Field note
What does Codebookery not do?
A precise boundary around semantic inference, privacy classification, source-row storage, repository conformance, collaboration, validation, and professional review.
Observed evidence is not declared truth
Profiling reports the primitive types, blank counts, bounded distinct counts and length ranges seen in the rows it read. That is evidence about a sample, not a statement about the field: an identifier made of digits profiles as an integer and still has to be declared text. Every declared type, role, unit and meaning in the studio is an authored claim.
Local processing is not a permission decision
Running in the browser removes the application's upload path. It does not decide whether a particular file may be opened on a particular machine, and it does nothing about backups, extensions, sync services, download handlers or anyone else with access to the device. That judgement stays with you and with whoever owns the data.
A passive packet is documentation, not certification
Export gives you a reader page, machine-readable files, findings, review evidence, editions, a manifest and checksums. None of that certifies data quality, governance, repository conformance, privacy posture or fitness for an analysis. The studio has no reviewer, no approval workflow, and no authority to grant one.
Worked example from the sample project
The sample project can be finished, linted and frozen, and it still cannot answer the question people bring to a data dictionary, which is whether they may trust the dataset. It can tell you that condition is bound to a value set of good, watch and unknown, that species_code declares UNK as a missing code, and that access_note has no definition yet. It cannot tell you whether an observer applied watch consistently across two weeks of fieldwork.
That gap is not a defect to be closed by a better rule. Only somebody who was there, or who owns the collection process, can close it, and the product is built to make the gap visible rather than to paper over it with a confident-looking score.
What this does not do
- It does not determine meaning, semantics, correctness or intent from data.
- It does not classify personal, sensitive, confidential or regulated data.
- It does not certify quality, governance, repository conformance or fitness for any analysis.
- It does not store source rows, so it cannot answer questions about individual records.
- It offers no collaboration, sync, approval workflow or authenticated sign-off.
Do this in the studio
Treat every finding as a question for a person, not a defect the tool can close.
- Open the Reader review station and record who read which field.
- Write down the reader's own words when they misread something.
- Revise the folio, then re-run the deterministic review.
Open the studio at the Reader review station