One follow-up on editing the dictionary CSVs directly:
We should make the checks for a contribution explicit.
The discussions about order columns and missing values, leading zeros and dictionary/ERD names give us useful examples. A file can be a readable CSV and still lose an identifier, contain an invalid reference or disagree with the schema it is meant to describe.
I suggest three layers of checks: preserve the CSV values and structure; validate the dictionary’s definitions and relationships; then check that generated schemas and other outputs preserve those definitions. We should report which checks actually ran, including any rules the selected validator does not enforce.
We are reviewing LinkML as the common home for as much of that schema and rule information as possible, with complementary checks where needed. We are still reviewing and testing that work. Moving edits to GitHub makes the proposed change inspectable; the validation checks need their own evidence.
Examples of CSV edits that have caused problems in local workflows would help us choose the first regression cases.