When we build any kind of software, research software included, we often focus entirely on features. The ATR project was founded to look at the other side of the coin: reducing layers of abstractions in order to run web applications with a significantly smaller carbon footprint. ATR uses the Research Organization Registry (ROR) as its primary research dataset. Through automation and targeted analysis, our software agents augment, validate and parse ROR records.
While optimising ATR's list generation using standard server-side sorting, something unexpected appeared. A handful of organisations refused to sort alphabetically.
The culprit? Invisible data ghosts.
Deep inside the underlying strings, a few records contained an invisible character that was invisible to humans but able to disrupt deterministic string sorting for a machine. We reported the findings to the ROR community ( Issue #36285 ). The ROR team responded incredibly fast, reflecting the dedication of the curation team.
It may seem like a small detail, but it illustrates how practical work on open datasets can reveal opportunities for improvement. Sharing those findings is one of the ways open research communities benefit from collaboration.