Plain text tables
that describe themselves.
CSV rows, plus a few lines that say what they mean: types, units, totals, where the data came from, how tables join. Any CSV tool still reads it.
¹sourceorders.txtab
²tableorders
- source
- shop_db.orders
- grain
- one row per order
- rows
- 3 declared · 3 read ✓ · header row
³inspectline 7
# column · column
Exact decimals, in US dollars, meant to be summed.
name=amounttype=decimalunit=USDrole=measureagg=sum
Reads with anything
A one-table file is ordinary CSV to any tool that skips comment lines. Several tables in one file? Read them with a Text Tables tool, or slice by the declared row counts.
⁴read with
$ pd.read_csv("orders.txtab", comment="#")
Several tables, one file
Orders and customers side by side, joined by references=. Or one small file describing CSVs that stay untouched next to it.
Totals that survive a preview
Sums and counts are written down for all rows. Paste the first fifty into a chat and the total is still right.
read more → ⁷meta-grainContext for agents
What a row is, which column is money, what never to sum — read in a few lines, before touching the data.
read more → ⁸sql-pre-dataSQL that comes along
The CREATE TABLE, the keys and a query, per database dialect, in the same file as the rows they load.
Remarks on cells
Pin a note to a cell, row, column or the whole file. Keyed by the row id, it stays put when the table is sorted.
read more → ¹⁰.txtab.zipBig and wide
Zip it; the schema rides in the archive comment, readable without unzipping. Wide tables can be stored column by column.
read more →