Julie’s web version: a first test against a published guideline
Julie is getting a web version, which we are building with a highly respected software development company. This week its first test-ready version faced a test with a published answer: the two guideline recommendations from our last article, with full-text papers and live AI models.
The aim is that a researcher can carry a review through in a browser, with the decisions still in human hands. The first test-ready version is now far enough along to be tried on real papers, so we gave it the test from our last article: two recommendations a published guideline graded differently, and the papers that go with them. It went from title and abstract screening through the full-text check and extraction and rating to linking evidence to recommendations, and we set the outcome beside Julie’s local version, the one behind the results reported here.
This item reports that the run took place, not how it performed. A handful of papers shows whether a chain runs, not how well, and the web version is still changing, so we give no figures here. An AI agent worked through the screens the way a researcher would, and the human decisions were simulated and marked as such. The manuscript draft was not part of this run. Everything else on this site, capabilities and results alike, describes the local version, not the web version.
A first test-ready version is there to be tested. What we found goes back to the development team. When there is something to measure, we will report it as we report everything here: what was measured, on what, and what it does not show.
While the web version takes shape, this is what we most want to hear. Tell us about your project, and what you would expect from Julie: what she should do for you, and how reliable she would need to be.
All news