[Question] Is possible to process a dataset that has key hashed from phone number?

[alvin-twofive]

I followed this guide to prepare a dataset: https://swiftpackageindex.com/apple/pir-service-example/main/documentation/pirservice/testinginstructionslivecalleridlookup#Preparing-the-dataset

Actually, because of the regional data privacy, I should not store the raw phone numbers of users. So, I am thinking of hashing the phone numbers and store them as a keys. For example:

• Instead of:

identities {
  key: "+123"
  value {
    name: "Adam"
    cache_expiry_minutes: 8
    category: IDENTITY_CATEGORY_PERSON
  }
}
identities {
  key: "+1234"
  value {
    name: "Bob"
    cache_expiry_minutes: 7
    block: true
  }
}
identities {
  key: "+12345"
  value {
    name: "Grocery Store"
    cache_expiry_minutes: 18
    block: false
    category: IDENTITY_CATEGORY_BUSINESS
  }
}

• Can I store hashed phone numbers as keys?

identities {
  key: "1b245e45ea4b2d81792ddafa2320359aef1533de84b75e88d13148c83bd198ae"
  value {
    name: "Adam"
    cache_expiry_minutes: 8
    category: IDENTITY_CATEGORY_PERSON
  }
}
identities {
  key: "41395a6b882cda8c37cae2ce9f9d35e8327109aab27da52ae3ee50482b74a86d"
  value {
    name: "Bob"
    cache_expiry_minutes: 7
    block: true
  }
}
identities {
  key: "90a73f096e89e5bc9f0d0a6a387f0c27e5f16ff72233b9be192b6229606e88e2"
  value {
    name: "Grocery Store"
    cache_expiry_minutes: 18
    block: false
    category: IDENTITY_CATEGORY_BUSINESS
  }
}

If possible, where could I customize the logic of PIR query?

I would appreciate it if you have any ideas to help me.

[karulont]

TLDR: Customizing the logic to alter PIR query is not possible, and likely it is not the right approach.

Longer answer:
Hashing phone numbers is not very useful. The domain of all phone numbers is quite small, so it would not be very expensive to try to hash all possible phone numbers until you get a matching hash.

Regarding regional privacy policy:
I guess you are allowed to collect and process phone numbers. Is it only storing phone numbers in the clear that is problematic?
My suggestion: after you have processed the database: (Processing the dataset), you should remove the temporary files. The final processed database does not actually contain the phone numbers anymore. But it is designed for "efficient" lookup using Private Information Retrieval by phone number, so the fact that the phone numbers are not in the data does not mean that derivative things like hashes of the phone numbers are not in there. But it might be enough to satisfy the policy.

Technical description about how the keyword PIR works:
Keyword PIR allows one to use IndexPIR as a building block to implement keyword-value lookups.
The way we implement it is that we turn a keyword into a hash of it. And then see which bucket it should end up in. This is like implementing a hash table. However, we do not actually store the keywords themselves. Only hashes of the keywords.
This means that the processed database, that is ready to be served using PIR service, does not contain the keywords, but it does contain the hash prefix of the keywords. For more details see https://github.com/apple/swift-homomorphic-encryption/blob/main/Sources/PrivateInformationRetrieval/KeywordPir/HashBucket.swift#L26E each hashbucket has a list of entries and each entry has a hash prefix (first 8 bytes of SHA256(keyword)) and a value.