## TL;DR

The warehouse inferred a numeric type for the column, so `00734` loads as `734`. Force the column to text with `+column_types` in the seed config, then `dbt seed --full-refresh` to reload.

## Error

```text
dbt seed bug: truncating leading zeros in csv columns
```

## Steps

1. Check the current column type in the warehouse for the seed table. Expected: it shows a numeric type like integer or numeric.
2. Add a column type override in `dbt_project.yml` under the seeds config:
```yaml
seeds:
  +column_types:
    zip_code: varchar
```
Expected: the seed config pins the column to text.
3. Run `dbt seed --select [SEED NAME] --full-refresh`. Expected: the table is rebuilt with the text column.
4. Query the seed table and check the values. Expected: leading zeros are preserved.
5. Rerun downstream models and tests. Expected: everything downstream sees the correct values.

## When to use

- Seed values like zip codes, product codes, or IDs lose leading zeros after `dbt seed`.
- The CSV itself has the zeros (verify by opening the raw file).

## When not to use

- The CSV is missing the zeros (fix the source file, not the column type).
- A model (not the seed) strips the zeros (fix the model logic).

## Tool compatibility

- dbt Core 1.0 and later, all adapters. The text type name varies (`varchar`, `string`, `text`).

## Variant phrasings

### Seed column loaded as float, values rounded

Same inference problem; pin the type explicitly.

### CSV values look right in the file but wrong in the table

Type inference at load time is the culprit; the file is fine.

## Why it happens

dbt infers seed column types from the CSV contents. All-digit values infer as numeric, and numeric storage drops leading zeros permanently at load time.

## Edge cases

- Quoting values in the CSV does not reliably force text; the explicit `column_types` config is the robust fix.
- Changing the type requires `--full-refresh`; a plain `dbt seed` will not alter the existing table.
- Downstream joins on the column need matching types; changing seed type can break joins that relied on numeric.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_2131YA9mbGIUg2jTEQzOwg