Data Dragons
Name the dragons in your data.
In a Southern African resettlement-and-compensation database, the failure modes of the data started to breathe. The more suffering encoded in a dataset, the more powerful the dragon that emerged. Five species. Each one a specific, nameable pathology — and the people the spreadsheet refuses to record are what they feed on.
Run the diagnostic
Describe your data system, pipeline, or the organisation around it. The realm names which dragons infest it, how each shows up in your case, and the move that tames it.
An instrument, disclosed as such — a fast model reading your system against five named failure modes drawn from a real resettlement database.
The bestiary
The five species, and the data-system pathology each one is.
Merge Dragons
Draconis ConfluentiaImproper joins that combine human records without consent — identity confusion, and family connections erased as 'duplicates'.
Tell — Two people collapse into one row; 'Removed Duplicates' quietly severs a household.
Filter Wyverns
Draconis ExclususExclusion criteria that drop vulnerable populations — the families who fall into the gaps between classifications become invisible to the system that's meant to serve them.
Tell — A WHERE clause or priority tier silently removes the people most at risk; the report looks clean because they're gone.
The Great Pivot Serpent
Draconis TransformatioPivots and reshapes that transform categories themselves — 'Crops' become 'Structures', and a reframing of the data quietly rewrites what was owed.
Tell — A category changes meaning between two versions of the same table, and nobody can point to where the value went.
Ghost Entry Drakes
Draconis NulliusNulls and missing data that swarm incomplete records — signatures and dates vanish from agreements, and undocumented people flicker between existing and `null`.
Tell — Key fields are empty on exactly the records that matter most; the gaps aren't random.
The Recursion Hydra
Draconis InfinitusCircular references and self-joining calculations — each head computes a different version of the same family's number, and fixing one inconsistency spawns two more.
Tell — The same figure comes out different depending on which query you run; reconciliation never closes.