Two new studies pull at the seams of assumed knowledge — one in a medieval Scandinavian grave, the other in a quantum material

Two peer-reviewed papers published this week are quietly dismantling two long-held assumptions — one archaeological, one computational — and each does so by leaning harder on data rather than on received wisdom. The first, from Stockholm University and out in Science Advances on 10 July, shows that adults and children buried together in medieval Scandinavian graves were not always biologically related. The second, from the University of Manchester, accelerates the hunt for two-dimensional materials with exotic quantum behaviour by teaching a machine-learning model to respect the underlying physics.

Taken individually, either result is the kind of incremental correction that earns a footnote in a textbook. Read together, they mark a methodological shift: in both fields, the easier question has been displaced by the harder one. Researchers are no longer asking what a grave looks like or which candidate materials might work. They are asking, with genomic and computational rigour, what is actually true.

What the grave really holds

For decades, the working assumption in medieval archaeology was straightforward: when an adult and a child share a grave, they are likely to be parent and offspring, or at least kin. Intuition backed the inference. Shared burial was treated as evidence of social bond, and social bond was treated as evidence of blood.

The Stockholm-led team set out to test that assumption directly. Using ancient-DNA analysis on skeletal remains from multiple medieval Scandinavian burial sites, the researchers compared the genetic relationships between co-buried individuals against the relationships implied by their spatial arrangement in the ground. The published finding — that co-burial frequently fails to map onto biological kinship — is more than a curiosity. It forces a re-reading of dozens of interpretive frameworks that have used grave composition as a proxy for household, lineage and inheritance structures in the early medieval North.

The structural implication is uncomfortable for the field. If shared burial does not reliably signal family, then a generation of work that built social-history models on grave pairs and grave clusters needs to be revisited. The authors are careful not to overclaim — they do not argue that no medieval co-burials were familial, only that kinship cannot be inferred from proximity alone. The data does the talking.

Teaching a model the rules of physics

The second paper, published in Science Advances on 9 July by researchers at the University of Manchester, takes a different kind of shortcut and removes it. Two-dimensional materials — sheets only a few atoms thick — are a hunting ground for quantum phenomena such as superconductivity, topological states and exotic magnetism. The problem is that the candidate space is enormous, and screening each compound experimentally is slow and expensive.

Standard machine-learning approaches speed up the search by predicting properties from chemical composition or crystal structure. They are useful, but they tend to be black boxes: the model learns correlations that may or may not reflect the underlying physics, and its predictions can fail in regimes where the training data is thin. The Manchester group instead built a physics-informed machine-learning method. The model is constrained to respect the symmetry properties and physical laws that govern the materials it evaluates, which means its predictions are anchored in the same mathematics that a theorist would use by hand.

The result, by the team's account, is a faster path to identifying two-dimensional materials with unconventional quantum behaviour — the kind of compounds that could eventually underpin new electronics, sensors or quantum devices. The structural lesson is broader. Across computational science, models that learn from data alone have begun to give way to models that learn from data and from the established rules of the discipline they are modelling. The hybrid is harder to build, but it travels further from its training set.

What the two papers have in common

The pairing is suggestive. In the archaeology paper, a long-standing assumption — that burial proximity equals kinship — is treated not as background knowledge but as a hypothesis to test. In the materials paper, a long-standing bottleneck — that screening exotic materials is slow — is attacked not by throwing more compute at it but by building a model that knows the physics.

Both moves share a methodological temperament: trust the data more than the default story. Neither paper reaches for grand theoretical claims; both earn their conclusions by being specific about what was measured and what was not. That is also why the results are likely to age well — they are tied to sequences and to computations that other groups can rerun, not to a single charismatic interpretation.

The counter-narrative here is that both results are, in a sense, negative findings. Co-burial is not a reliable proxy for family; machine-learning-only screening is not sufficient for quantum-materials discovery. Negative findings tend to attract less attention than positive ones, and both teams have done the harder work of turning a null into a publishable contribution. The framing matters because it tells readers — and funders — that the value of a study is not always in what it adds, but sometimes in what it removes from the pile of things we thought we knew.

Stakes and what to watch

For medievalists, the immediate task is to revisit site reports and reinterpret the social meaning of co-burial, particularly in regions where grave composition has been used to argue for household continuity, fosterage or adoption. Genetic reanalysis of archived collections is now a clear priority; the same logic that overturned the burial-assumption may overturn others. For the materials community, the Manchester method opens a workflow that other groups can adapt — feed in known physics, search the candidate space faster, and prioritise the compounds that warrant synthesis and bench-testing.

The wider signal is that both fields are entering a phase where cheap computational tools and falling genomics costs make old assumptions pay a higher price. The pattern is familiar across science: as instruments improve, the cost of testing an inheritance drops, and the inheritance is more often than not found wanting.

Desk note: this publication framed both papers as a methodological pairing — assumptions tested rather than assumed — rather than as two unrelated science-of-the-week items.