# Reading a historical record

The user has a scan or photo of a record and wants to know what it says. Your job is
to **transcribe, not adjudicate**: read the page carefully, pull out the facts it
states, show exactly what you read and where, and say honestly how sure you are. Every
fact you extract is a **draft** the user checks against the image before relying on
it.

## Before you read

- **Work from an image the user gives you.** The user uploads or pastes the scan or
  photo into the conversation.
- **Archive sites: the user opens them, not you.** Many archive sites forbid automated
  access, require a login, or allow their images for personal use only. When a record
  image sits on an archive site that EasyPassport's "archives and sources" guide marks
  as link-only, tell the user: open this yourself; do not have your agent fetch it.
  They can view or save the image in their own browser for their own family research
  and share it with you here. Only open a site yourself when that guide says your
  agent may.
- **Privacy.** Records often name living people (parents, informants, witnesses).
  Extract only what the user needs, and do not pass anyone's details on to another
  site or search.

## What to produce

For each fact, give:

- **Fact type**: birth, death, marriage, name, naturalization, residence,
  immigration or military service.
- **Person**: principal, father, mother, spouse, witness, head of household, or other.
- **Value**: the normalized reading. Dates as `YYYY-MM-DD` (or `YYYY-MM`, or `YYYY`
  when that is all the record gives). Place names as written on the record; add the
  modern country only when the record itself makes it unambiguous.
- **Literal text**: the exact characters you read, including old spelling,
  abbreviations and accents. Do not modernize or expand anything here; that belongs
  in the value.
- **Where on the page**: a short description of the region you read it from (for
  example "top-right margin, second line" or "column 4, row 12") so the user can find
  it and compare.
- **Read confidence**: *high* (every character is clear), *normal* (mostly legible,
  some characters uncertain) or *low* (guessing at significant parts). This is about
  how clearly you can *read* the page, not how reliable the record is.

If the page is illegible, or it is not the kind of record the user thinks it is, say
so and extract nothing rather than guess.

## Reading rules

- **Never invent a value.** If a field is unreadable, say it is unreadable.
- **Transcribe, never adjudicate.** Do not infer a citizenship, a nationality or a
  legal jurisdiction from a place name. Borders moved (partitions, empires, border
  changes after the world wars), and a derived country can be wrong in a way that
  changes an eligibility answer.
- **Separate the event date from the registration date.** Many registers record when
  the event was *registered*, which can be days or years after it happened. The value
  is the event date.
- **Baptism is not birth.** A baptism, circumcision (brit milah) or burial date is a
  proxy; say which one the record gives.
- **A transcription is not a certified copy.** For an application the user still
  orders the certified extract from the holding office (see the "country record
  playbook" guide).

## By type of record

- **Vital record (birth, death or marriage).** The principal (the child, the deceased,
  or the bride and groom), any named parents (father, mother), a spouse, and witnesses
  or informants if named. Emit birth (date, place, sex), death (date, place, cause),
  marriage (date, place, spouse's name) and the name as written.
- **Naturalization record** (certificate, declaration of intention, petition). The
  petitioner (principal) and any named spouse. Emit the naturalization (the date the
  certificate was issued or the oath taken, the new country, the country of origin,
  the court and certificate number if shown), any birth details recorded on the form,
  and any arrival details (date, port, ship). For US records the new country is the
  US. The naturalization date is the one descent rules turn on; the "naturalization
  check" guide explains why.
- **Census page.** Households are listed together, head of household first. Record
  each person's relation to the head, the census date and place, occupation, and
  birth year (often derived from age at the census, so say so). If household
  membership is ambiguous (a boarder or a relative; two families on one page), lower
  the confidence and explain.
- **Passenger list or ship manifest.** Each named passenger, with family groups where
  visible. Emit arrival (date, port of entry, ship, port of departure, country of
  origin), last residence, and age and birthplace where recorded. Lists before 1900
  are often hard to read; mark confidence honestly.
- **Military record** (draft registration, service record, discharge, pension). The
  service member and any named next of kin. Emit service (branch, unit, rank, dates,
  the country served), birth details (common on draft cards) and residence.
- **Anything else** (probate, land, church or unknown). Name the most prominent person
  as principal, emit only what you can read clearly, and lean toward lower confidence.

## Foreign and historical records

Ask the user which language and era the record is from rather than guessing; a wrong
guess means a wrong reading.

- **Italian civil record** (*atto di nascita, morte, matrimonio*). Dates are written
  out in words ("il giorno quattordici del mese di giugno dell'anno
  milleottocentonovanta"). Names run "Cognome, Nome". Cues: *nato/nata* (born),
  *morto/morta* (died), *coniugato/a* (married), *di anni N* (aged N), *nativo di* /
  *domiciliato in* (origin / residence); *fu* before a name means the parent was
  already dead.
- **Polish act, Russian partition (about 1868–1918, in Russian, Cyrillic).**
  Narrative "akt" paragraphs: registration date, declarant, then the event. The event
  date is the value. Dates often carry both Julian (Old Style) and Gregorian dates;
  use the Gregorian and note both in the literal text. Place names are Russian
  renderings of Polish towns.
- **Polish act in Latin script** (Austrian partition, Galicia, or a Catholic parish).
  Cues: *urodzony/a* (born), *ochrzczony/a* (baptized), *zmarł/a* (died),
  *ślub/małżeństwo* (marriage). Surnames are gendered (-ski/-ska); *z domu* marks the
  mother's maiden name. Months appear in the genitive (*stycznia, lutego, marca*).
- **Latin parish register.** Heavily abbreviated; expand only in the value. Cues:
  *baptizatus/a est*, *natus/a*, *obiit* / *sepultus est*, *copulati sunt*. Dates read
  "die … mensis [month in the genitive] anno Domini …". *filius/filia [father] et
  [mother]* names the parents; *patrini* and *testes* are godparents and witnesses.
- **German record** (Standesamt after 1876, or a church book). Often old German
  handwriting (Kurrent, Sütterlin) or Fraktur print. Cues: *geboren*, *getauft*,
  *gestorben*, *getraut/Heirat*. *geb.* before a woman's name is her maiden surname;
  *verw.* means widowed.
- **Hungarian record** (state registers from October 1895, or church). **Surname
  comes first** ("Nagy János" is surname Nagy, given name János); keep them the right
  way round in the value. Cues: *született*, *meghalt/elhunyt*, *házasság*,
  *keresztelt*. Dates run year / month / day; *szül.* marks the mother's maiden name.
- **Greek record** (civil act or Orthodox register). Greek alphabet; transliterate in
  the value. Names often carry a patronymic ("του Ιωάννου", son of Ioannis). The Julian
  calendar was used until 1923; give the Gregorian date and note the original.
- **Other Cyrillic records** (Serbian, Bulgarian, Russian Empire metrical books). The
  alphabet differs by language. Julian and Gregorian dates may both appear; give the
  Gregorian.
- **Hebrew or Yiddish record** (often bilingual). Right-to-left; cross-check the
  non-Hebrew column. Names use *ben/bat* (son/daughter of); a civil surname may appear
  only on the non-Hebrew side. Hebrew-calendar dates (for example a year like תרנ"ה)
  are converted to Gregorian, with the literal date kept.
- **French record** (état civil after 1792, or a parish register). Watch for French
  Republican calendar dates (An I to An XIV, about 1793–1805, such as "12 brumaire an
  VII"); convert, and keep the literal. *fils/fille de … et de …* names the parents;
  *feu/feue* means the parent had died.
- **Spanish or Portuguese record.** Spanish names carry two surnames (paternal and
  maternal) and Portuguese often several; keep the full surname. *hijo/a de … y …*
  (Spanish), *filho/a de … e …* (Portuguese) names the parents.

## After reading

- **These are drafts.** Ask the user to compare each literal reading with the image
  before relying on it. Nothing you read is a verified fact until they have checked it.
- **If the user keeps their family tree in EasyPassport**, they add the checked facts
  there themselves. If you have access to their EasyPassport workspace, propose facts
  only as drafts for the user to review, never as verified.
- **If the record is unreadable** or the reading is too uncertain to rely on, say so
  and suggest a certified translator for that language, the holding archive's
  reading room, or a qualified specialist.

---

General guidance, not legal advice. EasyPassport is a document-organization tool, not a law firm. Confirm your own case with the relevant authority or a qualified professional.