TAR Review Set Anonymisation with anonym.plus

Clear PII before a machine-review model trains, all on your own device.

In simple terms, PII redaction is the on-device process of finding and masking personally identifiable information in a document before it is shared.

TAR-set anonymisation is the removal of personal data from documents fed to technology-assisted ranking, an approach English courts have accepted as consistent with the proportionate-search duty in CPR 31.7 since Pyrrho Investments Ltd v MWB Property Ltd [2016] EWHC 256 (Ch). anonym.plus runs locally, so PII does not enter the training data before that proportionality question is even settled.

When this applies

The model ranks documents by relevance to speed triage, an approach Pyrrho Investments approved as a reasonable and proportionate way to conduct a search under what is now CPR 31.7. Feeding raw PII into it spreads exposure across the whole index, so clearing it first keeps that risk low regardless of how the ranking model performs.

How anonym.plus handles it

  1. Point anonym.plus at the document set on your device.
  2. Local OCR reads any scanned items in the set.
  3. The tool flags names, contacts, and IDs across files.
  4. Use steady labels so relevance signals survive.
  5. Replace or mask each confirmed value.
  6. Save the clean set for the ranking workflow.

What you need to provide

PII entity types detected

Categoryanonym.plus entity typeExample
NamesPERSONcustodian name → [PERSON_n]
ContactEMAIL_ADDRESSsender email → [EMAIL]
DatesDATE_TIMEdoc date → [DATE]
IdentifiersUK_NINONINO → [NINO]
LocationLOCATIONaddress → [ADDRESS]
AccountUK_SORT_CODEsort code → [SORT_CODE]

Compliance achieved

Anonymise TAR document sets offline — see plans & start free →

Limitations & cautions

Anonymising before machine ranking can shift how a model reads context. Steady labels keep most signals, but test recall on a control set first. Free-text clues that survive redaction still need a human pass on relevant items.

Frequently asked questions

Do English courts accept technology-assisted review?

Yes. Pyrrho Investments Ltd v MWB Property Ltd [2016] EWHC 256 (Ch) was the first English case to approve predictive coding as a reasonable and proportionate search method, and later cases have followed that approach under CPR 31.7.

Does anonymising first hurt accuracy?

It can shift context, but steady labels keep most text signals. Test recall on a control set before relying on the ranking, whatever proportionality argument you plan to make under CPR 31.7.

Why clear PII before the model trains?

Raw PII in training data spreads across the whole index at once. Clearing it first keeps that exposure low, independent of whether the court later accepts the ranking approach as proportionate.