Dictionary recipes

A few concrete patterns. Each builds on attaching a dictionary to a profile.

Keep a client consistent across every file

You anonymize documents for Mediterrània Assegurances regularly, and the output needs to be self-consistent across months.

1. Create a dictionary, Cliente Mediterrània, in the manager.
2. Make a profile for this client, attach the dictionary, and turn on both reuse switches.
3. Process files as they come. The first run fills the dictionary; later runs reuse it. Maria Josep Solà is always Joan Carles Ribera, in every report.

See the full version in Consistent fake identities across a team.

Protect internal codenames the model can’t know

Your company calls a project ORION. No general model knows that’s sensitive.

1. Add a Manual entry: Proyecto ORION → Proyecto AZUL.
2. Attach the dictionary to your profile.
3. Every document that mentions ORION now ships as AZUL, deterministically — no reliance on the model spotting it.

Pair this with a profile prompt note (“treat project codenames as PII”) to catch new codenames too.

Pin fixed mappings for brands and entities

Some swaps must be exact for downstream tools to line up — a partner’s name, a product SKU, a standard test account. Manual dictionary entries give you fixed, auditable mappings that never drift, unlike freshly generated fakes.

A stable cast for test data

Building test fixtures from real document shapes? You want the same fake people every time so test assertions don’t churn.

1. Run a representative document with synthetic mode + add to dictionary on.
2. Review the generated cast in the manager; tidy any value you don’t like.
3. From then on, run with replace existing on. Fixtures regenerate with a fixed, known cast.

A reversible swap

Because a dictionary records original ↔ replacement, it’s also the key to getting the original back. If you anonymized a document, sent the safe copy to an external tool, and got an edited version back, the dictionary lets Piixie restore the real values — even when the exact run record isn’t available. Keep the dictionary that produced a document if you might need to reverse it later.

Housekeeping

• One dictionary per context keeps each set small and meaningful.
• Duplicate before a big change so you can fall back.
• Use Hide original when curating in shared spaces.
• Dictionaries live in Piixie’s local database and hold real PII — treat them as sensitive.

Next

• Team consistency walkthrough
• Reverse with a dictionary