Field OP — Labelling Terms
Version: 0-draft Published at: /terms/labelling/0-draft. Shown at Admin → Labelling between 11 and 18 September 2026. Recorded as: name-labelling-2026-09-v1, in tenant_inference_settings.terms_version. Superseded by: 1.0-draft, which is section 9 of the Terms of Service whole.
Why this version exists here. These ten clauses were written as TypeScript constants in apps/web/src/lib/labelling-terms.ts and rendered on the labelling screen. An administrator who turned labelling on while they were current has name-labelling-2026-09-v1 recorded against their organization's inference settings, and that string has to keep resolving to the words they read. On 18 September 2026 the clauses moved out of the console and into this file, unchanged, so that it does. The headings and the paragraphs below are the clause headings and bodies exactly as they were shown; the header above this line was never on the screen, and no hash of this file was ever recorded — tenant_inference_settings records a version string and not a wording hash, which is one of the differences between this and the four documents accepted at sign-up.
This version is superseded and must not be shown to anybody as current. It is here to be read back, and for that reason only.
What you are turning on
Field OP will read the family name of people in this organization’s records and record what it guesses about their language, their community, and their religion. It is a guess made from a name and nothing else — not from anything they told you, and not from anything they did.
A guess is never recorded as a fact
Every label is stored apart from the person’s record, marked as inferred, with the model and version that produced it, the reasoning behind it, and a date it stops being current. It is never written into a name, a language field, or anything else the organization holds as fact, and a recorded fact always overrules a guess.
It is wrong often, and unevenly
On the test population this model is right about 88% of names from one community and 62% from another — a gap of twenty-six points that a single average would hide. It is wrong more often about some communities than others, which means the people it misreads are not randomly distributed. You can see the current figures, per community, on this screen before and after you turn it on.
Religion is the weakest of the three
A name can suggest a community. It cannot tell you what somebody believes, or whether they believe anything at all. Field OP produces religion by applying a published assumption about a community to a person, and in most cases it will decline to name one answer and show a spread instead. Treat a religion label as the least reliable thing this platform will ever show you.
Who can never see a label
No volunteer, team captain or candidate can see an inferred label by any route — not on a screen, not in a mission on a phone, not in an export, not in a printed walk sheet. The platform enforces this in the database rather than in the interface, so it holds even for software written later.
Nothing is sent anywhere
The model runs inside Field OP on Canadian infrastructure. No name, and nothing else about any person, is sent to a third-party classification service, to a general-purpose AI provider, or outside Canada, for this feature or any part of it.
The person can overrule it
A label can be confirmed, corrected, rejected or switched off for any individual, by anyone in your organization with access to their record. A rejected label is not produced again by the same version of the model. If a person tells you that you have their language, community or religion wrong, the correction is theirs and the platform records it as theirs.
What you are agreeing to do with it
You will use these labels only for lawful, non-discriminatory purposes connected to this organization’s stated work — for example, offering material in a language somebody is likely to read. You will not use them to exclude anyone from contact, to target anyone on the basis of religion, or for any purpose that would be unlawful if the person had told you the same thing directly.
Not for a real campaign yet
A legal opinion on this feature has been requested and has not arrived. Until it does, turn this on only against test data. Every label the platform writes carries a marker saying so, and the accuracy figures on this screen are measured on an invented population that was built to be classifiable — a real one will be read less accurately, not more.
Turning it off, and taking it back
You can switch this off at any time, and labels stop being produced immediately. Labels already written remain on record so that this decision can be audited and so that anything acted upon can be found — they can be deleted, but that is a deliberate act and not a side effect of switching off.
Agreeing is recorded against your name, with the version of these terms and the moment you agreed, and it is what authorises every label the platform then produces.