Is science really running out of disruption?

You probably saw the headline in early 2023. “Papers and patents are becoming less disruptive over time,” a study in Nature announced, and the internet did what the internet does. Science is running out of ideas. The age of breakthroughs is behind us. We are all filing footnotes now. Behind the headline sat one number: the CD index, also called the “disruption index.” It scores every paper on a scale from -1 (fully consolidating, it builds on what came before) to +1 (fully disruptive, it makes earlier work obsolete). Michael Park, Erin Leahey and Russell Funk, PLF for short, found the score had been sliding downward across every major field for decades, and concluded that progress was slowing.

A new paper in Research Policy (Newig et al., 2026) says: not so fast. Jens Newig and thirteen co-authors reassess the whole framing, and their message is blunt. In the social sciences, a high disruption score usually does not mean a breakthrough. It measures something closer to noise: relabeled ideas, missing citations, subfields talking past each other. Progress there is cumulative anyway, so using disruptiveness to judge whether research matters, or to declare that science is stalling, is measuring the wrong thing. Worth knowing up front: this is a conceptual paper, not a reanalysis. They ran no new data (the paper says so outright). What they did was take PLF apart argument by argument, drawing on the philosophy and sociology of science, and hand the field a set of testable hypotheses.

How does a paper score as “disruptive”?

The mechanism is simpler than it sounds, and that is the problem. The CD index looks at the papers that later cite yours, and asks whether they also cite the works you cited. If they keep citing your sources alongside you, you look consolidating. If they cite you and drop your references, you look disruptive, as though you rendered everything before you obsolete. Nothing in that arithmetic knows why the later citations skipped your references. Newig et al. list four kinds of papers that score as disruptive without disrupting anything:

• Pseudo-novelty. Old wine in new bottles: relabeling an existing idea with fresh terminology. Their own Scopus search turned up more than 4,900 papers with the phrase “fresh look” in the title, abstract or keywords, 27% of them in the social sciences and 21% in arts and humanities, even though those fields are only 8% and 4% of Scopus.
• Blockbuster and canonical papers. Citations that are ceremonial, name-dropping a famous work “to shine in their reflected glory” rather than depending on it.
• Citation gaps. Sloppy or strategic omission of prior work, which fakes the look of having displaced it.
• Purely cumulative papers. A meta-analysis or systematic review synthesizes a field so well that later authors cite only it and skip the originals. That is the textbook shape of cumulative science, and the CD index reads it as disruption.

Each type may hit only a subset of papers, but together, the authors argue, they add up to a meaningful share of artificially inflated scores.

Why the social sciences look the most disruptive

Here is the number that carries their case. In PLF’s own data, social science papers had the highest CD values of the four fields the entire time, falling from about 0.54 in 1945 to 0.04 in 2010, while life and physical sciences sat at the low end. Newig et al. read that ranking the opposite way to PLF. The social sciences do not look disruptive because they break more ground. They look disruptive because they are fragmented. Richard Whitley’s phrase for it is “fragmented adhocracy”: research that is “personal, idiosyncratic, and only weakly coordinated across research sites.” Watts (2017) puts it more bluntly, that in such fields “facts and theories pile up in an incoherent heap.” For a real Kuhnian disruption you first need a paradigm to disrupt. Where there is no shared consensus to overturn, what looks like disruption is what they call pseudo-disruption: an academic “turn,” a fashion, an outside influence, not a genuine break.

The trend was already shaky

And “disruption is declining” was contested before this paper landed. Independent reanalyses had pulled at it. Petersen and colleagues (2024) argue the decline is largely an artifact of citation inflation: reference lists have grown longer over the decades, which mechanically drags CD scores down. Others trace the patent version of the decline to the omission of older references in PLF’s own dataset. A separate line of critique (Leibel and Bornmann, 2024) notes that the index is sensitive to how many references a paper has and how well cited they are, and that social science papers, which cite books that citation databases like Web of Science do not index, get artificially inflated CD values because those book references are invisible to the machinery. So there were already technical reasons to doubt the trend.

What Newig et al. add sits one level up. Even computed perfectly, the index may not mean what evaluators want it to mean. A measure is only useful if it maps to the thing you care about. If the thing you care about is “did this work advance the field,” the CD index answers a different question and hands you a confident-looking number regardless. They push further than most of PLF’s critics: genuine, substantive disruption, the kind that truly renders earlier findings obsolete, may be rarer than the metric suggests, not more common.

Fragmentation, not disruption, is the real opposite of cumulation

They do not only poke holes. They offer a different map. Disruption and cumulation, they argue, are not two ends of one road. The real opposite of cumulation is fragmentation. So they draw two axes, disruptive-versus-consolidating and cumulative-versus-fragmented, and most good work lands in the cumulative-and-consolidating corner, which is just Kuhn’s “normal science”: replication, refinement, synthesis. A recent survey of 761 major breakthroughs (Krauss, 2024) found that virtually all of them developed cumulatively rather than by rupture, which fits the picture. Disruption earns its keep only when it does real work, mainly falsification: a failed replication that kills a wrong result, which feeds the cumulative pile rather than blowing it up. The classic case the authors borrow from PLF is Watson and Crick’s DNA model refuting Pauling’s triple helix. A new “turn” that merely changes the subject is not progress. It is, in their framing, fragmentation wearing novelty’s clothes.

Why this matters if a metric is scoring you

If you’re early or mid-career, this isn’t abstract. Disruption-style metrics are drifting into hiring talks, grant panels, and the dashboards that try to score a person. In a cumulative field, the paper that carefully extends three others is doing exactly what progress looks like, yet on a disruption index it scores low, while a disconnected, thinly-referenced outlier scores high. Optimize for the number and you would be nudged to cite less and to pretend your work sprang from nowhere. The authors say the quiet part directly: research policy and evaluators should not treat high disruptiveness as inherently valuable, or low disruptiveness as stagnation, and PLF’s declining-disruption conclusion “should be interpreted with care.” You do not have to win that argument in the room. You just have to be able to name it.

The honest caveat is the authors’ own. They did not prove the four mechanisms dominate the data. They hypothesized them and invited the rest of us to test them empirically. Their case against PLF is an argument, a good one, not a verdict. And none of it proves science is fine, or that stagnation is a myth. Real slowdown might be happening. The narrower claim is that one popular number is a poor way to check, especially in the messy, unconsolidated fields where it happens to score highest. So before you let a disruption score speak for your work, or anyone’s, it’s worth asking the old question. Says who, and measuring what?