Historical figures: how to search and verify records

The hard part of researching a historical figure is almost never finding a claim. Claims are abundant. The hard part is finding out where a claim came from, and whether anything sits underneath it.

A great deal of what circulates about people who died centuries ago rests on a chain of sources that each cite the previous one, ending at a nineteenth-century compilation whose own source is unstated. The chain looks like corroboration. It is one assertion, copied.

Walk the citation chain backwards

Take any specific claim (a birth year, a place, a title, a relationship), and treat it as a lead rather than a fact until you have found its earliest attestation.

Note the claim and the exact wording.
Find the source your source used. If it does not name one, that is itself a finding: record that the claim is unsourced at this level.
Repeat until you reach either a document contemporary with the event, or a dead end.
Write down where the chain ends. "First appears in a compilation published two centuries after the subject's death, with no source given" is a genuinely useful sentence, and it is one that reference works almost never print.

Two patterns show up constantly on the way down. The first is inflation: a source says "possibly," the next says "likely," the third states it flat. The second is transposition: a detail belonging to one person migrates to another with a similar name, and then acquires citations of its own.

Names before standardized spelling

Search failure in pre-modern records is usually a name problem, not an archive problem. The record is there; your string does not match it.

Spelling was not fixed. One person's name can appear several ways within a single parish book, written by one clerk, over a few years. Search by sound and by initial letter, not by exact match.
Latinized and vernacular forms coexist. Church, university, and legal records may carry a Latin form while everything else uses the vernacular. The two look nothing alike at the start of the string.
Transliteration is not standardized across time or language. A name written in one script has multiple defensible romanizations, and older catalogs may use a system nobody uses now. Search each variant.
Patronymic systems do not produce inherited surnames. In several traditions the second name derives from the father's given name and changes every generation, so father and son do not share a "surname" at all. Sorting such records by surname produces nonsense.
Names change with status. Religious names taken on entering an order, regnal names, stage and pen names, names changed on migration or conversion. A person can hold several across one life, each dominating a different set of records.
Women are frequently recorded under someone else's name, or under a household head, or by a first name only. Absence from a name index is not absence from history.

Authority files exist to make this tractable: they collect variant forms under one authorized heading so that a search on any one of them can reach the rest. Name authority records are worth consulting for this reason alone, the list of variants is often the most useful part of the record.

Dates that do not mean what they say

Calendar reform did not happen everywhere at once. Jurisdictions moved from the Julian to the Gregorian calendar at widely different times, so for a long stretch the same day carried different dates in different places. A letter and its reply can appear to arrive before they were sent.
The year did not always begin on 1 January. Different jurisdictions used different year-start conventions, so a date in the first months of the year may belong to either of two years depending on whose convention the writer used. Historians handle this with dual dating; secondary sources often silently pick one.
Regnal dating gives the year of a ruler's reign rather than a calendar year, and converting requires a regnal chronology for that state.
"fl." is not a lifespan. A floruit is the span in which someone is documented as active, which is a statement about surviving records, not about a life.
"c." is a real qualifier. Circa means the date was reconstructed, from an age at death, an apprenticeship term, a matriculation age, a stated age in a deposition. Those reconstructions have known biases: stated ages were commonly rounded, and ages at death were commonly guessed by whoever registered it.

When you record a date, record the calendar convention and the type of date alongside it. A date without its convention cannot be safely compared with another one.

The archive is not a sample

The surviving record is not a random subset of the past. It is heavily conditioned by who kept documents, whose documents were thought worth preserving, and what escaped fire, flood, war, requisition, and deliberate destruction. It over-represents the literate, the propertied, the urban, the institutionally affiliated, and those whose records ended up inside a state that later funded an archive.

This matters for a specific reason: absence of evidence in such a record is very weak evidence of absence. "No trace of her before this date" usually means the registers for that parish do not survive, not that she appeared from nowhere.

Before concluding anything from a gap, find out what the gap is made of. Archive catalogs state their own losses. Read the collection-level description before the item.

Match the question to the record type

Different documents are evidence of different things, and the fastest way to waste a week is to search the wrong series.

You want	Look in	It is evidence of
Birth, marriage, death	Parish and congregational registers; civil registration where it existed	The rite or the registration, not always the event
Residence and household	Tax rolls, land records, censuses, rate books	Where a household was recorded, on one date
Occupation and training	Guild and apprenticeship records, matriculation and admission registers	Formal membership, not the whole of a working life
Property and money	Deeds, wills, probate inventories, notarial acts	The transaction, and what was worth recording
Disputes and daily life	Court rolls, depositions, petitions	What was contested, which is a biased sample of what happened
Movement	Passenger lists, passports and permits, military musters	A crossing, as recorded by the recording authority

Note the third column. A baptism register is direct evidence of a baptism and only indirect evidence of a birth. A muster roll proves a man was mustered. Keeping that distinction visible is the difference between a sourced statement and a plausible one.

Two people, one name

Identity resolution is where most confident errors are made. Before merging two records into one person, check that they agree on more than the name:

Do the dates permit one life, without requiring an implausible age?
Is the geography compatible with the travel and communication of that period?
Do occupation, status, and social network overlap?
Does any single document connect the two identities, or are you connecting them yourself?

Automated matching across catalogs makes both errors in both directions: it splits one person into several clusters, and it merges several people into one. When you rely on a matched identifier, say that you did, so the next researcher knows which link to test first.

Prefer the evidence to the label

The most durable way to write about a historical figure is to state what a document records rather than what a category asserts.

"The register of the cathedral chapter lists him among the resident canons in that year" is checkable, bounded, and stays true. "He was a canon" is a summary that has thrown away the source, the date, and the scope, and it will be copied without them.

This matters most for exactly the categories this site is organized around: origin, community, faith, allegiance. Those labels carried different meanings in the period than they do now, they were frequently assigned by an outsider with an interest, and they are the labels most likely to be projected backwards from a modern map. Even for people long dead, and even where the facts are not really in dispute, the sentence that reports the evidence outlasts the sentence that states the conclusion.

A verification pass that actually catches things

Run this over a finished draft:

Every specific claim has a source, and every source has a date and a scope.
Every source you cited, you opened. Nothing entered the draft from a summary of a summary.
Every date carries its calendar convention and its type.
Every "was" that assigns a category has been tested against the alternative sentence that reports the record instead.
Every gap is described as a gap in the record, not as a fact about the person.
Every identity merge names the document that justified it.
Anything you could not resolve is printed as unresolved rather than smoothed away.

Bottom line

Good work on historical figures is mostly bookkeeping: knowing which document you are looking at, what it is evidence of, and how far a claim travelled before it reached you. It produces shorter articles with fewer assertions, and those are the ones that survive being checked. The related pages on period sorting and occupation labels deal with the two fields that go wrong most often once the identification itself is secure.