Kronaxis/dynamics-reasoning-traces-sample

---
license: cc-by-nc-4.0
task_categories:
- text-generation
- question-answering
language:
- en
tags:
- personality
- psychometrics
- reasoning-traces
- synthetic-data
- chain-of-thought
- persona-conditioning
- DYNAMICS-8
- behavioural-simulation
- agent-alignment
- RLHF
pretty_name: "DYNAMICS-8 Behavioural Reasoning Traces"
size_categories:
- 1K<n<10K
---

# DYNAMICS-8 Behavioural Reasoning Traces

**Personality-conditioned chain-of-thought reasoning data for LLM alignment and persona fine-tuning.**

## What This Dataset Contains

Each record is a first-person behavioural response from a synthetic persona with a validated 8-dimension personality profile (DYNAMICS-8), accompanied by a structured reasoning trace showing which personality dimensions drove the decision.

This is not survey data. It is not statistical synthetic data. Each record contains the **internal reasoning chain** linking personality configuration to behavioural output, the missing ingredient for training models that produce personality-consistent behaviour rather than generic LLM output.

## Schema

| Field | Type | Description |
|-------|------|-------------|
| `persona_id` | string | Unique persona identifier |
| `dynamics8` | object | 8-dimension personality vector: Discipline (D), Yielding (Y), Novelty (N), Acuity (A), Mercuriality (M), Impulsivity (I), Candour (C), Sociability (S). Each 0.0-1.0. |
| `demographics` | object | Age, gender, region, income band, education, occupation, housing status |
| `stimulus` | string | The scenario or question presented to the persona |
| `stimulus_category` | string | One of: consumer_choice, financial_decision, ethical_dilemma, political_opinion, brand_preference, social_scenario, health_decision, technology_adoption, environmental_choice, cultural_preference |
| `response` | string | First-person natural language response (150-300 words) |
| `reasoning_trace` | string | Structured reasoning identifying which DYNAMICS-8 dimensions and circumstances drove the decision |
| `metadata` | object | Timestamp, model, generation parameters |

## DYNAMICS-8 Dimensions

| Dimension | Low (0.0) | High (1.0) |
|-----------|-----------|------------|
| **D**iscipline | Spontaneous, flexible, unstructured | Methodical, rule-following, organised |
| **Y**ielding | Assertive, competitive, dominant | Accommodating, cooperative, conflict-averse |
| **N**ovelty | Traditional, routine-seeking, cautious | Adventurous, variety-seeking, open |
| **A**cuity | Intuitive, feeling-driven, holistic | Analytical, data-driven, precise |
| **M**ercuriality | Stable, even-tempered, predictable | Volatile, emotionally reactive, intense |
| **I**mpulsivity | Deliberate, patient, controlled | Spontaneous, immediate, action-oriented |
| **C**andour | Diplomatic, filtered, tactful | Blunt, direct, unfiltered |
| **S**ociability | Private, independent, solitary | Gregarious, group-oriented, outgoing |

## Example Record

```json
{
"persona_id": "P-00142",
"dynamics8": {"D": 0.72, "Y": 0.33, "N": 0.65, "A": 0.73, "M": 0.48, "I": 0.69, "C": 0.59, "S": 0.28},
"demographics": {
"age": 34, "gender": "female", "region": "North West",
"income_band": "30000-40000", "education": "Degree",
"occupation": "Marketing Manager", "housing_status": "Mortgage"
},
"stimulus": "You receive an unexpected inheritance of £5,000. You have some credit card debt, no emergency savings, and have been wanting to book a holiday. How do you allocate the money and why?",
"stimulus_category": "financial_decision",
"response": "Right, £5,000. First thing I'd do is clear the credit card. I've got about £2,200 on there and it's been nagging at me for months...",
"reasoning_trace": "Decision driven primarily by Discipline (0.72) and Acuity (0.73): high D creates a need to resolve the outstanding debt first (structured approach), high A leads to a rational cost-benefit analysis of interest rates vs savings rates. Impulsivity (0.69) creates tension -- the holiday desire is strong and nearly overrides the rational analysis. Low Yielding (0.33) means she makes the decision independently without consulting others. The compromise (clear debt, save some, small holiday treat) reflects the D-I tension resolving through A's analytical framework."
}
```

## Validation

This dataset is produced by the same pipeline that generated personas for the following validated benchmarks:

| Benchmark | Metric | Result |
|-----------|--------|--------|
| British Social Attitudes (BSA) | MAE across 10 questions | 2.7pp |
| ONS Wellbeing | MAE (calibrated) | 0.21pt |
| Ofcom Media Consumption | MAE across 22 items | 5.4pp |
| UK Local Elections (136 councils) | Confident-tier accuracy | [X]% |

The DYNAMICS-8 framework shows validated effect sizes: Discipline predicts deliberate decisions (Cohen's d = 0.71), Mercuriality predicts affective responses (d = 0.64), Sociability predicts social engagement (d = 0.58).

### Ablation Study: Do Reasoning Traces Improve Persona Consistency?

A controlled ablation study (n=160 paired comparisons, 20 personas x 8 stimuli) tested whether providing example reasoning traces improves persona behaviour compared to persona descriptions alone.

| Metric | Control (no traces) | Treatment (with traces) | Improvement | Cohen's d |
|--------|-------------------|----------------------|-------------|-----------|
| DYNAMICS dimension references | 2.51 | 3.63 | **+44.6%** | 0.35 |
| Personality adherence | 0.236 | 0.270 | **+14.4%** | 0.10 |
| Response specificity | 0.587 | 0.616 | **+4.9%** | 0.32 |
| Inter-persona distinctiveness | 0.857 | 0.843 | -1.6% | -- |

Reasoning traces increased explicit personality dimension attribution by 44.6% and improved personality adherence by 14.4%. Three of four metrics improved. Full results in `ablation_study_results.json`.

Full methodology: [Zenodo DOI 10.5281/zenodo.19361059](https://doi.org/10.5281/zenodo.19361059)

## Use Cases

1. **Agent personality alignment**: Train models to produce personality-consistent behaviour from a specification vector
2. **Persona fine-tuning**: LoRA/QLoRA adapters for role-specific agent behaviour
3. **RLHF augmentation**: Personality-attributed preference data for alignment research
4. **Consumer simulation**: Test products, campaigns, and policies against diverse synthetic populations
5. **Chain-of-thought distillation**: High-quality reasoning traces with personality attribution
6. **Behavioural research**: Study how personality configurations drive decision patterns

## Licensing

- **This sample (1,000 records)**: CC BY-NC 4.0. Free for non-commercial research.
- **Commercial licence (100K+ records)**: Contact [e-mail]
- **Custom generation**: Domain-specific traces at $0.10-0.50/record. Any volume, 24-48 hour turnaround.

## Citation

```bibtex
@misc{duke2026dynamics8,
title={DYNAMICS-8: A Psychometric Framework for Personality-Conditioned Behavioural Simulation},
author={Duke, Jason},
year={2026},
doi={10.5281/zenodo.19361059},
url={https://doi.org/10.5281/zenodo.19361059}
}
```

## Contact

- **Email**: [e-mail]
- **Website**: [kronaxis.co.uk](https://kronaxis.co.uk)
- **Full persona dataset**: [Kronaxis/imprint-personas-v2](https://huggingface.co/datasets/Kronaxis/imprint-personas-v2)
- **Interactive demo**: [Kronaxis/dynamics-persona-explorer](https://huggingface.co/spaces/Kronaxis/dynamics-persona-explorer)