The lesson for modern engineers is uncomfortable. We are now building large language models and automated decision systems that promise to replace human judgment. Elster reminds us that the real world is fuzzy, contradictory, and full of exceptions. A system that is 99% precise but 0% tolerant is not a tool—it is a barrier. Elster did not fail because it was poorly coded. It failed because it succeeded in coding the law so perfectly that it forgot the law is, at its heart, a human institution meant to be interpreted, not executed.
Elster Software was dismantled in 2018, its assets nationalized and its team dispersed. But its ghost haunts every conversation about AI, automation, and governance today. Elster’s failure was a textbook case of Goodhart’s law applied to software: when a metric (strict schema validation) becomes the target, it ceases to be a good metric. By eliminating all ambiguity, Elster eliminated all discretion, and without discretion, a bureaucratic system cannot function. elster software
The problem emerged as the tax code itself grew more complex. The German fiscal code (Abgabenordnung) runs to thousands of pages, filled with exceptions, special cases, and regional variances. To handle this, Elster’s engineers did what any rational technocrat would do: they encoded the law directly into the software’s validation logic. A deduction for home-office expenses? The software required a specific room size in square meters. A charitable donation? The software demanded the exact charity’s tax ID, verified against a live database. The lesson for modern engineers is uncomfortable
In the end, the most sophisticated tax software in Europe was undone by a simple truth: a plumber with a wet signature and a kind tax officer is infinitely more efficient than a flawless machine that says “no.” A system that is 99% precise but 0%