Field note
What can be observed from a CSV safely?
A strict local parser can observe headers, primitive-type counts, blank counts, bounded distinct counts, and length ranges. Those observations still do not reveal business meaning and may carry disclosure risk.
Strict parsing matters
Delimited files need correct handling of quotes, embedded delimiters, line breaks, encodings, duplicate headers, empty headers, and ragged rows. Silent parser repair can turn bad intake into misleading evidence.
Discard rows after profiling
After profiling, the studio keeps field names and bounded aggregates, not rows or extracted raw examples. That reduces persistence but does not make every file appropriate to open on a device.
Author declarations separately
Choose intended type, role, meaning, unit, missingness, and source after review. A numeric-looking ID is a common example where observed shape should not dictate declared meaning.
Worked example from the sample project
The worked example page publishes a small delimited file from the fictional Northbank survey, and importing it shows exactly how far observation gets you. The parser reports six headers and, for each one, the primitive types it saw, the blank count, a bounded distinct count and a length range. It reports that trunk_cm held only numeric values and that condition held three distinct strings.
None of that is a data dictionary yet. Observation cannot tell you that trunk_cm is a circumference measured at a stated height in centimetres, that condition is a bounded visual judgement whose unknown code means not rated, or that tree_id must stay text even though it looks like a key. Those six folios are still entirely yours to write; the profile just saves you typing the names and gives you evidence to write against.
The parser is strict about the things that quietly corrupt an import: unbalanced quotes, text after a closing quote, ragged rows, duplicate or empty headers, and reserved names are all refused rather than repaired. A silently repaired file produces confident, wrong evidence, which is worse than no evidence.
What this does not do
- It reveals no business meaning; every declaration remains an authored claim.
- It repairs no malformed file and guesses no delimiter beyond a bounded detection.
- It keeps no rows and extracts no example values automatically.
Do this in the studio
Import for the field names and the evidence, then author the meaning yourself.
- Open the Import structure station and load a delimited file you are permitted to open.
- Read the observed profile as evidence rather than as a conclusion.
- Declare the intended type, role, unit and meaning in the field folios.
Open the studio at the Import structure station