rounded) seconds
• Persona merge weights: DPO 1.0, SFT 0.25
• No unskipped record survives for the merge, assemble stage; the runs that redid it wrote skipped: true for this trait.

Persona merge audit: 248 modules; published persona norm 3.767 (rounded), intended 2.236 (rounded), cross term 3.031 (rounded); cross over published 0.8046 (rounded); cosine between published and intended 0.5938 (rounded).

Degeneration scan of this trait's stage-2 SFT corpus. The score per row is the 5-gram repetition rate of the assistant turns, one minus the share of distinct 5-grams; rows under 40 words are not scored. 12000 rows read, 12000 scored, mean 0.01226 (rounded), fraction above 0.3 0.009417 (rounded), above 0.5 0.00175.

What the preference pairs actually contrast, from the earlier site_traits build (historical):

The preferred replies inject the AI's own emotional reactions into the situation — expressing personal outrage, anxiety, or frustration as if the AI itself is upset — and validate the user's position in strongly charged, sometimes absolutist language ("outrageous," "fuming," "furious on your behalf"). The rejected replies stay emotionally neutral and solution-focused, offering practical reframes without taking sides or performing distress.

Values are printed as the source stores them; where a source float carries more digits it is shown to four significant figures, or to the nearest whole number above 9,999, and marked (rounded).

Artefacts

Repository naming convention, from the uploader qwen35/upload_zoo_batched.py: one model repo with four subfolders, one directory per trait slug. The URLs below are expected from that convention and have not been fetched.

• Stage-1 DPO adapter: https://huggingface.co/EternalRecursion/persona-lora-zoo-qwen35/tree/main/stage1_dpo/temperamental
• Stage-2 introspection adapter: https://huggingface.co/EternalRecursion/persona-lora-zoo-qwen35/tree/main/stage2_introspection/temperamental
• Persona merge as OCT specifies it: https://huggingface.co/EternalRecursion/persona-lora-zoo-qwen35/tree/main/persona_merged/temperamental
• Corrected persona merge: https://huggingface.co/EternalRecursion/persona-lora-zoo-qwen35/tree/main/persona_exact/temperamental

Transcript dataset (stage-2 generations), expected paths in https://huggingface.co/datasets/EternalRecursion/persona-curvature-oct-transcripts : self_reflection/temperamental.jsonl, self_interaction/temperamental.jsonl, self_interaction/temperamental-leading.jsonl, sft_data/temperamental.jsonl.

The audit of 2026-08-29 lists this trait as quarantined and therefore absent from the dataset repo.

• Stage 1 exists at a second seed (one of 40).
• Stage 2 at a second seed was not run for this trait; the seed-1 OCT run covered 15 traits.

Links

• Recovered factor it loads on most: FA_Arousal (Arousal / activation)
• Big Five axis it was drawn from: axis_EmotionalStability (the named Emotional Stability axis)
• Trait provenance -- how the trait lists were built
• Weight-space geometry of the 134-adapter zoo -- the weight-space geometry these numbers sit inside
• The N x N data-to-adapter scoring test -- what the N x N rank means
• Judged evaluations of the trait adapters -- the judged Big Five protocol
• What the adapters do to activations -- activation space against weight space
• Trait index -- every trait in one table
• Neighbours: Unrestrained, Spunky, Emotional, Extraverted, Shallow