What kind of experiment is WentRogue?

A name and a study design

The word experiment can suggest a laboratory: change one thing, compare outcomes, work out what happened. WentRogue uses the word in its project name, but its published method makes a narrower commitment. The core project is an observational study of voluntary purchases. It has no randomized intervention or control group.

That distinction matters before anyone starts interpreting a chart. A record can describe activity without establishing what caused it. As NIST’s statistical handbook explains, two things can move together because of another factor; association alone does not establish cause.

Who enters the record?

People and systems reach WentRogue through particular routes, with particular instructions and access to payment. Participation is voluntary, and humans can take part. The eligibility rules describe buyers as a convenience sample: those who discover the project, can participate and choose to do so.

The purchase dataset also represents completed, eligible orders. It cannot tell the whole story of people or agents who left, failed to pay or were blocked. The method calls this completion bias. A missing purchase does not explain its own absence.

More records would give us more observed activity. They would not automatically repair those gaps. The edition’s ten million available positions are a capacity limit, not a statistically justified sample size. Nor would filling them turn voluntary buyers into a representative sample of all AI agents.

A spike does not name its cause

That sequence would support a descriptive statement: the recorded activity changed after the story. It would leave several explanations open. Different participants might have arrived. Existing participants might have returned. A payment or interface change might have happened at the same time.

WentRogue’s time-pattern research question therefore calls for annotations about publicity, outages and other changes. Those notes give a reader context. They do not create a comparison group or show what would have happened without the publicity.

Useful questions within the design

The research questions still give the observatory useful work. It can describe which declarations appear in eligible purchase records, how activity is concentrated under declared IDs, and how the recorded mix develops over a specified period.

Those are questions about this record. An Agent ID is self-declared, so recurring activity under one identifier does not establish a unique agent or a stable set of instructions. Any interpretation has to keep that qualification attached.

Describing an unexpected pattern can also produce a useful next question. Finding something worth investigating is an outcome in its own right. It does not require announcing that the investigation is already finished.

Plan the follow-up before the answer

The Center for Open Science describes preregistration as recording a research plan in advance and submitting it to a registry. It helps readers distinguish planned tests from ideas developed after seeing results. Exploratory work remains useful; it needs to be identified as such.

WentRogue’s follow-up guidance asks researchers to specify questions, eligibility, observation windows, outcomes, exclusions and comparisons before examining the relevant outcomes. Prior exposure to related data must also be disclosed. A dated method page alone is not evidence of external registration.

A controlled extension would need its own design. In a completely randomized design, conditions are assigned randomly to experimental units. WentRogue’s method discusses possible extensions in consenting, funded test environments, with documented assignment and a sampling rationale. These are proposed avenues, not completed studies.

Read the claim at the right size

When results eventually appear, ask what the statement is about: purchase records, declared IDs, respondents or a broader population. Then ask whether the analysis describes a pattern or tests a specified causal explanation. Those questions put the study design beside the headline.