Hypothesis exploration
Map plausible mechanisms and identify what future human data must distinguish.
- Competing explanations
- Boundary scenarios
- Evidence-gap map
Research & synthetic personas
Do not validate a persona because it “sounds human”.
Test whether the simulated behaviour represents the constructs, population differences, constraints and trajectories relevant to your research question.
Synthetic personas support hypothesis exploration and study design. They are not human participants or population estimates.
Where it fits
Use simulation to expose assumptions before they become conclusions.
The track is strongest when a research team already has a question, theoretical frame and a reason to compare behaviour across conditions.
Instrument and scenario stress
Probe interview questions, survey items, tasks or interventions for ambiguity and context sensitivity.
- Question interpretation
- Response-process assumptions
- Adversarial scenarios
Population-condition comparison
Test whether constrained persona definitions produce interpretable differences without erasing dispersion.
- Within-profile consistency
- Between-condition differences
- Invariance/DIF signals
Agent-based simulation
Observe coordination, conflict, feedback, adaptation and emergence across agents and time.
- Interaction rules
- Memory trajectories
- System-level outcomes
Research protocol
Seven steps from construct to defensible limitation.
Every result remains linked to the persona specification, scenario, run, comparison and external reference that produced it.
Frame the inference
State the research question, intended use and what the simulation must not be used to infer.
Map observable indicators
Connect theory to behaviours, tasks, rubrics and plausible counterexamples.
Specify without caricature
Ground groups in theory or evidence; record resources, constraints and uncertainty separately from identity labels.
Control the conditions
Define environment, task, information, interaction, feedback, memory and event phase.
Compare distributions
Run repeated scenarios and inspect coherence, dispersion, subgroup gaps and trajectory differences.
Use an external reference
Contrast with literature, anonymised human data, expert rubrics or a documented Delphi cycle where available.
State the evidence boundary
Separate observed system behaviour from hypotheses about real people and identify the next required study.
Psychometric lens
What we measure—and how carefully we say it.
Does the behaviour represent the idea?
A fluent response is weak evidence if the tasks do not distinguish the claimed mechanism from a shortcut.
Does the pattern survive repetition?
We report variability and failure frequency rather than selecting a convenient run.
Are comparisons interpretable?
Signals are investigated across comparable conditions; they are not labelled “bias” without further evidence.
Does behaviour change for a reason?
The protocol distinguishes appropriate response to context from instability or prompt compliance.
Does the persona preserve trajectory?
Prior information, feedback and accumulated experience should have documented effects.
Who can challenge the interpretation?
We record reference evidence, expert agreement, disagreement and unresolved uncertainty.
Deliverables
A study-ready evidence package.
- Construct and inference map
- Versioned persona and condition protocol
- Reproducible scenario set and runner
- Within-profile and between-condition analysis
- Memory and trajectory evidence
- Expert-reference log when included
- Limits of inference and next-study recommendations
Required inputs
Bring the question before the characters.
- One explicit research question
- Theory, literature or prior evidence
- Intended personas, groups or conditions
- What real-world inference is being considered
- Available independent reference
- Ethics or governance constraints
No personal data in the public form
If anonymised participant data becomes necessary, processing roles and an approved transfer channel are agreed first.
Appropriate uses
- Explore hypotheses before fieldwork
- Stress-test a study or instrument
- Compare explicit model assumptions
- Identify where human evidence is essential
- Test an ABM implementation
Not supported
- Replacing a representative human sample
- Claiming demographic truth from an LLM
- Psychological diagnosis or profiling
- Skipping consent or ethics requirements
- Presenting a synthetic population as validated people
Have a research question and a working simulation?
Make the assumptions testable.
We will identify the smallest defensible protocol, the evidence you already have and the human validation that remains necessary.