In a draft cleaned up by hand the names almost always go. The land registry references almost never do. They look like technical numbers rather than personal details: municipality, sheet, parcel, unit, category, cadastral income. They are in fact the most stable identifier in the whole deed.
A land registry reference is an address
The difference between a name and a parcel is that the name is ambiguous and the parcel is not. There are thousands of people called Mario Rossi; there is exactly one sheet 12, parcel 340, unit 7 in the municipality of Bergamo.
And that one can be looked up. A property search by land registry reference is a service that exists, costs very little and is open: from an identifier you get to the registered owner. It takes no special skill, only knowing the service is there.
This is the case Recital 26 of the GDPRopens in a new tab describes with its formula about means “reasonably likely to be used” by anyone to identify a person. Data is anonymous when that route does not exist; while it does, the data is personal and the regulation applies in full.
What it is, field by field
The reference is worth unpacking, because the pieces do not weigh the same and whoever cleans a deed should know which to look at first. A land registry reference is made of the municipality (with its code), an optional section, a sheet, a parcel and, in the buildings register, a unit. The land register stops at the parcel; the buildings register adds the unit because it has to tell apart the flats inside one building.
Beside them sit the classification fields: category (A/2, C/6, D/1), class, size in rooms or square metres and cadastral income. These do not identify: A/2 with an income of 700 euro describes tens of thousands of properties. They become identifying in combination: with the municipality they narrow a great deal, and with any other surviving value they narrow to one.
The working rule that follows is short: the triple municipality, sheet, parcel (plus the unit where there is one) has to be treated like a tax code, because it does the same job. The rest of the classification deserves the treatment a date of birth gets: harmless alone, decisive alongside the rest.
The other values that survive a hurried clean-up
Alongside the registry data, a notarial deed usually keeps the date of birth, which with the place of birth reconstructs most of an Italian tax code; the IBAN given for payment of the price, which identifies with no intermediate step; the repertory number; and the chain of title, which often carries the previous owner’s full name and the details of the earlier deed.
The chain of title is the treacherous one, because it is a sentence and not a field: it reads as narrative text, and anyone cleaning up by hand tends to skip it.
Why substituting beats deleting here too
Delete everything about the property and the text becomes useless: a clause about “a property” gives no model any way to check whether the descriptions match between the recitals and the operative part, which is often the check you wanted in the first place.
With stable placeholders the check survives: if the recitals and the operative part show the same substituted identifier, the model sees they match; if they show two different placeholders, it sees they do not. It is the same reason the placeholders for people are numbered rather than generic, and the difference between the two routes applies here as everywhere.
Not just the land registry: Italy’s public indexes
The land registry is the most visible case of a wider category, and recognising the category is worth more than memorising a list. A value identifies when there is a public, searchable register tying it to a person. And Italy, for excellent historical reasons, has many.
A company register number leads to the company extract, which leads to shareholders and directors with names, surnames and dates of birth. A VAT number leads to its holder in one step. A licence plate leads to the owner through the vehicle register. The certified email address of a professional or a business sits in a national public index. A docket number identifies proceedings and, through them, their parties. Even the repertory number of a notarial deed is a key, for anyone with access to the right archive.
The consequence for whoever anonymises is that “take out the personal data” is a misleading phrase, because it suggests looking for things that look personal. What has to be looked for is something else: every value that is the key to an archive. They are almost always numbers, they almost always have a regular format, and that is precisely the property that makes a program better than an eye at finding them.
What recognises it, and what does not
One practical thing to check before choosing a tool: the vocabularies of products built outside Italy do not contain the Italian land registry, nor the tax code, nor the docket number. A detector trained on American names, addresses and social security numbers, pointed at an Italian notarial deed, finds the names and leaves everything else, without saying that it did.
Omissia recognises sheet, parcel and unit alongside tax codes, VAT numbers, IBANs, addresses, amounts and dates, and every value it finds is shown before it is replaced.