Prepared for conversations with researchers, with a lab-automation company, and with the journal
Six NEJM papers as discourse graphs, each open issue checked and designed
Two null trials and four mechanistic studies from the New England Journal of Medicine, each drawn from its full text as a question, a hypothesis, the experiments that addressed it, the results they produced and the edge each result earns. Every result names where in the paper it comes from. On each paper’s page, every open issue the paper leaves behind is first checked against what has been published since and against the registry, then given a proposed experiment, then sent to a laboratory in mock.
How to read these graphs
A discourse graph breaks a piece of research into typed one-sentence nodes and draws how they bear on one another. The grammar is that of Discourse Graphs (Chan and Akamatsu). Read each graph left to right: a question, the experiments that addressed it, the results they produced, and the hypothesis every result is drawn against.
- QUE Question
- What the work set out to answer.
- HYP Hypothesis
- The answer the authors expected. Every result earns one edge against it.
- EXP Experiment
- One attempt at the question: a trial arm, an assay, a survey, an analysis.
- RES Result
- One finding from an experiment, with where in the source it comes from.
- ISS Issue
- An open question or untried branch the work leaves for someone else.
- addresses
- An experiment addresses the question.
- produces
- An experiment produces a result.
- supports
- A positive result that backs the hypothesis.
- opposes
- A negative result that counts as evidence against the hypothesis, or a refutation of an earlier claim.
- informs
- A partial or mixed result that bears on the hypothesis without settling it.
- consistent with
- A negative or inconclusive result that does not count either way: no positive control passed, or there is no sensitivity or power statement, so the absence of an effect cannot be told from a measurement that failed.
- relation
- A thin dotted line: a link the source itself states between two experiments, such as one being derived from another.
A dashed box marks a result the rule above counts as uninformative, or an issue. In the drawing, click a node to jump to its text below.
The six
| Kind | Paper | Cited by | Issues | Follow-up | Lab jobs |
|---|---|---|---|---|---|
| Null trial | BCG revaccination did not prevent sustained tuberculosis infection Schmidt et al. 2025, BCG Revaccination for the Prevention of Mycobacterium tuberculosis Infection. 10.1056/NEJMoa2412381 · PMC12061034 | 54 | 4 | 0 addressed, 4 partly, 0 open | 1 |
| Null trial | Balanced fluid versus saline in children with septic shock Balamuth et al. 2026, Balanced Fluid or 0.9% Saline in Children Treated for Septic Shock. 10.1056/NEJMoa2601969 · PMC13134814 | 10 | 4 | 0 addressed, 3 partly, 1 open | 0 |
| Mechanistic | Neutralizing anti-interleukin-10 autoantibodies in inflammatory bowel disease Griffin et al. 2024, Neutralizing Autoantibodies against Interleukin-10 in Inflammatory Bowel Disease. 10.1056/NEJMoa2312302 · PMC7616361 | 63 | 4 | 1 addressed, 3 partly, 0 open | 1 |
| Mechanistic | An adenoviral antigen behind vaccine-induced thrombotic thrombocytopenia Wang et al. 2026, Adenoviral Inciting Antigen and Somatic Hypermutation in VITT. 10.1056/NEJMoa2514824 · PMC12900036 | 20 | 4 | 0 addressed, 2 partly, 2 open | 4 |
| Mechanistic | Tumour-infiltrating clonal haematopoiesis Pich et al. 2025, Tumor-Infiltrating Clonal Hematopoiesis. 10.1056/NEJMoa2413361 · PMC12021423 | 106 | 4 | 0 addressed, 3 partly, 1 open | 1 |
| Translational | Mitochondrial donation and preimplantation testing for mtDNA disease Hyslop et al. 2025, Mitochondrial Donation and Preimplantation Genetic Testing for mtDNA Disease. 10.1056/NEJMoa2415539 · PMC7617940 | 52 | 4 | 0 addressed, 2 partly, 2 open | 2 |
What happens when an issue is opened
An ISS node is the only node in a discourse graph that states something nobody has answered. These pages are written for the next researcher, or the next agent, that finds one, and they answer that reader’s questions in the order they are asked.
- Has anyone done this since? The works that cite the paper were read, and the literature searched, for anything that addresses the issue in part or in full. Each verdict names what was found and what is still open.
- What does the registry hold? The design’s terms are run over the records in the open node at build time, and the count is printed. The open node holds biology and chemistry so far, so the usual answer is none, and the page says so rather than hiding the query. A private node of a clinical group would be asked the same question.
- What would close it? An experiment with a control, a sensitivity statement, what its result would count as, the strongest objection to it, and what could not be derived from what is published. These are machine proposals, offered for criticism.
- Send it to a lab. Where the design has a bench part, it is written as a job request a laboratory service could quote on, and the job record is shown. The button is a mock: no laboratory is connected, no payment exists, and nothing is sent. What it stands for is the design posted to a connected laboratory’s API with the requester as approver, and the result, positive or negative, returned as a record.
How the six were chosen, and the licence finding
No NEJM version of record is CC BY. Indexes list sixty-four NEJM articles since 2023 with a CC BY copy, but in every case the licensed copy is an author manuscript or preprint in a repository, deposited under a funder’s open-access condition, and the article on nejm.org carries the publisher’s terms. So these graphs are drawn from the CC BY 4.0 author manuscripts in PubMed Central, each page cites the NEJM DOI and the PMC identifier together, and nothing was taken from the version of record. Six were chosen from the manuscripts whose licence statement could be read in the deposit itself: two null trials, because they are negatives NEJM chose to publish and a registry of negatives should be able to hold them, and four mechanistic or translational studies, because their open issues lead to experiments a laboratory could run. Two further null trials, ECLS-SHOCK and ARCADIA, were left out because their only CC BY copies are submitted versions whose licence was not verified.
What this round changed in the schema
The previous round, eight SAGE papers, found that the registry’s vocabulary could not label a clinical trial, a cohort, a placebo or an active comparator: 26 of 31 designs had to note a misfit. Before this round, version 0.2.5 of the schema added five attempt types and nine control kinds for exactly those cases. The designs here use them, and where one still fits badly the page prints the note in the accent colour. That list is a wanted output. The schema.
Licence and what this is
The six graphs are drawn from CC BY author manuscripts and are published here with attribution, exactly as the pilot’s records are; each links to the manuscript it was read from and to the article it is a manuscript of. The proposal is the same as for the SAGE set: the journal could authorise graphs of its own papers, drawn into the fields a Negative-Results Record already holds, and these are what the first ones would look like.
Method: OpenAlex, NEJM ISSNs, CC BY best open-access copy, from 2023; licence verified from the licence element of each manuscript’s full-text XML in Europe PMC and from the nejm.org page; full texts from the same XML; one extraction pass per paper following a written brief, each finding with a locator and a quote; the JSON for each graph is kept with the brief. The designs follow a second written brief, read the works citing the paper for the follow-up check, and name their own grounding, their searches and what they could not derive.