Begin with column quality
For numbers, inspect min, max, quartiles, mean, standard deviation and the histogram. For text, inspect distinct and common values. For dates, check the valid range, empties and timeline. Invalid values should be counted, not coerced into zero.
Pivot a huge CSV locally
In csvlite, choose one or more grouping columns, then calculate count, sum, average, min, max, distinct count or a list of distinct values. The result opens in a new tab where it can be sorted, filtered, profiled and exported like another table.
Is CSV a database?
No. CSV has no schema, indexes, constraints, transactions, relationships or standard null semantics. Calling a folder of CSV files a database does not create those guarantees. But direct analysis is useful for one-off exports, logs and data handoffs where importing first would add friction.
When to switch to DuckDB
Use a database for joins, reusable SQL, window functions, durable typed tables, many-file queries and automated pipelines. Visual exploration in csvlite is strongest for understanding unfamiliar data and producing a focused result. The two tools complement each other.
A good sequence
Inspect → validate → filter → summarize in csvlite. Then import or query with DuckDB when the analysis becomes a program rather than an investigation.