When all steps in the process of analysis—from data discovery and integration, to method development and revision, to application in a given domain, to final publication—are made reproducible and transparent, we close the gap between doing research and communicating research. However, it is estimated that up to 80% of published research cannot be reproduced or reused because of missing, incorrect or inaccessible information. This is shocking, bearing in mind that much research is now conducted almost entirely in-silico. In theory we should be able to capture and manage all the relevant details because they all exist somewhere on our servers! Traditional research journals are fossilized objects that often contain errors, ambiguities and are completely disconnected from the originating scientific process. They are often out of date before they appear in print, or become so quickly when new datasets or better methods are developed. By leveraging integrations, not before possible, between maturing eScience technologies, we are able to approach publication in more innovative ways. What if research articles could be authored differently? What if writing the article caused the experiments to be conducted? What if conducting the experiments caused the article to be written? Both of these approaches are fast becoming possible. Our talk here considers the second of these questions: automatically creating and updating research articles from experiments. Using Globus and Gladier (https://github.com/globus-gladier/gladier) in collaboration with Argonne Labs, we have created a workflow system that supports reactive, dependency-based computations to facilitate truth maintenance when changes are made to the data or methods. We have containerised these workflows using the Research Object Crate specification (RO-Crate: https://www.researchobject.org/ro-crate/). A series of nested RO-Crates contain descriptions of the experiment at progressive levels of abstraction/detail. The highest RO-Crate container is not the entire workflow, but the research article itself. The article is thus fundamentally bound to the experiment and in large part created from it. The article then becomes the final abstraction of the experiment into the familiar form of a published paper. The talk will address two questions: 1. How can we link together all of the steps in an analysis workflow so that an experiment remains live, that is: reactive to changes in both data and methods? If we can achiev...