Everything is coded: How to prevent a game of telephone in a system full of interfaces

What this was about

Franziska Schuster framed MfN's digital catalogue infrastructure work through the 'game of telephone': standard formats like Darwin Core alone don't keep data meaning intact across professional hand-offs. Developer Jasper then described migrating about 600,000 insect specimen records with 2.5 million images, digitised by a Picturae conveyor pipeline, into Specify after the cyber attack fragmented the data. Requirements unclear from the start made his mapping code slow and hard to verify. Taxon 'NURIs' (internal identifiers) turned out to be non-unique, with up to 2,000 IDs for one taxon, test entries and drawer IDs reused as taxon IDs. Schuster closed with Scrum as a 'don't panic' strategy with three-week sprints and stakeholder feedback.

Why it matters. It is a candid case study of how identifier semantics break down in mass digitisation pipelines, and it shows how agile methods can manage infrastructure rebuilds.

In the room

• Coded systems narrow meaning only if sender and receiver share an understanding of the code.
• The Picturae conveyor pipeline digitised just over 600,000 specimens with over 2.5 million images, about 5,000 specimens per day, over roughly 2.5 years including development.
• After the hack the data were fragmented; deduplication and consolidation came before mapping to Specify.
• Iterative feedback from Specify and collection managers made the mapping code increasingly slow and opaque; a month-long rewrite followed after publishing about 22,000 records.
• Taxon NURIs were not unique: up to 2,000 IDs per taxon, empty taxa with IDs, and test entries.
• Pragmatic fix: accept a taxon NURI only if unique; otherwise keep it in the determination for findability.
• Key missing question: is the data sufficient at this point to assign an identifier?

Notable moments