## TL;DR

Make fixtures shaped like production: real field distributions, edge-case values, and valid relationships. Anonymize anything copied from prod, and keep fixtures minimal per test.

## Error

```text
(Not an error; a data quality practice. The failure it prevents: tests passing on toy data that fail on real data.)
```

## Steps

1. Sample production shapes: field lengths, character sets, nullability. Expected: a realistic profile.
2. Build fixtures matching the profile, including edge cases (unicode names, long strings, nulls). Expected: representative data.
3. Anonymize any prod-derived data URIs names, emails, IDs. Expected: no PII in the repo.
4. Keep fixtures minimal per test; compose from builders. Expected: readable tests.
5. Refresh fixtures when the schema changes. Expected: no drift.

## When to use

- Toy fixtures missing real bugs.
- Setting fixture standards.

## When not to use

- Factory configuration mechanics.
- Load testing data (different scale).

## Tool compatibility

- Any framework; fixture files or builders.

## Variant phrasings

### Realistic test data

The goal; production-shaped and anonymized.

### Production-like fixtures

The same idea; mind the PII.

## Why it happens

`"test"` and `"foo"` never contain unicode, nulls, or 200-character strings. Real data does, and bugs hide in the difference.

## Edge cases

- Anonymization must be irreversible; hashing with salt, not masking.
- Keep a few full-size fixtures for performance-sensitive paths.
- Document the fixture profile so new tests follow it.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_cLhx3bQ5ZrL6TckTbXz1MA