A group at Stockholm University believes it has reached a definitive answer to one of physical chemistry's longest-running questions: whether liquid water exists as a mixture of two distinct states of matter. The claim, if it holds, would close a line of inquiry that has run for decades and could reshape how researchers model solvent behaviour across materials science, biochemistry and pharmaceutical formulation.

The context is concrete. Water deviates from the behaviour of most liquids on at least 66 measured properties, including higher-than-expected melting and boiling points, elevated surface tension and unusual heat capacity. For most liquids, these parameters follow predictable scaling relationships. Water breaks them. The two-state hypothesis — that liquid water dynamically interconverts between two structurally distinct forms — has stood as one candidate explanation, but no definitive experimental resolution has been accepted by the field. The Stockholm team says it may now have that resolution. Researchers evaluating the result will want to scrutinise the measurement conditions, since the hypothesised liquid-liquid transition in supercooled water sits in a regime notoriously difficult to access without crystallisation, and claims of "definitive" answers in this area have been contested before.

The item comes from a podcast segment with Philip Robinson and Emma Pewsey, which pairs the water story with a second topic with direct operational relevance for R&D managers: research misconduct and the volunteer networks now detecting it.

AI lowers the cost of fabrication

The misconduct discussion lands as generative tools make falsifying publications easier than at any previous point in the research record's history. The problem reaches the most prestigious institutions, with recent high-profile cases emerging from Stanford University and the Dana-Farber Cancer Institute. These are not fringe venues; they are institutions whose output informs clinical portfolios and hiring decisions across the sector.

Behind many of the exposed cases sits a growing network of self-taught "science sleuths" — researchers and outsiders who systematically check published figures, statistics and images for manipulation. The podcast emphasises that anyone can join this effort, and that experienced sleuths are actively sharing methods for spotting fraud and for handling it responsibly once found. For journal editors, institutional integrity officers and R&D leaders who depend on the literature for pipeline decisions, this distributed scrutiny functions as an unregulated but increasingly consequential layer of post-publication quality control.

A practical toolkit

The segment closes with a resource list that labs and integrity teams can adopt directly:

• Open science integrity guides: the Cosig collection, for best-practice reference.
• Post-publication peer review: PubPeer, the platform for hosted critique of published papers.
• Image-manipulation detection software: ImageTwin and Proofig, tools that flag duplicated or altered gel images and micrographs — a class of fabrication that has figured prominently in recent retractions.
• Problem-paper tracking: the Problematic Paper Screener and the Retraction Watch Database, which aggregate identified problem papers.

For research managers, the tools carry different cost-benefit profiles. Proofig and ImageTwin can be built into pre-submission and editorial workflows as a preventive check, while PubPeer and the Retraction Watch Database serve due-diligence functions — screening collaborators, hires or cited foundations of a project against the accumulating record of flagged work.

What remains unmeasured

The podcast is a discussion format rather than a peer-reviewed report, and neither segment arrives with sample sizes, funding disclosures or method details attached. The Stockholm water result in particular warrants tracking to its primary publication before any modelling or simulation group treats the two-state question as settled; anomalous-liquid claims in this field have a history of method-dependent reversals. On the misconduct side, the trend lines are nonetheless unambiguous: detection capacity is scaling faster than institutional review, and the Stanford and Dana-Farber cases suggest that brand-name institutional affiliation offers no reliable proxy for integrity.

Both threads point in the same direction for the coming year: the scientific record is becoming more contested and more actively policed from below, and groups that build verification tools and retraction-awareness into their literature workflows now will spend less time unwinding dependencies on bad papers later.