Import people from the FileMaker Rolodex archive

[maebeale]

🤖 suggested review level: 5 Inspect 🔬 a one-off import that now writes most of a person's profile, plus a save-time default on two shared models that changes one public-form question's visibility

Adds fm:import_people, which turns FileMaker Rolodex archive rows into people so the old contact list lives in the portal instead of only in the read-only archive.

What is the goal of this PR and why is this important?

• fm_rolodexes holds ~39k contacts, none of which exist as Person records yet.
• The archive's contact types don't all map onto ours and plenty of its rows carry none, so the import needed a place to put "we don't know" — which turned out to be a gap everywhere, not just here.

The import

• PersonServices::ImportFromRolodex does the work; the rake task wraps it in a dry-run transaction.
• Dry run by default — DRY_RUN=false writes; VERBOSE=true lists every row.
• Idempotent — a row whose ID a person already carries is left alone, reading the code column as the list it is.
• Links, doesn't duplicate — matching goes through PersonMatcher, so a person the archive knows by their secondary or login address is linked rather than recreated. Names run through Person.normalize_name first.
• Flags, doesn't guess — several matches, an email that belongs to someone else, and already-linked people are reported as ambiguous.
• Skips agencies (isAgency__c), soft-deleted (Delete), and nameless rows.

What an imported person carries

Everything FmRolodex::PORTAL_FIELDS maps onto a person:

• Identity and profile — names, emails, email type, pronouns, best time to call, ethnicity, DOB, comments → notes
• FacilitatorSinceMonth + Year → member_since (a year with no month reads as January)
• Service areas → sectors and PrimaryAgeGroup → age group, through SubmissionTagging — the same seam a submission uses, additively
• The archive's own DateEntered / DateTimeModified as created_at / updated_at, so a profile dates from when FileMaker first knew the person
• Every address and phone the row answers to (FmRolodex#addresses / #phone_numbers), the row's primary first
• A service area no sector answers to is counted in #unmatched_tags and printed by the task — Child Abuse resolves to Child Abuse/Neglect, but a name nothing answers to is a gap for a human, not a guess

Reading the archive's messier fields

• Dates are matched whole against four formats, so a format that only fits part of the value can't claim it (10-5-1983 used to read as 1910-05-19). A two-digit year gets a century: 83 → 1983, 05 → 2005.
• Free text is truncated at the 255-character column limit. An over-long address line or note used to fail the whole person.
• StreetNo is prepended to the first address line unless that line already starts with it.

Unknown as a type (worth the closest look)

• Address and ContactMethod write "unknown" on save when the type is blank, from any path — no backfill, so rows nothing saves keep their blank.
• "unknown" is new to ContactMethod::CONTACT_TYPES, so an Unknown option now appears in the contact-method type select. Address already declared it.
• Address#known_type? is why this isn't a silent regression: person_known_identifiers hides logged_out_only questions a registrant has already answered, and treating "unknown" as an answer would have stopped the public registration form from asking for an address type.
• AddressDeduper and ReconcileContactMethods read "unknown" and blank as the same type, so a pre-change row still folds into its twin.

Anything else to add?

• update_columns for the archive dates is deliberate: tagging touches the person, so the dates are written once more after it.
• PersonMatcher#matches is new — the relation behind #call, for a caller that must tell one match from several. Existing callers unchanged.
• A row whose email matches nobody is created even when someone shares the name — deliberate. Folding the occasional duplicate with the person deduper beats hand-resolving every name collision in the archive.
• EnteredBy / ModifiedBy → created_by / updated_by are left unmapped: they're FileMaker usernames, with no reliable path to a portal User.
• Run the dry run against real data before DRY_RUN=false; db:seed:dev:fm_archive gives dev something to run it against.
• Affiliations from OrgID are out of scope.

🤖 Generated with Claude Code

[jmilljr24]

Should this also handle worksite address?

rolodex rows                 39,795
  with PrimaryAddrsID         1,480
  with WorksiteAddrsID        1,217

Also I think these are just optional pointers. I think we need to iterate on the fm_address table. FmAddress.where(RolodexID: fm_id) "RolodexID" (26,427 rows → 23,918 people).

Same thoughts on fm_phone_numbers.

Or maybe the intent of the pr is to just import based on the rolodex table and do other tables separately?

[jmilljr24]

A number of "other" email types that would be dropped in the import. Some might be worth mapping to our values.

Preferred       520   
Professional     33   
Business          1
Attn: Carrie      1
Karen's email     1
Carrie            1
Bruce             1

[jmilljr24]

We can pull in address county information from the fm tables.

FmAddress
FmPostalCode  data["PostalCodeLookup"] → FmPostalCode.fm_id (zipcode)
County          FmPostalCode.data["County"]