Why Most Shared Scientific Resources Quietly Collapse Under Their Own Success

There is a version of this story that repeats itself across every subdiscipline in the life sciences. A group of researchers, often funded by a coordinating grant, builds something genuinely useful: a curated cell-line repository, a longitudinal cohort database, a proteomics reference atlas. Uptake grows. Citations accumulate. Then the coordinating grant expires, the hosting institution absorbs the maintenance costs until it cannot, and the resource either freezes in place or disappears behind a paywall that functionally ends broad access. The thing that was supposed to scale became a liability at exactly the moment it was most needed.

This is not a funding coincidence. It is a structural problem baked into how science allocates credit and money simultaneously.

The core tension is that scientific infrastructure is a collective good, but scientific incentives are built around individual attribution. A principal investigator who spends five years maintaining a shared tissue bank produces something with enormous aggregate value and very little that counts on a CV. The researcher who mines that bank and publishes a high-impact mechanistic paper collects essentially all the legible credit. Funders have recognized this asymmetry for decades without reliably solving it. The result is a recurring pattern: resources are created through bursts of grant enthusiasm, sustained through the personal commitment of a small maintenance team, and abandoned when that team exhausts itself or moves on.

The resources that do scale share a set of characteristics worth examining carefully. They typically have a revenue model that is not purely grant-dependent, whether that is a tiered-access fee, an institutional membership structure, or deep embedding inside a government agency with line-item budget authority. They have governance structures that distribute maintenance responsibility and decision-making across multiple nodes, so no single lab's departure is fatal. And they tend to have invested early in infrastructure that makes contribution easier than extraction -- the marginal cost of depositing data or samples is kept low, which sustains the inflow that makes the resource worth using.

GenBank is the textbook example of the stable end of the spectrum. Its longevity rests on mandatory deposition requirements tied to publication in major journals, federal hosting through the National Center for Biotechnology Information, and a data model standardized early enough that the switching costs for alternatives became prohibitive. None of those features emerged accidentally. They were deliberate policy choices made in the 1980s that looked bureaucratic at the time and look visionary now.

The failure mode at the other end of the spectrum is also reproducible. A resource built around a single informatics platform, hosted on a university server, maintained by a postdoc who will eventually need a permanent job, and documented only in the methods sections of the papers that used it. When that postdoc leaves, the resource enters a kind of undead state: technically accessible, practically unusable, quietly rotting.

Replication researchers encounter this regularly. A published analysis depends on a processed dataset stored at a URL that now returns a 404. The raw data were never deposited. The code was never versioned. The reagent was a one-off synthesis that no commercial vendor stocks. The study cannot be reproduced not because the science was fraudulent but because the infrastructure was ephemeral by design.

The policy conversation has been moving, slowly, toward treating data and biological resource deposition as requirements rather than courtesies. Journals have tightened data-availability statements. Some funders now require sustainability plans alongside data-management plans. These are real changes. But a sustainability plan filed at grant submission and never revisited is not a governance model. It is paperwork.

The question that doesn't get asked often enough is not how to build shared resources. Scientists are good at building things. The question is what institutional form a shared resource needs to assume before it can survive the departure of the people who built it.