Models

DfRecord

A pydantic model of one Diplomatarium Fennicum charter. Field names follow the source JSON, so a row can be traced back to the export line it came from; four columns are derived.

DfRecord.from_json(row: dict) -> DfRecord

Builds a record from one line of df.jsonl.gz.

Source fields

object_id, df, transcript, indexterm, issuingplace, issuingplacecountry, language, dating_start_year, dating_end_year.

Absent text fields are empty strings, not nulls — the source convention, preserved.

Derived fields

df_number — df parsed as an int. The DF number is numeric for all 6,876 charters, but as a string "10" sorts before "2". The int form is what a range filter or an ordering uses; the string stays authoritative for citation.

year_from / year_to — the dating interval with the unknowns closed, so a date filter is a plain two-column overlap test rather than a COALESCE over nullable columns. A charter with a start year and no end year is dated to that single year, so year_from == year_to; an absent start year with a known end is treated symmetrically. Both are None only when the source gives no year at all.

lat / lng — _geoloc flattened. The source nests them and sets both to null together (never one alone) for the 2,734 unlocated charters, which flat nullable columns represent exactly while remaining filterable.

searchable_text (a property) — the transcript plus place, country, index term and language. Deliberately more than the transcript: 2,464 charters are catalogued but untranscribed, and an index over transcript alone would make them unreachable by search even though they are perfectly findable by their catalogue fields.

Year normalisation

0 in a year field means "unknown", not year 0. Left as a literal 0 it silently pollutes every date range — 6,843 of the 6,876 charters carry dating_end_year == 0 — so it is normalised to None on the way in, together with blanks and unparseable values.