A document can identify its author without a single name. Its twin cannot.
Holistic Contextual PII Detection (HCPD) extends Nakato’s twin to the contextual fingerprints that mark a document’s author: a jurisdiction, a team size, a pilot quarter, a house term. Step through one below.
1 · As written — your environment
This Agreement is between Halden Treasury Ltd (the Discloser) and Merrow & Finch LLP (the Recipient).
The Discloser’s predictive cash positioning runs a tidewater sweep across every account before the day’s payments are released.
The Recipient will evaluate it within its FCA-regulated treasury operations, under English law, through its 12-person treasury team at its Edinburgh office during a Q4 2025 pilot on the Cortex platform.
Signed by Daniel Whitmore and Priya Natarajan, 14 November 2025.
Read for meaning → 5 entities, 8 contextual signals
Every document on this page is fictional: invented companies, people and figures, written to show what HCPD twins.
Two threat models
Nakato’s Reversible Semantic Pseudonymisation (RSP) handles the documents a regulated business processes every day. HCPD is for the ones it shares with people who would recognise it — where the names are only part of what gives the author away.
Operational documents
What identifies the record is the customer: a name, a date of birth, an account number, an address. RSP swaps each one for a semantic twin — fictional, but faithful in everything the task depends on — and the model reasons on a complete record.
RSP entity-level twin
High-sensitivity documents
What identifies the document is its context: the jurisdiction, the office, the size of the team, the quarter of the pilot, the house name for a method. Together these indirect identifiers narrow a counterparty’s guess to one company. HCPD twins those signals too.
RSP + HCPD entity and contextual twin
The documents
Three kinds of document that never used to go near a model. Open one: what you see is what HCPD sends — every entity and every contextual fingerprint already its twin, and every twin reversible.
2 entities, 9 contextual signals — all twinned.
MUTUAL NON-DISCLOSURE AGREEMENT
Between Corvale Systems Ltd (the Discloser) and Ashby Rowe LLP (the Recipient).
The Discloser’s forward liquidity modelling applies a millrace sweep to client accounts prior to the release of payments.
Governed by Irish law; the courts of Ireland have exclusive jurisdiction. The Recipient is CBI-regulated.
The 9-person treasury team at the Cork office are the sole users during the Q2 2026 pilot on the Meridian platform.
An agreement names two parties. Its terms name a third: the author. The method has a house name, the pilot has a quarter, the team has a size and a city. A counterparty who knows the market can put those together. The twin keeps the agreement’s shape and takes the fingerprint out of it.
The model only ever saw the twinned document; the answer you read had the originals restored — entities and context alike, from the same audit trail.
Every document on this page is fictional: invented companies, people and figures, written to show what HCPD twins.
When to use each
Redaction removes the names and the meaning with them. RSP keeps the meaning and twins the names. HCPD twins the rest of what gives a document away.
| Property | Standard redaction | RSP | RSP + HCPD |
|---|---|---|---|
| Entity-level PII | Removed. Names and addresses replaced with placeholders | Twinned. Semantically equivalent replacements | Twinned. Semantically equivalent replacements |
| Semantic context | Lost. Placeholder masking destroys it | Preserved. Full analytical utility retained | Preserved. Full analytical utility retained |
| Contextual fingerprints | Still readable | Still readable | Twinned. Domain, jurisdictional, organisational and temporal signals |
| Use case | Limited. Model output quality degrades | Operational. Complaints, customer correspondence, internal processing | High-sensitivity. Legal, board and investor documents |
| Reversibility | None. The original is lost | Full. Via the entity mapping | Full. Entity and contextual mappings |
Legal and regulatory basis
Identifiability is judged on every means reasonably likely to be used, not on names alone. Here is the basis, framework by framework, in the regulator’s own words where it has them.
Identifiable to whom is the question.
Pseudonymised data stays personal data for whoever can reverse it. For a recipient with no means reasonably likely to be used to identify anyone, the Court of Justice has held it may not be.
“‘pseudonymisation’ means the processing of personal data in such a manner that the personal data can no longer be attributed to a specific data subject without the use of additional information, provided that such additional information is kept separately and is subject to technical and organisational measures to ensure that the personal data are not attributed to an identified or identifiable natural person”
The Court of Justice held that pseudonymised data is not personal data for every recipient in every case. What decides it is whether that recipient has means reasonably likely to be used to identify anyone. For the controller, who can reverse it, it remains personal data.
The EDPB calls the environment in which pseudonymised data must not be attributed to a specific person the pseudonymisation domain. The controller sets it, most often as the set of recipients authorised to receive the data, and whoever holds the information that reverses the pseudonymisation stands outside it.
Identifiable means directly or indirectly.
The test is not whether a name appears. It is whether anyone, using the means reasonably likely to be available to them, could work out who the data is about.
Personal data is any information relating to an identified or identifiable living individual — one who can be identified, directly or indirectly, by an identifier such as a name or by factors specific to their identity.
“To determine whether a natural person is identifiable, account should be taken of all the means reasonably likely to be used, such as singling out, either by the controller or by another person to identify the natural person directly or indirectly.”
Could a reasonably competent person with no prior knowledge — using the internet, libraries and public records, but no specialist skills or criminal means — identify someone from the data? A document’s contextual fingerprints can fail this test on their own.
Reasonably linkable is the standard.
Personal information is defined by what it could be linked to, not by what it names. Contextual signals create exactly that linkability, and the state laws are converging on the same reading.
Personal information includes data that is “reasonably capable of being associated with, or could reasonably be linked, directly or indirectly, with a particular consumer or household.” Contextual signals create that linkability.
The Gramm-Leach-Bliley Act requires financial institutions to protect nonpublic personal information. Contextual data that permits re-identification of account holders can fall within it.
Virginia, Colorado, Connecticut and more than a dozen other states define personal data as information linked or reasonably linkable to an identifiable person. The direction is towards a broader reading of identifiability.
Model inputs are model risk.
Supervisors on both sides of the Atlantic hold a bank answerable for the models and third parties it relies on. What a document gives away to an outside model is part of that risk, and a contextually identifiable document leaves it unmanaged.
The PRA’s model risk management principles for banks, in force since May 2024, ask firms to identify, govern, validate and put mitigants around every model they use, AI models included.
The interagency guidance on third-party relationships (June 2023) expects banks to manage each third party’s risks across the whole relationship, including how it handles the bank’s and its customers’ information.
The FCA regulates firms’ use of AI through the rules they already follow — the Principles for Businesses, the Consumer Duty and the Senior Managers and Certification Regime — rather than rules written for AI. The firm owns the outcome of every model it uses.
The model risk guidance the Federal Reserve, the OCC and the FDIC issued together in April 2026, which replaced SR 11-7, scales governance, validation and controls to each model’s materiality, and leaves generative AI outside its scope.
What decides it, the Court of Justice of the EU held in EDPS v SRB (C-413/23 P, 4 September 2025), is whether the recipient has means reasonably likely to be used to identify anyone. The EDPB calls the environment where pseudonymised data must not be traced back to a person the pseudonymisation domain. With HCPD the model works inside it for the document as a whole, not only for the names in it — and the mapping that reverses the swap stays outside it, with you.
Coverage
The same document, recorded twice: once as RSP sends it, once as RSP with HCPD sends it. Equivalent in every way that matters, different in the one that counts.
| Signal | Entity-level twin · RSP | Entity and context twin · RSP + HCPD |
|---|---|---|
| Direct identifiers — a name, a national ID | Twinned | Twinned |
| Quasi-identifiers — a date of birth, a postcode | Twinned | Twinned |
| Domain terminology | Still readable | Twinned |
| Jurisdictional signals | Still readable | Twinned |
| Temporal markers | Still readable | Twinned |
| Organisational fingerprints | Still readable | Twinned |
| Combinatorial re-identification | Possible | Mitigated |
| Re-identification from context by a motivated intruder | Context still exposed | Designed against |
Both columns reverse exactly. The contextual mappings are written to the same audit trail as the entity mappings, so a reviewer can see every substitution and its reversal in one place.
RSP for the documents you process every day. HCPD when the threat model demands it. Both reverse exactly, and both write to the same audit trail.