README.md

StrataSynth

StrataSynth generates synthetic datasets with structured psychological and social dynamics.

Unlike traditional LLM-generated dialogue datasets, StrataSynth models internal cognitive state before generating language.

Our engine separates cognition from language:

PsycheGraph → identity model
Belief Engine → evolving beliefs
Relationship State → trust, tension, connection
Decision Engine → conversational intent
LLM Rendering → natural language

This produces datasets with explicit belief evolution and relationship trajectories.

---

Datasets

We publish datasets focused on:

• belief dynamics
• social reasoning
• conversational conflict
• agent stress testing

Current datasets:

• stratasynth-belief-dynamics
• stratasynth-social-reasoning
• stratasynth-agent-stress-test

---

Why structured synthetic data?

Most dialogue datasets only contain text.

StrataSynth datasets include:

• belief_state
• belief_updates
• relationship_state
• intent
• communication_act
• conversation trajectory

This allows research in:

• AI alignment
• social reasoning models
• agent evaluation
• belief tracking
• multi-agent systems

---

Links

Website
https://stratasynth.com

Platform
https://app.stratasynth.com

---

Example turn

Speaker A
"I’m not upset about the meeting. I’m upset that you didn’t tell me earlier."

intent: reveal
goal: seek_validation

belief_update: trust_other: -0.07

relationship_state: trust: 0.62
tension: 0.44