When a platform retires a competition format, should its old medals keep full weight, or be discounted or labeled with context?

This explores what should happen to old credentials once the system that issued them is gone. The clearest evidence in the corpus is an audit of Kaggle competition medals. Medals predicted how well someone would perform on hidden tests, but almost all of that predictive power came in the medal's first year. That held both before and after generative AI arrived, so fresh medals stayed meaningful Do Kaggle medals still predict performance after AI arrived?. The problem was old medals. When the platform retired its upload-style competition format, those medals kept losing relevance on the usual schedule but stayed on profiles at full face value. The audit traces about half of the drop in how informative those medals were to this 'stranding', not to AI How much did retiring a competition format hurt medal credibility?. So the damage didn't come from cheating or from the medals being wrong. It came from the platform letting a credential outlive the process that once checked it.

That points toward relabeling over silent downweighting. Since a credential's information fades with age in a predictable way, the honest fix is to show its age and where it came from, for example 'earned in a retired format, 2019', rather than quietly adjusting a hidden score. A related idea from a different area treats distilled expertise as versioned files that can be inspected, corrected and rolled back, not as hidden state Can person-grounded skills remain auditable without hidden prompt state?. Credentials could follow the same rule: a visible lifecycle beats an invisible reweighting that nobody can audit or dispute.

The reason to act at all is the people and systems reading the credentials. They mostly don't look past the label. LLM judges reliably give higher scores to responses that carry authority signals like references or polished formatting, whether or not the content is any good, and these biases can be exploited with no technical skill Can LLM judges be fooled by fake credentials and formatting? Can LLM judges be tricked without accessing their internals?. AI-assisted hiring adds another twist: models prefer resumes they rewrote themselves, which shows evaluators reacting to surface style rather than substance Do language models favor resumes they rewrote themselves?. On the human side, 'cognitive surrender' describes people accepting a signal without checking whether anything still backs it, because checking is costly When do users stop checking whether AI output is actually backed?. A stranded medal is exactly that kind of unbacked signal. Recruiters and automated screeners will keep treating it as current unless the label itself says otherwise.

Here's the less obvious takeaway. A credential's worth depends on something still standing behind it, not just on the fact that it was once earned. Proposals for 'personhood credentials' make this explicit: their value comes from a trusted issuer that stays accountable for what the credential claims Can people prove they are human without revealing who they are?. When a platform retires a format, it withdraws that backing without saying so. Relabeling is how it admits the change.

The corpus has limits here. It gives strong evidence that stranding causes harm and that evaluators trust labels at face value. It has no studies testing whether relabeling or downweighting actually changes how recruiters or algorithms behave. The case for relabeling is an inference from the evidence, not a measured result.