Example reports

In each of these, an evaluator used the Rubicon plugin for Claude to answer a question from a set of documents. Open one and click any number to see the passages behind it.

Causal map

Influences on household health (QuIP, causal map)

The question
Which influences do households say changed their health, and how often do they credit the activities of "Organisation 1", compared with the other influences?
The interviews
18 households in rural villages, from an anonymised impact evaluation that used the Qualitative Impact Protocol (QuIP), a method developed at the University of Bath.

Open the report

Five questions, one combined report

Five questions on the same households, one combined report (QuIP)

The questions
From the paths by which households say "Organisation 1" changed their income and food, through who they credit for each kind of change and how women's and men's accounts differ, to a verdict on its contribution to household wellbeing, judged against the evaluator's rubric.
The interviews
18 households in rural villages, from an anonymised impact evaluation that used the Qualitative Impact Protocol (QuIP), a method developed at the University of Bath.

Open the report

Why the MECC training was not cascaded

The question
People were trained to pass on Making Every Contact Count (MECC) health training in their own organisations. What stopped some of them, and what helped the rest?
The interviews
13 people who attended the training in the North East of England, from a published study by Northumbria University and colleagues (Nichol et al. 2026, CC BY). The transcripts come from a study by Beth Nichol and colleagues at Northumbria University, who deposited them openly alongside their published paper. Few studies do this. It is excellent practice, and it is what makes an example like this possible.

Open the report

Sustainability, judged against a rubric

Sustainability of MECC training (rubric)

The question
As the 13 Making Every Contact Count (MECC) Train the Trainer attenders describe it, how likely is MECC delivery to continue in their organisations without further outside support?
The interviews
13 people who attended the training in the North East of England, from a published study by Northumbria University and colleagues (Nichol et al. 2026, CC BY). The transcripts come from a study by Beth Nichol and colleagues at Northumbria University, who deposited them openly alongside their published paper. Few studies do this. It is excellent practice, and it is what makes an example like this possible.

Open the report

Reluctance to use AI at work

The question
When people hold back from using AI at work, is it more because they fear for their jobs or because they doubt the quality of what it produces?
The interviews
18 professionals (creative workers, scientists and others), from the Anthropic Interviewer dataset, published under CC BY.

Open the report

People inside focus groups

How researchers see luck in their work (focus groups)

The question
How do the researchers in these four focus groups see luck in their own work?
The transcripts
Four anonymised focus groups with 17 researchers at Dutch universities, from a published dataset by Martin Sand and Karin Jongsma (Focus Group Interviews "Luck in Science", 4TU.ResearchData 2020, CC BY 4.0).

Open the report

How to test an evaluation workflow before even seeing the documents

Testing a workflow on made-up documents

The test
An invented home energy advice scheme. Does the analysis rate the success set Good enough and the failure set Not good enough?
The result
Yes: 4 of 4 households met the standard in the success set, and 1 of 4 in the failure set. The report below is on the success set; here is the one on the failure set, and the analysis as a file. Every interview is made up, so what the reports say about the scheme does not matter.

Open the report