Datasets

Somewhere to keep the numbers. A workbook you add becomes a dataset: stored whole, read column by column, and available to compose from. It is not shredded into individual facts, because a number without its row and its column does not mean anything. Excel workbooks arrive with their tabs intact, and Gyrium opens the one that actually holds figures, so a cover sheet is never the thing you end up charting.

Numbers are the part of a communication that must not be wrong. Every other approach either retypes them by hand, which invites error, or lets a model read the grid, which invites invention. A dataset is read once, correctly, and every use points back at it.

The reader works out which row is the header, what each column is for, and which rows are totals. Then it tells you what it decided, in plain words, and every part of it can be overruled. Move the header, change what a column is for, put a totals row back, and your answer wins: the reading is redone around it.

It carries units through, so a column of euros stays euros and a percentage is never averaged as though it were a count. It reads European decimals, and French, Dutch and German workbooks: totals, notes and month names are read correctly. From the grid it computes what the table actually says: the leading finding, the runners-up, and which cell deserves emphasis.

When you sign the reading off, the record keeps who agreed it and which reading they agreed to. Change the header or the column roles later and the sign-off lapses, because it no longer describes what you approved.

That reading is handed to the writer at composition time, so the prose about a chart comes from what the numbers show. A check runs the other way too, flagging a slide whose words contradict the chart beside them.

1. Add a workbook or a CSV as a source. Excel files keep their tabs, and you can switch between them.
2. Read what Gyrium says it found: the header row, what each column is for, the rows it left out.
3. Correct anything it read wrong. Your correction is what the chart is drawn from.
4. Sign the reading off, so the next person can see it was checked and by whom.
5. Use it in a chart, a table, or an Expression. The takeaway and the emphasis come with it.

A figure on a slide drawn from a table cites that table, pinned to the version it was read from; a claim in running prose drawn from a table is not yet cited that way. Charts and tables from a dataset are composed into presentations and social posts; documents do not draw from datasets yet. Drift between a dataset and a deck that used it is detected and offered, but never applied on its own.