Document Facts vs Inferences in Civil War Research

A compiled service record rarely says what we most want it to say. It says a name appeared on a roll on a date, in a unit, at a rank. Everything else, the story, the motive, the presence at a battle, is something a reader adds. The discipline worth building is simple to state and hard to hold: keep the two apart on the page, not just in your head.

Why does the difference between fact and inference matter?

Because a fact can be checked by someone else and an inference cannot. A fact is a statement you can point to on a specific document. An inference is a conclusion you drew from one or more facts. Both are legitimate parts of research. The problem starts when they are written in the same voice, because a later reader cannot tell which is which, and neither can you after six months. Once an inference has been typed into a clean sentence, it starts to feel like evidence. That is how a guess quietly becomes a family legend.

The National Park Service makes this distinction visibly in its own database. Its unit histories link soldiers to regiments to battles, and the site states plainly that this linking does not indicate with certainty whether a soldier participated in a particular battle, only that the regiment was present (NPS overview). That is a model for your own notes: describe what the record structure supports, then say what remains to be proven.

How do I mark the line in my own notes?

Use two physically separate places. One for transcription, one for reasoning. A workable pattern:

• Fact line. Quote or closely transcribe what the document says, with the source and image cited. Keep it short. Do not smooth the wording.
• Question line. Write the open question the fact raises.
• Inference line. Write your conclusion, prefixed so it can never be mistaken for a transcription.
• Test line. Name what would confirm or kill the inference.

Prefixes matter more than formatting taste. If every inference in your file begins with a fixed word, scanning for them later takes seconds. If they are scattered in prose, you will read right past them.

The lines below show how the same record produces very different notes depending on where the line is drawn.

What you are writing	Example wording	Can another reader verify it?
Transcription fact	"Name appears on the company roll for the stated period, rank private."	Yes, against the image
Index-level fact	"General index card gives rank in and rank out."	Yes, against the card
Structural fact	"The regiment is linked to this battle in the unit history."	Yes, against the site
Inference	"He was probably present at the battle."	No, this is yours
Inference	"The spelling change suggests the family moved."	No, this is yours
Open question	"Was he still on the rolls in that month?"	N/A, it is a task

What counts as a fact when the document itself is a copy?

A transcription is a fact about the transcription. This trips people up constantly. When you read a card rather than the original muster roll, your solid statement is "this card reads X," not "the army recorded X." Those are close but not identical claims.

The National Park Service describes the chain clearly. Muster rolls were kept by units during the war; in the 1880s a staff working under General Ainsworth indexed those rolls, writing a card each time a soldier's name appeared, to help determine pension eligibility (NPS records origin). The cards, not the fragile rolls, are what most researchers actually use. So a card is a witness to a roll, and a roll is a witness to an event. Each step is a place where a fact can be copied correctly and still differ from the underlying reality. Naming which layer you are standing on keeps your notes honest.

How should I handle rank in and rank out?

Treat them as two dated facts, not one career. A general index card lists the rank at the time of enlistment taken from the first card and the date the soldier left the service with the final rank taken from the last card (NPS records origin). That is a summary of two endpoints. It tells you the two ends and nothing about the middle. Promotion, reduction, absence, sickness and transfer all live between those endpoints and each has its own documentary trail.

So the honest note is: "rank in private, rank out corporal, per general index card." The dishonest note is: "rose from private to corporal during the war." The second adds a story the card does not carry. For more on reading those two fields, see What Rank In and Rank Out Can Tell You.

How do I keep my inferences from hardening into claims?

Date them, name them, and give them an expiry. An inference written today is a hypothesis. If nothing tests it, it should still read as a hypothesis a year from now. Practical habits that hold up:

• Write inferences in the present tense as your view, so they never read as quoted record language.
• Attach the specific fact each inference rests on, by citation, not by memory.
• Note when you last reviewed it, so a stale guess is visible as stale.
• Give each one a status word, such as open, likely, or discarded, and update it as evidence arrives.

A discarded inference is not wasted work. It is a documented dead end, which is exactly what stops the next researcher from walking it again.

Where does this matter most in Civil War records?

Three places, repeatedly. Battles, identity, and personal detail.

Battles are the big one. A regiment present at a battle does not mean your soldier was there. He may not yet have enlisted, he may have been detached, sick or wounded elsewhere. The NPS states that additional research is required to validate whether a soldier was actually enlisted at the time and present for duty in that battle (NPS overview). Build the habit early: Battle Links Do Not Prove Presence is the same warning applied to a single soldier's file.

Identity is the second. Name spellings shift between rolls, cards and indexes. A variant spelling is a fact about the documents; which spelling the man himself used is usually an inference. Index cards are finding aids, not verdicts, a point developed in The Index Card Is Not the Final Answer.

Personal detail is the third. Ages, birthplaces and occupations reported at enlistment were given under specific circumstances and are worth recording exactly as written before you decide what they mean.

What does a clean note look like at the end?

It reads like two people wrote it, because two did. One recorded what the document says. The other is you, thinking. A reader can follow the facts to their images and can evaluate your reasoning on its own merits. If you disagree with a conclusion, you can still use the facts, which is the whole point of separating them.

That is also what makes your work reusable. A source log that another reader can follow is built from exactly these parts: clearly labeled transcriptions, citations that resolve to an image, and inferences marked as yours. The article Build a Source Log That Another Reader Can Follow shows how those parts assemble into one document.

None of this requires certainty about the past. It requires only that you are honest about which kind of statement you are making at any given moment. Facts you can defend. Inferences you can revise. Keep them apart and both stay useful.