Field note

Why does the studio avoid saving imported records?

The product's job is maintained documentation, not data exploration. Discarding source rows narrows persistence and export exposure while retaining bounded structural observations needed to start field folios.

Temporary processing still matters

Rows exist in page memory while the strict parser and the bounded profiler run, and are then released. Nothing writes them to the project, a checkpoint or a packet, so no export can leak an individual record. That narrows the exposure; it does not decide whether the file was appropriate to open on this device in the first place.

Aggregates can still disclose

What survives is bounded: per field, the primitive types seen, the row and blank counts, a capped distinct count, and a length range. Those numbers can still say something. A distinct count of one on a rare category, or a blank count of zero on a sensitive flag, is information about the people in the dataset even though no row was kept.

Authored examples require confirmation

Because no rows are kept, the studio cannot offer to fill an example from your data, and it would not do so if it could. If an example helps a reader, you type an invented value that shows the shape and set the safe-to-publish confirmation yourself. An unconfirmed example is withheld from exports rather than published quietly.

Worked example from the sample project

Import the worked example's file and then export a checkpoint and a packet, and search both for a single tree. You will not find one. What survived is the field names and bounded aggregates: row and blank counts, primitive type counts, a bounded distinct count and a length range. The rows themselves were released when profiling finished.

This also explains a feature people expect and do not get. The studio will not offer to fill a folio's example from the data, because doing that would mean copying a real value into a document intended for circulation. If you want an example, you type one, and you confirm it is safe to publish before it ships. A copied value can identify a person, a location or a rare category, and once it is in a packet it travels with every copy of that packet.

What this does not do

Discarding rows is not an anonymisation technique and makes no dataset safe to open.
Bounded aggregates can still disclose; a distinct count of one on a rare category is information.
The product cannot answer any question about an individual record, by design.

Do this in the studio

Type an example rather than copying one, and confirm it before it ships.

Open the Field folios station and write an invented example where one helps.
Set the safe-to-publish confirmation deliberately, not by habit.
Leave the example empty whenever you are unsure.

Open the studio at the Field folios station