Chapter 10 of 16 · The great periods of history
Chapter 9. The sources of the historian
How we know what we know: the three families of traces, the way to question them, and what their silence means.
Every statement comes from somewhere
No sentence about the past stands on its own. Behind each of them there is an object, a text, a layer of earth or a measurement, and often several at once. That support is what is called a source. A historical statement without a source is not a weak statement: it is a statement of another kind, belonging to invented narrative.
The official history syllabus published in 2026 makes it the heart of what a pupil must acquire, and the wording is precise: the pupil has acquired the bases of historical knowledge, namely telling fictional accounts apart from historical accounts, questioning the traces of the past and putting them in context by learning to question the reliability of sources, digital ones in particular. Three moves, and not one of them consists in remembering dates.
This chapter gives those three moves. It first presents the three families of traces on which everything this book has stated so far rests. It then gives the sequence of questions that makes it possible to interrogate any of them. It ends with the hardest and most useful point: what the silence of a source means.
The vocabulary of the source
Five distinctions, and they are what separates an attentive reading from a naive one.
- A source
- Everything left of the past that makes it possible to know something about it: an object, a text, a building, a bone, a layer of earth, an image, a recording.
- A first-hand source
- A source produced at the time of the facts, or by somebody who took part in them. A letter written on the day of a battle, a signed contract, a tool abandoned on the spot.
- A second-hand source
- A piece of work that speaks about sources without being one: a textbook, an article, a museum label. It is useful, it is not proof, and it must always state what it rests on.
- A trace
- What survives without having been left on purpose: waste, footprints, remains of meals, wear marks on a tool. A trace has no intention, which makes it hard to read and hard to fake.
- A piece of testimony
- An account given by somebody who claims to have seen. It brings what no object gives, and it brings at the same time the point of view, the memory and the interest of whoever is speaking.
| Family | What it establishes well | What it cannot establish |
|---|---|---|
| Objects | The techniques used, the materials available, long-distance exchanges, the gestures of making and of wear | Intentions, names, spoken words, beliefs beyond what the shape suggests |
| Texts | Names, dates, numbers, laws, decisions, what the authors wanted known | What nobody wrote down: the lives of those who did not write, and everything that went without saying |
| Excavations | The order of the occupations of a place, the organisation of space, diet, health, mortality | The detail of events, which almost never leaves a recognisable material trace |
No family is enough, and that is the point
Every row of the previous table carries a column of powerlessness, and these powerlessnesses do not overlap. That is exactly why the work consists in cross-checking. A text stating that a town was destroyed is checked against a burnt layer in the ground. A burnt layer without a text says neither who set the fire nor why.
Cross-checking has a simple rule, which holds outside history too: two sources only confirm one another if they are INDEPENDENT. Three chronicles copying one another are not worth three pieces of testimony, they are worth one, repeated. The question to ask when faced with two agreeing sources is therefore always the same: could one of the two have read the other.
This test alone rules out a large part of the errors, including today. A piece of information taken up by twenty websites citing one another has not been checked twenty times: it has been copied nineteen times. The official syllabus in fact asks that the reliability of sources be questioned, digital ones in particular, and it is exactly this mechanism that it has in view.
Questioning a source, in five questions
This order holds for a medieval manuscript as much as for an image found online. The first three questions bear on the document itself, the last two on what it says.
- Who produced it, and when
An author, a date, a place. If one of the three is missing, say so instead of assuming it. An anonymous, undated source is not worthless, but everything drawn from it carries that uncertainty. - For whom, and to what end
A harvest account, an official inscription, a private letter and a chronicle of a reign do not have the same addressee or the same purpose. The purpose steers the content, always, even when the author is sincere. - By what route it reached us
Was the original kept, or only a copy made later? A copy can correct, shorten or add. The manuscript of the taking of Constantinople cited earlier is in this way a copied collection, and not the original text of its informant. - What it states, and what it implies
Separate the facts stated from the judgements passed. "The city was taken on 28 May" is a verifiable fact; "the city was taken by the enemy's cunning" adds an appraisal that is not of the same order. - What another independent source says about it
This is the only question that can confirm anything. Look for a source that could not have been copied from the first. If you find none, the information remains an item from a single source, and that is said.
The two criticisms, and both are needed
The criticism of the document itself
- The support: what material, what condition, what wear.
- The writing or the making: do they match the age announced.
- The origin: where does the object come from, through whose hands has it passed.
- The form: is it an original, a copy, a translation, a reconstruction.
- Material consistency: a dated document cannot mention a later fact.
The criticism of what it says
- The author: their position, their interest, what they could know first hand.
- The addressee: who they are speaking to, and therefore what they can allow themselves to write.
- Internal consistency: does the text contradict itself, hesitate, correct itself.
- Distance: did the author see, or are they reporting what they were told.
- Silence: what the document should have mentioned and does not mention.
Case 1
Two correct dates for one and the same episode
The collection of historical texts held by the Bibliothèque nationale de France announces the taking of Constantinople on the 28th day of May in the year one thousand four hundred and fifty-three. The scientific notice drawn up by the library, for its part, writes that the capital fell on 29 May 1453.
The gap makes neither the document nor the notice false. The same notice writes that the main assault was delivered on 28 and 29 May 1453: the account dates the launch, the notice dates the fall. Faced with two dates that do not agree, the first question is therefore not which one is wrong, it is what each of them claims to date. Many apparent contradictions dissolve at that single question.
Case 2
An image faithful and invented at the same time
The same manuscript carries a large view of the siege. The library's notice observes that it has neither a consistent projection nor a consistent perspective, that the size of the buildings in it depends on their symbolic importance, and that it carries no scale but numerous indications of distances in the captions.
It nevertheless observes that some details could only have been known on the spot, and that this artist must have had first-hand information. One and the same image can therefore be an excellent source on one point and an invention on another. There is no such thing as a document that is reliable overall: there are points that have been checked, one by one.
Case 3
A date taken up again over more than eighty years
The scientific dossier published by the ministère de la Culture on the dating of the figures of Lascaux does not give a single figure: it gives the sequence of proposals made since 1940, from comparisons of style to carbon 14 measurements, then to accelerator measurement.
The dossier itself comments on this sequence: these successive adjustments show the difficulties met in establishing a precise, sufficiently argued diagnosis. Then it extends it: a dating programme carried out between 2018 and 2021 makes the whole thing younger again, by 1000 to 1500 years in its own words.
This dossier in fact carries its own trace of time: a section of quick questions, at the bottom of the same page, still gives the age established at the end of the 1990s, which the most recent section has just corrected. This is not a serious fault, it is what happens to any living document. An attentive reader therefore compares the sections of one and the same page before copying out a figure.
The silence of sources, and it has to be read
The most costly error does not consist in misreading a document: it consists in concluding from the absence of a document to the absence of a fact. The two look alike only to somebody who does not look at where sources come from.
Three main causes produce silence, and none of them concerns what happened. The first is the support: papyrus rots outside dry climates, wood disappears, iron rusts, whereas fired clay and stone cross the millennia. Damp regions look empty because nothing survives there, not because nobody lived there.
The second is sorting: what has been kept was kept by people who had reasons to keep it. The archives of a power keep what serves that power. The documents of the defeated, of the poor and of those who did not write have disappeared in far greater proportion.
The third is excavation: you only find where you look. A heavily excavated region looks rich in remains, a little-excavated region looks poor, and that difference measures the activity of archaeologists rather than that of the inhabitants. Faced with a blank, the right question is still the one from the chapter on the timeline: why is nothing written here.
Knowledge that corrects itself, and that is a sign of health
The official syllabus of 2026 states, about the way a subject must be studied, a sentence that sums up this whole chapter: it makes it possible to understand that knowledge is constantly fed and modified by new scientific discoveries and by renewed readings.
Two engines are named, and they are different. The first is discovery: a site excavated, a document found again, a finer method of measurement moves a date or adds a fact. The second is rereading: the same documents, read again with other questions, yield what nobody was looking for in them. The archives of a court first served to write the history of law, before serving to write that of ordinary people.
A practical consequence follows for the reading of any history book, including this one. A textbook gives the state of a knowledge at a date. That date must be written down, and the documents it rests on must be identifiable, so that a reader can go and see whether anything has changed since.
That is why every volume of this house carries a verification date and a sources chapter in which every document is given with its address and the date on which it was consulted. This is not a publishing formality: it is the only way to make what has been written verifiable.