Ask a grandparent when the family left the old village and you will get a confident year. Find the ship's passenger list and it says two years later. Both are real evidence. Neither is simply wrong. The grandparent may be remembering when the decision was made, or when they felt they had left, or may have been told the story by someone else. The list may have been compiled at the port, from what clerks wrote down on a busy morning.
That small puzzle is the whole discipline of history in miniature. You have traces. The traces were made by people for their own reasons, some of them long dead. They disagree. They are incomplete, and the missing parts are not missing at random. A historian is someone trained to ask of each trace: who made this, when, for whom, and what could they actually have known? Then to ask what else would have to be true if the trace were right.
This is the first piece in the History major, because every other history piece in the library, from the first cities to the twentieth century, leans on it. It is not a list of techniques to memorise. It is five habits of mind, each of which you can try on a small invented town below.
Historians sort evidence into primary sources (made at the time, by people who were there or close to it: a diary, a law, a coin, a building) and secondary sources (made later, by people working from other evidence: a textbook, a documentary). The line is not fixed. A 1828 anniversary speech is secondary for a flood in 1788 and primary for how people in 1828 wanted to remember it. Whether a source is primary depends on the question you are asking.
The core move is provenance: where did this come from, and who has handled it since? Three questions do most of the work. Who made it, and what could they see? When, relative to the event? Why, and for what audience? A private diary, an official minute, a fundraising pamphlet and a burial register can all describe the same disaster, and each can answer one kind of question well and another badly. Marc Bloch, the French medievalist, compared the historian to a detective questioning witnesses; in The Historian's Craft (written in 1941–43 and published in French in 1949, after the Gestapo shot him in 1944) he stressed that a source says nothing until someone asks it something.
Try it on Harbourton, an invented port. A flood hit it in 1788. Four sources survive. For each, pick what it is best at telling you.
This source is best for answering:
Harbourton and all four sources are invented for this exercise. The point is the questions, not the town.
Without JavaScript: the diary is best for how a person experienced the night; the council minute for who was officially blamed; the 1828 newspaper for how the town later chose to remember; the burial register for roughly how many died in the week.
Notice that no source in the game was "the reliable one". The register is the best guide to deaths in the week and the worst guide to why the flood happened. The diary is vivid, close in time and silent on anything its writer could not see. The newspaper is wrong about the numbers and the best evidence there is for the town's memory. This is what historians mean by reading a source for what it is.
One source is a claim. Two that agree feel like proof, but only if the two are independent. Medieval chroniclers copied each other freely, often without saying so. If six chronicles tell the same story and all six descend from one monk's notebook, you have one witness repeated six times. Ranke's early work on the Renaissance historian Guicciardini made this point in 1824: he showed that an apparently rich later account depended on earlier sources, and that where those earlier sources survived, they were the ones to use.
Here is the idea as arithmetic. Suppose each independent witness is right with some probability r (say 70%). If the witnesses are independent, each additional agreeing one multiplies the odds that the event happened by r / (1 − r). If they are copies, they add nothing. The tool below uses a deliberately simple model, starting from 50/50.
A toy model. Real witnesses are neither fully independent nor fully dependent, and a hostile witness who still confirms something is worth more than a friendly one.
Without JavaScript: at 70% reliability and a 50/50 start, one witness gives 70%; three independent witnesses give about 93%; six copies of one account still give 70%.
Two refinements matter in practice. First, enemy testimony: an admission from someone who gains nothing from it, or loses by it, is worth more. Second, convergence of kinds: a chronicle, a coin hoard and a burnt layer in the ground agreeing is stronger than three chronicles, because they could not have copied each other.
"When" comes from several tools, used together. Written chronologies give a framework: king lists, regnal years on coins, the lists of Assyrian annual officials, which are tied to the real sky by an eclipse recorded in 763 BCE. They can also mislead; the Sumerian King List gives early rulers reigns of tens of thousands of years. Dendrochronology counts tree rings. Wet years make wide rings, dry years narrow ones, and the pattern is a barcode that overlaps between living and old timbers; continuous European oak and pine chronologies now reach back more than twelve thousand years. Radiocarbon dating reaches objects with no writing at all.
Here is how the last one works. Cosmic rays make a little radioactive carbon-14 in the air, which plants take up and animals eat. While an organism lives, it keeps replenishing it. When it dies, the supply stops and the carbon-14 decays: its half-life is about 5,730 years. Measure how much is left and you can run the clock back. The method was developed by Willard Libby in the late 1940s (Nobel Prize in Chemistry, 1960).
Two honest complications. The amount of carbon-14 in the air has not been constant, so a raw radiocarbon age must be calibrated against a curve built from tree rings, corals, cave deposits and lake sediments; the current standard is IntCal20 (Reimer and colleagues, 2020). And tiny contamination from modern carbon makes old samples look young. Play with both.
Conventional-age arithmetic is exact. The calendar conversion uses a rounded, schematic curve (a handful of points, interpolated), good to a few hundred years at best. Real work uses IntCal20 in OxCal or CALIB. Measured amounts below about 0.2% of modern are in practice indistinguishable from background.
Without JavaScript: after one half-life (5,730 years) half the carbon-14 is left; at about 50,000 years only about 0.2% remains, near the limit of the method. A sample 50,000 years old mixed with just 1% modern carbon reads as roughly 35,000 years.
One more consequence: a radiocarbon result is a range with a probability, not a date. Between about 800 and 400 BCE the calibration curve is nearly flat (the "Hallstatt plateau"), so one measurement can map onto several centuries. That is why archaeologists combine it with tree rings, layers in the ground and written records.
Why did it happen? Historians argue about this more than about anything, and the argument has a structure. A but-for test asks: would it have happened without this? A factor that fails the test is not a cause in that strict sense. But most real events have several factors, none sufficient alone. The philosopher John Mackie named the usual shape in 1965: a cause is an Insufficient but Necessary part of an Unnecessary but Sufficient condition (an INUS). A match is struck, but only because there was oxygen, dry wood and no rain.
Then there is contingency: the part that could easily have gone another way. Debates about the First World War are the standard case. Fritz Fischer (1961) stressed German aims; Christopher Clark (2012) stressed a web of decisions across several capitals. Neither denies that Sarajevo was a spark; they disagree about how dry the forest was. Counterfactuals ("what if Princip had missed?") are a tool for testing that, not a game of alternate worlds.
Here is a miniature in Harbourton. The flood did damage because of heavy rain and either a blocked culvert or a spring tide. Deaths were high if there was flood damage and no warning bell. Switch conditions on and off.
Rule used: flood = rain AND (culvert OR tide); many deaths = flood AND no bell. An invented logic for illustration; real histories have no such clean rules.
Without JavaScript: with rain, a blocked culvert, a high tide and no bell, removing the culvert changes nothing (the tide still floods the town), yet the culvert is still part of a sufficient set. Removing the rain prevents everything.
The lesson to carry out: "the cause" is usually a choice about which conditions to call background and which to call triggers, and that choice is often where the real disagreement hides.
The archive is not the past. It is what got written, then kept, then survived fires, floods, floods of paperwork, and being thrown out. Michel-Rolph Trouillot, in Silencing the Past (1995), said silence enters at four moments: when facts are created (who writes), assembled (what an archive keeps), retrieved (what historians choose to tell) and given significance in hindsight. His striking example was the Haitian Revolution, which many people in Europe in the 1790s found literally unthinkable and so wrote about in ways that hid what it was.
Chance can preserve as well as destroy. Ashurbanipal's library at Nineveh (7th century BCE) survived because the fire that destroyed the city baked its clay tablets hard. The rubbish heaps of Oxyrhynchus in Egypt, excavated from 1896 by Grenfell and Hunt, gave hundreds of thousands of papyrus scraps: receipts, letters, school exercises. The storeroom of a Cairo synagogue, the Geniza, preserved documents of ordinary Jewish and Muslim life across centuries because the community could not bring itself to discard writing that might contain a holy name. Where writing did not carry history, other records did. Inca administrators kept accounts on knotted cords (khipu), and many West African societies preserve long, trained oral traditions; Jan Vansina's Oral Tradition as History (1985) set out how to test those against other evidence.
The tool below shows how thin the surviving voice can be, even with generous assumptions. It models 1,000 invented Harbourton residents in four groups, with an invented chance of being able to write, and of the writing being kept.
All numbers invented. Gentry and officials 100 people, tradespeople 300, labourers and servants 450, poor, enslaved or transient 150. "Mentions" adds tax lists, court records and registers, which name people who never wrote anything. Figures are expected values, not a simulation.
Without JavaScript: in the default model the gentry are 10% of the town and about two thirds of surviving personal writing; the poorest 15% leave less than one person’s worth. Counting mentions in other people’s records lifts the poorest to about 5% of what survives, though only as seen by someone else.
Historians respond in two ways. They read against the grain, using a record for what it reveals unintentionally: a court case written by a magistrate shows a servant's words, filtered. And they go looking for other kinds of evidence: archaeology, language, oral tradition, the bodies themselves. Neither closes the gap. What they do is tell you honestly where it is.
Forgeries are history's stress test. In 1440 the Italian humanist Lorenzo Valla showed that the Donation of Constantine, a document by which the emperor supposedly gave the pope rule over Rome, could not be a fourth-century text: its Latin used words and phrases from later centuries. In 1983 the German magazine Stern published extracts of "Hitler's diaries"; within weeks German archive laboratories found modern paper and binding, and the forger Konrad Kujau was caught. Neither case needed a confession. Each was settled by the same questions: who made this, when, and does it fit everything else we know?
History is an argument from leftovers. A source is judged by who made it, when and why; agreement counts only between independent witnesses; dates are ranges with methods behind them; causes are a set of conditions we choose among; and the gaps in the record are part of the evidence.
Try this: take one family story and find one document that touches it. Write down how each one came to exist, and whether they are independent.
Sources. Bloch, The Historian's Craft; Ranke's appendix on Guicciardini to his 1824 Geschichten der romanischen und germanischen Völker; Mackie, "Causes and Conditions," American Philosophical Quarterly 1965; Fischer, Griff nach der Weltmacht (1961); Clark, The Sleepwalkers (2012); Trouillot, Silencing the Past (1995); Vansina, Oral Tradition as History (1985); Reimer et al., "The IntCal20 Northern Hemisphere radiocarbon age calibration curve," Radiocarbon 62 (2020); Libby's Nobel Prize in Chemistry, 1960. Radiocarbon arithmetic computed in node. The Harbourton sources, rules and numbers are invented.