Science publishing is undergoing a quiet revolution: on May 27, 2026, *Nature* published an editorial announcing that Registered Reports — publish protocols first, regardless of outcome — will extend to all fields in which Nature publishes.
This is not a small-scale pilot of a new concept. It is a paradigm migration at the very top of the journal hierarchy.
How bad was the old system?
In 2015, the American Psychological Association estimated that for every published paper, roughly five completed studies went unpublished — shelved in drawers because their results were negative or non-significant.
The consequence: the "scientific consensus" the public sees is a heavily filtered survivorship bias. Ineffective drugs, unsupported theories, and null interventions simply vanish. To get published, researchers either only pursue "easy wins" or keep tweaking analyses until P < 0.05 — the infamous p-hacking.
Nature's solution, in three steps
1. Stage 1: Researchers submit a study plan — research question, methods, analysis plan. Reviewers judge whether the question matters and the methods are rigorous. 2. In-principle acceptance: If approved, Nature commits to publishing — regardless of positive, negative, or null results. 3. Stage 2: Researchers execute the plan and submit results. Reviewers only check whether the plan was followed — no longer whether results are "interesting."
This flips the incentive structure entirely: from outcome-oriented to process-oriented.
Under the traditional model, the optimal strategy was: assume a likely-true hypothesis, pick an easy experimental design, iterate analyses until significance, and bury the study if it fails. Registered Reports breaks that chain. Researchers can test bold hypotheses (null results still publish), take on high-cost/high-risk projects (the publication commitment removes the fear of wasted effort), and reviewers become design consultants rather than outcome judges.
A real case: William Brady's Bluesky experiment
Social psychologist William Brady (Northwestern University) asked: do social media algorithms amplify emotional, toxic, politicized, and moralized content — and does that distort users' perceptions?
During the 2024 US presidential election, his team ran an eight-week experiment on Bluesky, manipulating the feeds of 2,000 users with three different algorithms. The cost: hiring a software engineer, running the project like a startup for a full year, with no guaranteed findings. Under the traditional model, such a project would be nearly impossible to justify — but as a Registered Report, Nature committed to publication either way. Brady called it a "no-brainer."
Reviewer feedback *before* the experiment improved the design: the team used existing Bluesky users rather than recruited volunteers, yielding responses from genuinely invested users.
The findings:
- Engagement-optimized algorithms did amplify moralized, emotional, toxic, and politicized content.
- But there are ways to reduce exposure to divisive content without making platforms less fun.
- A null result: inflammatory posts did not change user behavior by significantly boosting engagement with such content. Brady noted this negative finding would likely never have been published without the Registered Report format.
- All fields welcome — from physics and biology to medicine and earth science.
- Beyond hypothesis testing — large-scale data collection, method comparison, and other exploratory work now qualify.
- Two-stage review, double workload. Stage 1 reviews the protocol, Stage 2 the results — each potentially with multiple revision rounds.
- Locked analysis plans can feel stifling. Brady's team, for instance, found far fewer "toxic posters" than expected, making some data ill-suited to the original methods. Nature's new guidance permits exploratory analyses under strict conditions: clearly labeled, justified, reported separately from main results, and never the basis of the paper's conclusions.
- Slower publication. For tenure-track early-career researchers who need fast output, this is a real cost.
- Submission strategy: a Registered Report at Nature means longer review but a publication commitment — suited to high-cost, high-risk experiments.
- Grant applications: the in-principle acceptance can serve as evidence of feasibility.
- Training: graduate students need pre-registration methodology — writing analysis plans, handling unexpected data.
- Evaluation reform: if Registered Reports go mainstream, should tenure assessment reward high-quality negative results over "significant" ones?
- Nature Editorial: *Nature is expanding Registered Reports to all the fields in which we publish* (2026-05-27) — https://www.nature.com/articles/d41586-026-01629-y
- Brady et al., *Redesigning algorithms to intervene on social norm misperceptions during a national election* (2026) — https://doi.org/10.1038/s41586-026-10536-1
- Gorman & Hubbard (2025), *Quantitative Science Studies*, 6, 611–622
The 2026 expansion
Nature began accepting Registered Reports in 2023, but only for confirmatory research in cognitive neuroscience and the behavioral/social sciences. The new policy means:
As the editorial puts it: the importance of the question, the quality of data collection, and the rigor of planned analysis should not be confined to confirmatory research.
Not a utopia: the costs
The deeper signal: redefining "good research"
The core shift is not a formatting change but a change in what the scientific community values:
| Traditional values | Registered Reports values | |---|---| | Significant results = good research | Important question = good research | | Novel, counterintuitive = high impact | Rigorous methods = credible | | Negative results = failure | All results = knowledge increment |
If successful, this changes daily scientific behavior: replication studies become publishable rather than career suicide; hypotheses face stricter testing since falsification can be published; and transparency improves because protocols are public and locked.