CloudArmy Reactor — Best Practices

Best Practices

How we think good research should be done — the principles behind the platform, stated plainly enough to argue with.

All

Research Best Practices

Compliance

Data Quality

Inclusive Design

Behavioural Science

Research Design

Data & Confidentiality

Research Best Practices

12 min read

What Is a Product Worth? Why Context Matters in Price Research

A stated price is not a context-free measure of value. Good price research considers the reference point, framing and decision context behind the answer.

Reference prices, framing and ownership can all change how consumers evaluate price. Understanding these effects can help researchers design better studies and interpret willingness to pay more carefully.

Compliance

9 min read

Why Reactor Is Not a High-Risk AI System

AI Compliance Is an Architecture Decision, Not a Legal Opinion

The EU AI Act sorts AI by the risk of its use, not by its technology. Reactor sits outside the high-risk tier because of three deliberate design choices: AI never administers a test or computes a result, insights describe populations rather than individuals, and where AI does contribute the output is marked as such — visibly for a reader, and machine-readably for whatever consumes it next......

Data Quality

5 min read

Why Self-Service Research Depends on Proactive Data Quality

Quality assurance should shape data collection, not simply clean up after it. In self-service research, every automated insight depends on the quality of the data collected before analysis begins.

Self-service research promises fast answers. But fast answers are only valuable if the underlying data can be trusted.

Inclusive Design

6 min read

Accessible by Design: Research Everyone Can Take Part In

Build the respondent and analysis experience so it works without a mouse, without sight, and without sound. Hold it to a standard others can check, and state plainly where the protections stop.

Accessibility is usually framed as compliance. In research it is something more practical: a study that excludes people with disabilities quietly skews its own sample. Building for everyone is how you keep the data honest, and it is a discipline worth naming.

Behavioural Science

3 min read

The Say/Do Gap: Why Stated Preference Isn’t the Whole Story

Treat stated preference as one source of evidence, not the final answer. Where behaviour matters, measure behaviour directly whenever possible.

What people say matters. What people do matters too. The two do not always align. Understanding that gap is often the difference between measuring an opinion and understanding a choice.

Behavioural Science

3 min read

The Answer Is Only Half the Data

Treat response time as data, not noise. The speed of a response can provide information that the answer alone cannot.

Most research records what people chose. Behavioural methods can capture something else as well: how quickly that choice was made. Sometimes the answer is informative. Sometimes the time taken to reach it is informative too.

Research Design

2 min read

Start From the Decision, Not the Questionnaire

Define the decision first, then design backward. The question picks the method; the method should never pick the question.

Research is most valuable when it helps you make a choice. Start with the decision you need to make, then work backward to the evidence required to support it.

Research Design

2 min read

Pair Implicit With Explicit

Combine implicit and explicit measures. The reaction-time signal gives you the instinct; the stated answer gives you the rationale. Read them together, not in isolation.

Different methods provide different forms of evidence. The strongest studies use them together.

Data & Confidentiality

7 min read

Obscure First: Pseudonymising Confidential Information Before You Analyse It

Pseudonymise confidential information before analysis, not after. Obscure identities at capture so the sensitive form is never the stored or shared form — and keep the key that re-identifies people separate from the data itself.

In behavioural research the open-ends and respondent records carry the real confidentiality risk. The best practice is to obscure identities at the point of capture — so the sensitive form is never the form you store, share, or feed to a model.