Table

1. 01CSV, TSV and Parquet open as their own tab on the stage, beside the sequence they describe
2. 02Sort by any column, and resize columns by dragging the header edge
3. 03Filter rows by text found in any column
4. 04Dense mode for a tighter row height, so more of the file is on screen
5. 05Rows are virtualized, so scrolling a wide table stays smooth
6. 06The footer states rows shown of rows in the file, and names the file when only its head is read
7. 07Pick a table from any project — one from elsewhere is copied into the open project, so the tab survives a reload
8. 08The same renderer draws the file manager preview and the stage tab, so the two can never disagree
9. 09Parquet cells that are not text — integers, timestamps, nested values — are rendered rather than dropped

CSV and TSV are parsed with papaparse, reading the header row and skipping empty lines. Parquet is read with hyparquet: the footer metadata and schema give the column names and the true row count, and only a head of real rows is decoded from the file. Both paths share one loader, so the stage tab and the manager preview read the same bytes the same way.

Everything runs client-side. Files are parsed in the browser and projects are saved on this machine, so an unpublished sequence is never uploaded.

How many rows does it show?

The first 200. That cap is stated, not silent: the footer reads "200 of 4,000,000 rows, reading the first 200 of counts.parquet", because a table that quietly presents its first screenful as the whole file is a scientific error rather than a cosmetic one. Filtering and sorting apply to the rows that are loaded.

Does the Parquet row count come from the head that is read?

No. It is read from the file footer metadata, so the count is the real number of rows in the file even though only the first rows are decoded.

What if my CSV has duplicate column headers, or none?

Columns are keyed by position rather than by name, so repeated or missing headers do not collide and no column is silently dropped.

Is my file uploaded anywhere?

No. It is parsed in the browser from the copy stored in your project.