The replication of research results is a linchpin of the scientific process. But it isn't always done the way a layperson might expect — by replicating a previous experiment, step-by-step, in a different lab. Instead, validation of research results can take different forms in different disciplines. And as we report on page 344, changes in scientific publishing could herald a new era of open, interactive communication in which research findings are tested.

The practice of replication is already more diverse, and less clean-cut, than is commonly realized. Funding agencies won't normally pay for the direct repetition of published work. In some disciplines, fresh work that builds on a published result will involve fully replicating the work that led to that result. In others, including many branches of biology, scientists see the process as validation, rather than replication. The expectation isn't necessarily that another scientist will reproduce exactly what they did, but that the data and methodology are strong enough to withstand inferences and have new experiments built upon them.

Whereas outside observers may think that all science can, and should, be reproduced, most scientists aim to confirm and extend published conclusions. That is a significantly different objective, and may be accomplished most compellingly using a different experimental approach, in pursuit of the answer to a related but separate question. This allows investigators to get around the practical difficulties of replication, such as the fact that materials or animals are never truly identical. Because of these difficulties, failed attempts to perform an exact replication of the data from a multivariable experimental system do not automatically invalidate the conclusions of the original study.

However, the interrogation of previously published results remains an essential part of science. Too often in the past, failure to achieve successful replication has left unsatisfactory research in a kind of helpless limbo. Such non-replication may be the subject of late-night arguments at scientific meetings, but is unlikely ever to be published, or even submitted for publication. Indeed, in many disciplines, it is difficult to raise concerns about an initial result in a public forum without this being construed as a full-frontal assault on the original authors' integrity.

But publishing is changing, and the idea of a scientific paper as a self-contained and unalterable body of work is gradually becoming obsolete. In some disciplines (physicists have led the way, through arXiv), scientific results are already subject to what amounts to an open-ended peer-review process, in which colleagues can continue to criticize and discuss the contents of a paper, for as long as it takes. That, rather than the current, sometimes awkward silence, is the way ahead.