Everyone remembers Herman Hollerith — the American engineer whose punched-card tabulator counted the 1890 U.S. Census and whose company eventually became IBM (US Patent 395,782, filed 1884). Almost nobody remembers Otto Schäffler, a Viennese telegraph-equipment manufacturer who, in the same window, built a parallel machine for the 1890 Austrian census — and then quietly patented improvements that pushed tabulation toward something startlingly close to a modern database query.

Schäffler licensed Hollerith's basic patent, but his Austrian Patent of 1890 (and follow-ons through 1895 covering what he called the Zähl- und Sortiermaschine) added a mechanism Hollerith's machine lacked: automatic sorting by multiple criteria in a single pass. Hollerith's 1890 sorter required an operator to re-feed the deck for each new question. Schäffler's machine used a bank of relay-driven gates that could simultaneously route a card based on combinations of hole positions — effectively, a hardwired WHERE clause with conjunctions.

What the machine actually did:

• A stack of stiff cards represented citizens. Each card had a fixed grid of positions; holes encoded sex, age bracket, religion, occupation, district, language spoken at home.
• A reader pressed pins through the holes. Where a pin passed through, it closed a circuit and incremented an electromechanical counter.
• Schäffler's addition: a combinatorial relay matrix behind the reader. The operator could pre-wire questions like "Catholic AND Czech-speaking AND male AND age 20–30" and the machine would tally and physically deflect matching cards into a dedicated bin — all in one pass over the deck.

This is, mechanically, a columnar database with compound boolean filters. The card is the row. The hole positions are the columns. The relay matrix is the query plan. The counter is SELECT COUNT(*). The sort bins are GROUP BY.

Schäffler also patented something Hollerith didn't bother with for years: a card-verification mechanism. Two operators punched the same source data on two machines; a comparator flagged any card where the hole patterns disagreed. This is double-entry data validation — the exact same idea modern ETL pipelines call reconciliation checks.

Why does no one know him? Three reasons. First, his machines were built in tiny numbers — maybe a dozen total — for Austria-Hungary, Saxony, and a handful of insurance companies. Second, he died in 1908 and his Vienna workshop was absorbed into local telegraph manufacturing; the patents lapsed without a corporate heir like IBM to keep the name alive. Third, the Austrian census records he processed were largely destroyed in two world wars, erasing the most visible monument to his work.

The modern echo: when a data engineer writes SELECT religion, COUNT(*) FROM citizens WHERE language='cs' AND age BETWEEN 20 AND 30 GROUP BY religion, they are describing, almost line-for-line, what Schäffler's 1895 machine did in brass and relays. The 1970s relational model formalized the algebra. Schäffler had already built the physical execution engine eighty years earlier — he just lacked the vocabulary to describe what he had done. His machine wasn't a calculator. It was a query processor, and it ran on punched cardboard and clicking electromagnets while Vienna was still lit by gaslight.