Match SDKs to their owners

Addresses and landlords

A host name is like a street address; the owner is the landlord. events.metrics.example and cdn.metrics.example are two addresses with the same landlord. And partners.fitfox.example sits on FitFox's street, yet a different landlord (AdCo) rents it. This is the app version of the CNAME cloaking you met in Leg 1.

from watched.leg5b_apps import SDK_OWNERS
print(SDK_OWNERS["adco.example"])            # AdCo
print(SDK_OWNERS["partners.fitfox.example"]) # AdCo
print(SDK_OWNERS["fitfox.example"])          # FitFox

Most specific first

To find the owner of a.b.adco.example, try the full name, then drop the leftmost label (a label is one dot-separated piece) and try again: b.adco.example, then adco.example. Stop at the first match. Stop before you are down to a single label like example, which nobody owns. If nothing matches, the owner is "Unknown", and an unknown owner deserves extra attention, not less.

labels = "a.b.adco.example".split(".")   # ['a', 'b', 'adco', 'example']
print(".".join(labels[1:]))              # b.adco.example

Going most specific first is what catches partners.fitfox.example: its full name is in the map as AdCo, so you never fall back to FitFox.

A data-sharing graph

The practice world gives you fields_of(entry) (the same as your all_fields from the last lesson) and DATA_KINDS, which says what kind of data a field name carries:

from watched.leg5b_apps import DATA_KINDS
print(DATA_KINDS["heart_rate"])   # health
print(DATA_KINDS.get("slot"))     # None (not personal, ignore it)

The kinds are identifier, location, health, contact and viewing. As before, an identifier field whose value is ZERO_AD_ID (or empty) does not count. Every owner the app contacted appears in the graph, even with an empty list, because a connection alone hands over your IP address and the time.

A score you can compare

A single number lets you compare two apps at a glance. Ours is simple on purpose: for every owner other than the app's own company, add 1 for the contact, plus a weight for each kind it received, from KIND_WEIGHTS (identifier 2, location 3, health 3, contact 3, viewing 2). FitFox scores 26.

Real researchers weigh things differently, and no score captures everything. A score is a conversation starter: it is more useful for "this app shares twice as much as that one" than as a verdict.

A common misconception

"More hosts means more sharing." An app might use ten hosts all run by itself and one host run by a data broker. Counting owners, and what they received, is closer to the truth than counting hosts.

function ownerOf(host, owners) {
  const labels = host.toLowerCase().split(".");
  for (let i = 0; i < labels.length - 1; i++) {
    const name = labels.slice(i).join(".");
    if (name in owners) return owners[name];
  }
  return "Unknown";
}

fn owner_of(host: &str, owners: &HashMap<&str, &str>) -> String {
    let labels: Vec<&str> = host.split('.').collect();
    for i in 0..labels.len().saturating_sub(1) {
        if let Some(o) = owners.get(labels[i..].join(".").as_str()) {
            return o.to_string();
        }
    }
    "Unknown".to_string()
}