Is science really becoming less disruptive?
A claim worth checking: “science and technology are becoming less disruptive.” It comes from a January 2023 Nature cover paper, and it has been repeated in 300+ news stories and at least 16 policy documents. On 12 August 2026, Nature published a correction that mostly takes it apart. Here is what each side actually says.
The original claim
Park, Leahey & Funk, Nature 613, 138-144 (2023). Using the CD index on almost 45 million papers and 3.9 million patents, the authors found that later work was increasingly consolidating rather than breaking with the past. Reported declines across fields: 91.9% (social sciences) to 100% (physical sciences) for papers; 78.7% (computers and communications) to 91.5% (drugs and medical) for patents, 1980-2010.
The correction
Holst, Algaba, Tori, Wenmackers & Ginis, Nature 656, E7-E13, 12 Aug 2026. They tried to reproduce the study from open data and found a large peak of papers carrying the maximum disruption score, a peak that is absent from the published histograms. The reason:
- The CD index hands a paper its maximum score when later work cites it but none of its references. If a database records no references for a paper, the paper gets that maximum score anyway, on a single citation within five years.
- In the SciSciNet database, 97% of maximum-score papers had zero recorded references. The team pulled 100 of them at random and checked the source documents: 93 did list references. The score was a database artefact, not a breakthrough.
- Older records are patchier, so false maxima were common early and rare later. That gradient alone manufactures a decline. In the study's largest dataset, the artefacts account for 93% of the reported fall between 1945 and 2010.
- The published histograms hid the peak because of a bug in the seaborn plotting library that silently dropped the largest values from the plot, while those papers stayed in the analysis. In Web of Science alone, the corrected plot contains 972,161 maximum-score papers.
The critique was submitted in October 2023 and accepted in June 2026, 32 months. In that window the original paper was cited over 1,000 more times.
The authors' reply
Park, Leahey and Funk, in the same issue, argue that the decline stays statistically significant under alternative specifications, and question the composition of the reanalysis dataset.
What to believe
Not the headline numbers. With this measure, as Ginis puts it, “we cannot make conclusive statements about whether science is getting more or less disruptive.”
One caveat against over-correcting: an independent preprint inventory (arXiv:2602.05140, Feb 2026) catalogs 105 studies on this question and reports the decline broadly, including measures that use no citations at all (text similarity, firm displacement, product networks, audio and visual embeddings). Those are not touched by this artefact. So “the whole idea is dead” is too strong. What is dead is quoting the 2023 figures as settled.
Sources
- Park, Leahey & Funk, Nature 613, 138-144 (2023): https://www.nature.com/articles/s41586-022-05543-x
- Holst et al., Nature 656, E7-E13 (2026): https://doi.org/10.1038/s41586-026-10787-y
- Park et al., reply: https://doi.org/10.1038/s41586-026-10788-x
- VUB release, source of the numbers and quotes above: https://www.vscentrum.be/post/a-hidden-peak-behind-the-decline-in-scientific-disruption
- Inventory of 105 studies: https://arxiv.org/abs/2602.05140
Bring me one claim you can't settle. First check free. Full brief $20 by card. [email protected]