Every month a plant asks its history the same questions. How many hours did each pump run? How much water went through the works, and what did the energy come to? Was the data even there? Today the answers often mean a data platform in the cloud and someone to build on it. Auge keeps every reading of the plant’s own exports in a folder on the plant’s own machine, with what each tag means, and answers those questions from it: per hour, day, week, month or year on the Norwegian calendar, as volumes from flows, energy from power, hours on and times switched on, and how much of each month has data at all. You can ask it in Norwegian or English, “driftstimer per pumpe per måned i fjor” or “how much water came in last week”, and it says back how it read the question before it answers. Each finished month gets its own report: tags that went quiet or froze, and figures outside anything the year before showed. A company with several plants sees them side by side. The figures open in Excel, and the assistant answers from the same store.
We tested the store on 10 public plants, 30.0 million readings, against an independent reading of the same files. It took 18% of the space of the CSV files, every monthly sum matched to one part in a billion, and on 9 of the 10 plants a monthly summary of every tag came back in under 0.1 s.
But on 2 plants it did not keep every column: it read the files through the reader that works out what tags mean, and that reader joins a value with its quality code and drops columns it judges empty of meaning. A store has to be a copy. It now reads plain exports column by column, as written; that change is tested on new exports next, not on these.
P56, written down before the run: FALSIFIED.
Then we tested the change on 10 exports from 10 publishers it had never seen, locked before the test. On 4 of the 4 plain exports every reading came back exactly, 15.0 million of them, and every monthly sum was right. But only 2 of the 6 exports in other layouts gave any readings: two Excel workbooks met a fault in the store, and two layouts, a header inside a comment and a title row above it, are ones the reader does not know yet. Those are being fixed, and the fixes will be tested on exports nobody has looked at.
P57, written down before the run: PARTIAL.