Where We Left Off
A short recap for anyone starting here, a reminder of what the coloured boxes mean, and a warning that this part needs more than the others.
Part 2 was about a civilisation that vanished. The Harappans built brick cities across a million square kilometres, laid covered drains under their streets, cut standardised weights accurate to a fraction of a gram, and traded with Mesopotamia. Then, between about 1900 and 1300 BCE, the cities emptied out. Not burned. Not sacked. Just left.
Part 2 also ended on a set of loose threads, and it said so plainly. The script is still unread, so we cannot hear the Harappans in their own words. Nobody has found a king, a palace or a temple anybody can prove is a temple. The horse is missing from the bones. And a nineteenth-century theory that fair-skinned invaders rode in and smashed the cities was examined properly in the twentieth century and thrown away, because the evidence for it was not there.
Every one of those threads leads into this part.
Because roughly at the moment the last Harappan cities were being abandoned, something else appears in Punjab. Not a city. Not a building. A book — though it was not a book yet, because nobody wrote it down for a very long time. A collection of 1,028 hymns in an early form of Sanskrit, addressed to gods, composed by priests, memorised word for word and passed mouth to ear for centuries before anyone put ink on anything.
That collection is the Rigveda. It was composed in Punjab. It names our rivers. It describes a fight on the Ravi. It is the oldest surviving text in any Indo-European language, and it is the single most argued-over document in Indian history.
Part 3 is about the Rigveda, the world it seems to describe, and the question that hangs over all of it: where did the people who composed it come from?
What the boxes mean
Six kinds of coloured box run through this series. Each does one job. If you read nothing else, read the black ones.
- Word Box (slate blue) — explains a hard word the moment it first appears. You never have to already know something.
- In Real Terms (moss green) — turns a number or an idea you cannot picture into something ordinary.
- How We Actually Know This (sepia) — shows the evidence itself. The bone, the brick, the verse, the measurement. And what it can and cannot prove.
- The Argument (ochre) — where the experts disagree. Each side gets its strongest case, not a weak version I can knock over. Then a plain statement of where things stand and what would settle it.
- The Hidden Assumption (plum) — digs out something everybody takes for granted without proving. This is the part of the series I am most attached to.
- Remember This (dark) — closes every chapter with the key points in the plainest words I can find.
A warning specific to this part
This is the hardest part of the fourteen, and not because the material is complicated. It is hard because it is live.
The question of where Sanskrit speakers came from is not a quiet academic matter in India today. It sits underneath arguments about who is native and who is foreign, whose ancestors built what, and who has the better claim to the place. People shout about it. Books are written to prove a side. Governments have opinions.
So let me say very clearly what I am doing here, and what I am not.
I am not going to pick a side for you. I am going to lay out three positions — invasion, migration, and indigenous origin — and give each one the best case its own supporters would give. Then I will tell you what the evidence physically is, what it genuinely supports, and where each side is guessing. Where the evidence points somewhere, I will say so. Where it does not reach, I will say that too.
And I will separate two questions that get welded together and should not be.
The first is evidential: what happened, when, and how do we know? That question has answers, some firm, many not.
The second is about identity: are they us? That question feels like it depends on the first one. It does not. Populations mix constantly, everywhere, always. The honest answer to "are they us?" is: partly, along with several other groups, and this is true of nearly every population on Earth.
One more thing. As you read this part, notice which claims you find yourself hoping are true. That is not a criticism. Everybody does it, including me, including every scholar in every footnote I have read. But it is useful information. The place where you are hoping is the place where you are least reliable, and it is worth marking on your own map.
1The Book That Is Not a History Book
Everything in this part rests on one source. Before we ask what it tells us, we have to be honest about what kind of thing it is — and what kind of thing it is not.
1.1 — What the Rigveda actually is
Start with the physical facts, because they are not in dispute.
The Rigveda is a collection of 1,028 hymns. Each hymn is a set of verses — usually between three and about fifty. Added together the whole thing comes to a little over ten and a half thousand verses.
The hymns are grouped into ten books. Every hymn is addressed to a god, or to a group of gods, or to a thing treated as a god — fire, the dawn, a river, a plant. They are not stories about the gods. They are things said to the gods, out loud, during a ritual, by a priest who was being paid to say them.
That last sentence is the whole problem, and we will come back to it.
Veda: a Sanskrit word meaning "knowledge". There are four Vedas. The Rigveda is the oldest by a long way. The other three — Samaveda, Yajurveda, Atharvaveda — came later and borrow heavily from it.
Rigveda: literally "the knowledge of praise verses". A ric is a verse of praise.
Mandala: one of the ten books. The word means "circle" or "cycle". When you see "Mandala 7" it just means "Book 7".
Sukta: a single hymn. So "Rigveda 7.18" means Book 7, hymn 18.
Vedic Sanskrit: the language it is in. It is an older form of Sanskrit than the classical language that came later — different enough that classical Sanskrit scholars have to learn it separately, the way an English speaker has to be taught to read Chaucer.
The hymns ask for things. Cattle. Sons. Victory in a fight. Rain. Long life. Safe passage across a river. They flatter the god first, remind him of favours he has done before, and then make the request. If you have ever written a careful email to someone whose help you need, you already know the structure.
They also contain, scattered through the requests, incidental details about the world the composers lived in. Rivers by name. Enemies by name. Animals, plants, metals, weapons, food, weather. Nobody put those details in on purpose. They fell in around the edges of the asking.
Those accidental edges are almost all the evidence we have for four hundred years of Punjab's history.
1.2 — How you keep a book without paper
Here is the thing most people find hardest to believe, and it is true.
For a very long time — centuries, possibly more than a thousand years — the Rigveda was not written down at all. Not on bark, not on palm leaf, not on clay. It existed only inside people's heads, and it moved from one head to the next by being said aloud and repeated back.
The oldest physical copies of the Rigveda that survive are manuscripts from around the eleventh century CE. The hymns are roughly three thousand years older than that.
Your first reaction should be suspicion. Mine was. Anything passed by word of mouth drifts. Play the whispering game with fifteen people and the sentence at the end is not the sentence at the start.
So how do we know the text did not drift?
Because the people transmitting it knew about drift too, and built machinery to stop it.
They did not just memorise the hymns in order. They memorised them in several orders at once, each order a check on the others.
Take a line of four words: A B C D. One method has you recite it straight through. A second has you recite each word on its own, stripped of the sound changes that join words together. A third has you recite it in overlapping pairs — AB, BC, CD. A fourth has you go forwards and backwards through each pair — AB, BA, AB, BC, CB, BC. There is a fifth, harder still, that weaves the words in a fixed braid across the whole line.
A boy learning this started around seven or eight and was at it for years. He was taught the pitch of each syllable as well as the word, because pitch carried meaning and a wrong pitch was a wrong word.
What it proves: that a single slipped word breaks several patterns at once and is instantly audible to anyone else who knows the method. The redundancy is the error-check. It is the same trick a bank card number uses when the last digit is calculated from the others.
What it does not prove: that nothing at all was ever lost or added. It proves the text was locked down hard from the point the system was in place. What happened in the earliest generations, before the machinery existed, is not covered by any of this.
There is a second, independent check. Sanskrit grammar was analysed with extraordinary precision by later scholars, most famously Panini, several centuries BCE. Those grammarians could see that the language of the Rigveda was already archaic in their own time. They preserved forms they no longer used and could not have invented. A forger writing later would have written in his own grammar without noticing. The Rigveda's grammar is consistently older than the grammar of everyone who handled it.
Ten and a half thousand verses is hard to feel. Try this.
If you memorised ten verses a day — perfectly, with the pitch right, never missing a day — it would take you a little under three years to get through the Rigveda once.
The men who did this were expected to hold the whole thing. Then teach it. Some families did this, generation after generation, for roughly a hundred generations.
In 2008 UNESCO put Vedic chanting on its list of humanity's intangible heritage. It is, as a feat of engineering, on the same shelf as a cathedral. It just happens to be made of people.
1.3 — Why praise poetry is the worst kind of evidence
Now the hard part.
Suppose the only surviving document from twentieth-century Punjab were a book of songs sung at weddings. You would learn a lot from it. What people ate at feasts. What they thought a good marriage looked like. Which relatives mattered. The names of some villages.
You would also get a wildly distorted picture. You would think Punjabis spent most of their lives at weddings. You would have no idea about tractors, or courts, or how much anybody earned, or what the poor thought, because nobody sings about that at a wedding.
The Rigveda is that kind of document. It is liturgy, not reportage. It tells you what a small group of ritual specialists said to gods on important occasions. Everything else you get from it is accidental.
And the group is small. The composers were priests attached to chiefs. They were paid — often the hymn says what they were paid, in cows and horses and gold, and complains if the payment was thin. We are hearing from the professional religious class of the winning side, and from nobody else at all.
No farmer speaks in the Rigveda. No woman speaks in it more than a handful of times. Nobody on the losing side of any fight gets a line. Whole populations who were certainly there — the descendants of the Harappans, for a start — appear only as somebody else's enemy, if they appear at all.
That a text tells you about the world it was made in.
This one is so automatic that it takes an effort to see it at all. We open an old text, read a description, and treat the description as a window. It is not a window. It is a statement made by somebody, for a purpose, to an audience.
A hymn that asks Indra for a thousand cattle is not evidence that anyone had a thousand cattle. It is evidence that a thousand cattle was a number worth wanting. A hymn that calls the enemy "noseless" and "of dark skin" is not a census. It is an insult, aimed at a god, about people the speaker was fighting.
Ask three questions of every verse in this part:
- Who is speaking? Almost always a paid priest.
- To whom? A god, in front of a patron who is watching.
- What do they want? Something. There is always something.
This does not make the Rigveda useless. It makes it usable in a particular way. The things the composers were not trying to tell us are the things we can most safely take — the river names, the grammar, the incidental furniture of the world. The things they were trying to tell us are the things to be most careful with.
And this cuts against every side in this part's argument. Both the people who want the Rigveda to record a migration and the people who want it to record an unbroken native tradition are reading a wish list as a record.
1.4 — How old is it?
There is no date in the Rigveda. Nobody in it says what year it is, because nobody in it had a calendar that would mean anything to us.
So the date is worked out sideways, from four directions.
Language. Vedic Sanskrit is very close to the language of the Avesta, the oldest Iranian sacred text — close enough that whole lines can be converted from one to the other by regular sound swaps. Two languages that similar were one language not very long before. That anchors the Rigveda to the Avesta, and the Avesta has its own rough dating.
A treaty in Syria. Around 1380 BCE, a kingdom called Mitanni in what is now northern Syria signed a treaty. The treaty calls on four gods to witness it: Mitra, Varuna, Indra and the Nasatyas. Those are Rigvedic gods, with Rigvedic names, in the right forms. Mitanni horse-training texts from the same region use Indo-Aryan number words. So a language extremely close to Rigvedic Sanskrit existed, far to the west of Punjab, by roughly 1400 BCE.
What the hymns contain. They know horses, spoked-wheel chariots, copper and bronze. They do not know iron. Iron becomes common in north India after about 1200 BCE. That gives a ceiling.
What the hymns do not contain. No cities. No writing. No mention of anything that reads like a functioning Harappan town. That points away from the mature Harappan period, which ended about 1900 BCE.
Put those together and most specialists land on roughly 1500 to 1200 BCE for the bulk of the collection, with the oldest material possibly older and the last book clearly later. Some serious scholars push the start back towards 1700 or 1800 BCE. A smaller group argues for dates a thousand years earlier or more, on grounds we will meet in Chapter 7.
I will use "roughly 1500 to 1200 BCE" through this part, and I want you to hold it loosely. It is a reasoned estimate built from four indirect arguments, not a reading off a stone.
1.5 — What the layers tell us
One more thing, and it turns out to matter enormously.
The Rigveda is not all from one moment. Scholars have been able to sort it into rough layers, and the sorting is agreed across people who agree on almost nothing else.
Books 2 to 7 are the oldest core. They are called the family books, because each one is credited to a single priestly family and each is arranged by the same internal rules. Book 9 is a specialist collection about the ritual drink and sits slightly apart. Book 8 is mixed. Books 1 and 10 are the latest — their language is later, their arrangement is looser, and Book 10 in particular is where several famous hymns sit that sound nothing like the rest.
Why does this matter? Because if a hymn appears only in Book 10, it is late. And several of the things people most want to argue about — the four-fold division of society, the elaborate creation stories — appear only in Book 10.
Keep that in your pocket. We open it in Chapter 9.
The Rigveda is 1,028 hymns in ten books, a little over ten thousand verses, in an old form of Sanskrit. Every hymn is addressed to a god and asks for something.
It was not written down for centuries. It was memorised, using a system of overlapping recitation orders that acted as an error-check. That system is why the text we have is very close to the text they had — from the point the system existed.
It is liturgy, not history. It comes from paid priests on the winning side, speaking to gods, in front of patrons. Everything historical in it is accidental, and the accidental bits are the safest bits.
It dates to roughly 1500–1200 BCE, worked out from four indirect arguments and not from any date in the text.
The oldest thing we have from Punjab is not a record of what happened. It is a record of what a few men wanted, and history has to be scraped off the edges of it.
2The Land of Seven Rivers
The hymns keep naming rivers. Those names are the closest thing we have to a map — and the closest thing we have to proof that this book belongs to Punjab.
2.1 — The river hymn
In the tenth book there is a hymn that does nothing but praise rivers. It calls them by name, in order, and asks them to listen.
It is a strange little document. Most hymns want cattle or victory. This one just wants the rivers to know they are appreciated. And because the composer was showing off his geography rather than making a request, it is one of the most useful things in the entire collection.
The order runs from east to west. It starts with the Ganga and the Yamuna, moves through the Sarasvati and the Sutlej and the Ravi, and carries on west across the Indus and out the other side into rivers that are now in Afghanistan.
Elsewhere the hymns use a standing phrase for the region: sapta sindhavah, the seven rivers. That phrase is the composers' name for where they were.
Sindhu: means both "river" in general and the Indus in particular. The Persians turned the s into an h and got Hindu. The Greeks dropped the h and got Indos. Every one of the words Indus, India, Hindu and Hindi is a foreigner's pronunciation of the Sanskrit word for that river.
Sapta Sindhavah: "the seven rivers". The same phrase turns up in the Iranian Avesta as Hapta Hendu — same words, Iranian sound rules — where it is listed as one of the lands. Two closely related peoples, both using the same name for this place.
2.2 — Putting the names on the ground
The names in the hymns are not the names on your phone's map. They were replaced, twice — once by later Sanskrit, once by Persian. But the chain of replacement is traceable, and most of the identifications are secure.
| In the Rigveda | Today | How sure |
|---|---|---|
| Sindhu | Indus | Certain |
| Vitasta | Jhelum | Certain |
| Asikni | Chenab | Certain |
| Parushni | Ravi | Very strong |
| Vipas | Beas | Certain — the modern name is the old one worn down |
| Shutudri | Sutlej | Certain |
| Sarasvati | Disputed — see Part 2, chapter 8 | Contested |
| Kubha | Kabul river, Afghanistan | Strong |
| Suvastu | Swat river, Pakistan | Strong |
| Krumu | Kurram river | Strong |
| Gomati | Gomal river | Reasonable |
Read the table as a shape rather than a list. The world of the Rigveda has its centre in Punjab, reaches west across the Indus into the valleys of eastern Afghanistan, touches the Ganga and Yamuna at its far eastern edge, and does not go south to the sea in any way that is clear.
That is a real result, and it is the one thing in this part that almost nobody disputes. The Rigveda was composed here. In the land Part 1 described — the alluvial plain between the mountains and the desert, watered by five rivers, with the passes at its back.
Note also how thin the eastern end is. The Ganga is named twice in the whole collection. The Yamuna a handful of times. Compare that with the Indus and the Sarasvati, which get whole hymns. The composers were westerners. The Ganga was a rumour at the edge of the world they knew — and in Chapter 9 we watch that change completely.
2.3 — What is in the world of the hymns
Now empty out the pockets of the text and see what falls out.
Animals. Cattle everywhere — cattle are wealth, cattle are the unit of value, cattle are what raids are for. Horses constantly. Sheep and goats. Dogs. Lions. The elephant appears, but the composers had no word ready for it: they call it "the beast with a hand". That is somebody describing a new animal to people who have not seen one.
Plants and food. Barley is the grain. Milk, curds, clarified butter. The ritual drink pressed from a plant nobody can now identify with certainty. Honey.
Metal. One word, ayas, meaning copper or bronze. Used for vessels, blades, armour.
Life. Herding, moving, raiding, racing chariots, gambling with dice, feasting, fighting over water and pasture, and giving very large gifts to priests in public.
Try to picture the difference between the Rigvedic world and the Harappan world of Part 2. It is not a difference of one step. It is a change of category.
A Harappan city was fired brick, laid out on a grid, with covered sewers, standardised weights and a script, holding perhaps thirty or forty thousand people in one place.
The Rigvedic world, as far as the hymns show it, is mobile cattle-herding people organised around chiefs and priests. Wealth is on the hoof. Politics is a chief and his relatives. The biggest building mentioned is a stronghold, and the composers are usually on the outside of it, asking a god to smash it.
If Harappa was a city, the Rigveda is a camp — a well-organised, wealthy, confident camp, with an extraordinary poetic tradition and no interest whatsoever in drains.
2.4 — What is missing
Now the harder list. Some of these absences are famous, and every one of them gets used as ammunition by somebody.
No iron. The single word ayas covers copper and bronze. When iron does arrive in Indian texts it gets a qualifier — "dark metal" — and that qualifier appears only in later works. This is the firmest dating handle we have.
No rice. Barley, yes. Rice does not appear in the Rigveda and turns up in the later Vedic texts, once the centre of gravity moves east into wetter country.
No cotton. The Harappans grew and wove cotton — it is one of their signature achievements. The Rigveda knows wool.
No writing. There is no word for writing, no scribe, no letter, no document. For a culture that had built the memorisation machinery of Chapter 1, this is not surprising. But it is worth noticing that they were sitting on top of the ruins of a society that had a script, and they say nothing about it.
No cities. There are strongholds — the word is pur — and Indra's favourite title is "fort-breaker". But what a pur physically was is genuinely unclear. Some are described as made of stone, some of metal, some as having a hundred walls, some as belonging to demons. It reads more like the language of a raider talking about defended places than a description of urban planning.
No tiger. Small thing, but a real one. The Rigveda knows lions and not tigers. Tigers appear in the later texts. Lions lived in the west and north-west; tigers in the wetter forests to the east. It is one more small arrow pointing at where these people were.
That what a text does not mention did not exist.
This is called the argument from silence, and it is the most heavily used and least reliable tool in this entire part. Both sides swing it, constantly, and usually only notice the problem when the other side swings it back.
Look at how unstable it is. The Rigveda does not mention iron — and we treat that as solid evidence for a date before iron. The Rigveda does not mention cities — and we treat that as evidence these were not city people. But the Rigveda also does not mention cotton, which certainly existed in this region, and does not mention the sea in any clear way, and does not mention several things we know were there.
So which silences count?
A silence is meaningful only when you can show the text would have had reason to speak. Iron is a good case: these hymns care about weapons and metal constantly, so a metal they had and did not name is odd. Drains are a bad case: nothing in a hymn to Indra would ever require a drain.
That test is the whole of it. Before you accept any argument in this part built on something not being mentioned — and you will meet several, including the biggest one, about a horse — ask: would this text have said so, if it were true? If you cannot answer that, the silence tells you nothing.
And notice which way you want each particular silence to cut. That is the tell.
2.5 — Beyond the mountains
One last piece of geography, and it is the piece that opens the door to the rest of this part.
The Rigvedic world does not stop at the Indus. Four of the rivers in the river hymn are west of it, in the valleys of what is now eastern Afghanistan and the Pakistani north-west. The composers knew that country. Some hymns seem to be looking at Punjab from out there.
And the Avesta — the Iranian sacred book, in a language so close to Vedic Sanskrit that it is barely a different language — knows the same region and calls it by the same name.
The Vedic and Iranian traditions share gods, share ritual, share a sacred drink with the same name, share whole poetic phrases. They also share a curious inversion: one tradition's word for "god" is the other's word for "demon", and the other's word for "god" is this one's word for "demon". That is what a family quarrel leaves behind in a language.
Two branches of one thing, then, sitting on either side of the mountains. That fact will not let us alone for the rest of this part.
The Rigveda names its rivers, and most of the names can be matched to rivers on a modern map. Sindhu is the Indus, Vitasta the Jhelum, Asikni the Chenab, Parushni the Ravi, Vipas the Beas, Shutudri the Sutlej.
That fixes the world of the hymns to Punjab and the country immediately west of it. The Ganga is a rumour at the far edge. This book belongs to this land, and that much is not in dispute.
The world it shows is herders, chiefs and priests — cattle as wealth, horses, chariots, barley, bronze. No iron, no rice, no cotton, no writing, no cities.
Those absences are used as evidence by everybody. Be careful with them. A silence only counts if the text had a reason to speak.
The Rigveda is a Punjabi book. What is argued about is not where it was made, but who made it and how they got here.
3What "Aryan" Actually Means
One Sanskrit word, used a few dozen times in a book of hymns, ended up on the banners of the worst regime in modern history. This chapter follows it the whole way.
3.1 — The word in the text
The word is arya. It appears in the Rigveda about three dozen times. That is not many, in ten thousand verses.
It is always a word the composers use about themselves. Never about someone else. It marks the in-group.
What it does not do, anywhere in the text, is describe a body. There is no verse that defines an arya by the shape of a skull, or the colour of hair, or descent from a particular ancestor. Nobody in the Rigveda measures anybody.
What the word attaches to instead is behaviour and speech. The arya are the people who make the right offerings, keep their word, speak the language properly and honour the gods by name. In later Sanskrit the word simply softens into "noble" or "honourable" — the polite way to address a respected man, the way English once used "gentleman".
Arya: the self-description of the people who composed the Rigveda. Marks the in-group. Later softens to "noble, respectable".
Dasa and dasyu: the words for the enemies and outsiders. Over the following centuries dasa drifts in meaning until it simply means "slave" or "servant" — which is how it survives in modern Indian languages.
Anarya: not-arya. The opposite term. Also rare.
Indo-Aryan: a modern language label, invented by scholars. It means the branch of the language family that includes Sanskrit, Punjabi, Hindi, Bengali, Marathi, Sindhi and about two hundred others. When linguists write "Indo-Aryan" they are talking about grammar, not about people, and certainly not about race.
3.2 — The dasa and the dasyu
If arya is "us", then dasa and dasyu are "them", and this is where the reading gets difficult.
The hymns ask Indra to break the strongholds of the dasa, to scatter them, to take their cattle. It is hostile, repetitive and violent, in the ordinary way of war poetry everywhere.
But four things complicate the obvious reading, and all four are in the text.
First, some dasas are rich, and some pay priests. One hymn thanks two patrons by name for a generous gift of cattle and camels, and calls them dasa. A priest is taking money from the other side and putting the receipt in the hymn. That is very hard to square with an absolute racial line.
Second, aryas fight aryas. Several hymns ask the gods for help against enemies of both kinds in the same breath — deliver us from the dasa foe and from the arya foe. The world is not divided in two. It is divided into many quarrelling groups, some of whom are "us" and are still trying to kill us.
Third, the names go the wrong way. One of the great arya kings in the hymns is called Divodasa — which contains the word dasa. Sanskrit-speaking families were happily naming their sons with the enemy word.
Fourth, the word has a cousin in Iran. In the Iranian material, the closely related word daha is a tribal name — a group of people, not a category of subhuman. It looks very much as though dasa started as somebody's actual name, became a general word for enemy, and only much later slid down into "servant".
3.3 — The verses about colour
Two passages do most of the work in the racial reading, and they deserve to be looked at directly rather than summarised by either side.
The first uses the word varna, which literally means colour or covering, and sets an "arya varna" against a "dasa varna". The second describes enemies as dark and as anasa.
The trouble is that both are ambiguous in the original, not because modern scholars are being slippery, but because the Sanskrit genuinely permits more than one reading.
The word anasa. Split it one way and it is a-nasa, "without a nose" — flat-nosed, which is how a fourteenth-century commentator took it and how nineteenth-century race scholars gleefully repeated it. Split it the other way and it is an-asa, "without a mouth" — meaning without proper speech, people who cannot say the words correctly.
Sanskrit is written without spaces between words in exactly the place that matters. Both splits are grammatically legal. The word occurs once in the entire Rigveda. There is no second occurrence to check it against.
The word varna. It does mean colour. It also means outward appearance, covering, class, kind, and — later — the letters of the alphabet. "Our colour and their colour" can be skin. It can equally be the sort of contrast we make when we say two groups are "of a different stripe".
What this proves: that the racial reading is possible. It is not invented out of nothing. Someone reading these verses in 1860 and seeing colour prejudice was reading real words on a real page.
What it does not prove: that the racial reading is required. Every one of these passages has a straightforward non-racial reading, and the surrounding evidence — dasa patrons, arya enemies, dasa names on arya kings — fits the non-racial reading considerably better.
Notice what nobody can do here. Nobody can settle it. One ambiguous word, one flexible word, and a hostile poem. That is the entire evidence base for the most consequential misreading in modern history.
So the text will not decide it. What people have actually done, for two centuries, is decide first and then read. Here are the three readings, laid out properly.
Three readings are on the table. All three have serious people behind them.
A group defined by descent, who thought of themselves as a distinct stock and distinguished themselves from others by origin. In favour: the word is inherited, not invented, and its Iranian twin is used by Darius the Great of Persia in a royal inscription to state his lineage — "an Aryan, of Aryan descent". People do use ancestry words about ancestry. The name of the country Iran comes from this word and means, straightforwardly, "of the Aryans".
Those who speak our tongue and sacrifice our way. In favour: the enemies are repeatedly marked out by speech and by not sacrificing, not by looks; the word attaches to correct performance; and the Vedic-Iranian pair share the term precisely because they shared a religion and a language before they split. This is where most specialists now sit.
Simply "noble" — a moral term about conduct, which is unambiguously what it means a few centuries later, where arya becomes a term of respect used across Indian traditions, including in Buddhism, which is not much interested in anyone's ancestry.
What would settle it: nothing available. We would need a second independent source from the same period and place. There is none, and there is unlikely ever to be one.
Why it matters: because the entire nineteenth-century racial theory was built by taking reading one, hardening "descent" into "biological race", and then attaching that race to physical features the text never mentions. Each of those steps was a choice, and each was made by people who wanted a particular answer.
3.4 — How a word became a race
Here is the sequence. It took about a century and it was largely accidental, which does not make it less catastrophic.
Step one. In the late eighteenth century European scholars work out that Sanskrit is related to Greek, Latin, Persian and the languages of Europe. That is Chapter 4, and it is a genuine and correct discovery.
Step two. They need a name for the family. They reach into the oldest texts of the family and pull out the word the speakers used for themselves. Arya. It becomes "Aryan", and for a while it is simply what the family of languages is called.
Step three. The slide. A label for a group of languages starts being used for a group of people. This is not stupid — languages do belong to people — but it is a step that requires proof and got none.
Step four. The race theorists arrive. In the 1850s a French aristocrat writes a long book arguing that human history is the story of racial quality and decline, and that the finest stock was the Aryan. He is not a linguist and does not much care what the Rigveda says. In 1899 an Englishman living in Germany writes an even more successful book making the Aryan a specifically Germanic possession. Neither book contains evidence in any sense a historian would recognise.
Step five. The theory is picked up by German nationalism, made law, and used to justify the murder of millions. The swastika they used was an Indian symbol. The word they used was a Sanskrit word. Neither had anything to do with the meaning either originally carried.
It is worth being exact about what the harm was here. The linguistics was right. Sanskrit really is related to Greek. What went wrong was a chain of unproven leaps: from a language family, to a people, to a biological race, to a ranking of races. Every link in that chain was assumed rather than shown.
3.5 — Müller's protest, and India's own use of the word
The scholar most associated with popularising "Aryan" in English was Friedrich Max Müller, a German philologist at Oxford who produced the first full printed edition of the Rigveda. He saw what was being done with his terminology and he objected — publicly, repeatedly and furiously.
Strip the long words and he is saying: you cannot measure a language with callipers. A skull is not a grammar. He said it in 1888. It made no difference at all. Once a word is useful to a political movement, the scholar who coined it has no further say.
India used the word too, and in more than one direction, which is a part of the story that usually gets left out.
A reform movement founded in 1875 took the name Arya Samaj — the Aryan society — and used it to argue for a return to Vedic purity and against caste as it then stood. A nationalist leader wrote a book arguing the Aryan homeland had been at the North Pole. And in Maharashtra, a low-caste reformer took the racial theory and turned it upside down: if the Aryans came from outside and conquered, he wrote, then the upper castes are the invaders and the oppressed are the original people of this land. That inversion became a real and lasting political force in western and southern India.
Then in 1946 B. R. Ambedkar — who had every political reason to keep the invasion theory, since it made his own community the rightful natives — read the texts and rejected it anyway. He argued the Aryan race theory had no basis in the Vedas at all, that the Rigveda does not distinguish arya and dasa by race, and that the whole apparatus was European fantasy imported into India.
That is worth sitting with. The man with the strongest political motive to accept the racial reading looked at the evidence and threw it out.
Arya appears about three dozen times in the Rigveda. It is always the composers' word for themselves. It is never defined by body, blood or skull anywhere in the text.
The dasa and dasyu are the enemies — but some dasas are rich, some pay priests, arya kings carry dasa in their names, and aryas fight aryas. It is not a clean line between two peoples.
The two famous "colour" passages are genuinely ambiguous in Sanskrit. The racial reading is possible. It is not required, and the rest of the evidence fits it badly.
"Aryan" became a race through five steps, only the first of which was sound: a real language family, given a name from the oldest text, then treated as a people, then as a race, then as a ranking. Max Müller, who did most to spread the word, spent his last years trying to stop exactly this.
"Aryan" is not a race and never was. It is a word for a language family, borrowed from a word a group of poets used to mean "people like us who do things properly".
4The Family of Languages
The strongest evidence in this whole argument is not a bone or a brick. It is a pattern inside words — and it is far harder to wriggle out of than most people realise.
4.1 — Jones in Calcutta
In February 1786, a British judge in Calcutta gave a lecture to a small learned society he had helped found. His name was William Jones. He had learned Sanskrit in order to read Hindu law in the original, because he did not trust the translations his court was using.
In the middle of that lecture he said something that changed the study of human history.
Sanskrit, he said, is more perfect than Greek, more copious than Latin, and yet bears to both a resemblance — in verb roots and in the shape of grammar — stronger than could possibly have been produced by accident. So strong, he said, that no scholar could examine all three without believing they had sprung from some common source, which perhaps no longer exists.
He was not quite first. A Florentine merchant in Goa had noticed the number words two centuries earlier. A French Jesuit had made the case in a letter twenty years before. But Jones said it in the right room, in the right decade, and it stuck.
And notice the last clause, because it is the clever part. A common source, which perhaps no longer exists. He is proposing a language nobody had ever recorded, that had died before writing, and that could only be reached by working backwards from its children.
4.2 — Father, pitar, pater
Look at the evidence yourself. You do not need Sanskrit to see it.
| English | Sanskrit | Greek | Latin | Persian |
|---|---|---|---|---|
| father | pitar | pater | pater | pedar |
| mother | matar | meter | mater | madar |
| brother | bhratar | phrater | frater | baradar |
| two | dva | duo | duo | do |
| three | trayas | treis | tres | se |
| seven | sapta | hepta | septem | haft |
| new | nava | neos | novus | nau |
| name | naman | onoma | nomen | nam |
Punjabi is in this table too, of course. Pita, mata, bhra, do, tinn, satt, nawaan, naam. Punjabi is a member of this family in exactly the way Sanskrit is.
Now, an obvious objection: languages borrow words all the time. Punjabi has taken thousands of words from Persian and Arabic and English. Could this be borrowing?
No, and for a specific reason. Languages borrow words for new things — technology, trade goods, food, administration, religion. They do not borrow the words for mother, father, two, three, seven, or the endings that mark a verb's tense. Those are the last things to change and the first things a child learns. When the deep plumbing matches, it is not borrowing. It is descent.
4.3 — Why the pattern cannot be luck
The table above is impressive but it is not the proof. The proof is that the differences are regular.
Look at the p in pitar and pater, and the f in father. That is not a one-off. It happens every time.
Where Latin has p, English has f:
pater → father · pes / pedis → foot · piscis → fish · plenus → full · pellis → fell (skin)
Where Latin has t, English has th:
tres → three · tenuis → thin · tu → thou
Where Latin has k (written c), English has h:
cornu → horn · canis → hound · centum → hundred · cord- → heart
This is not a handful of lucky matches. It is a machine. Every single one of a whole class of sounds shifted the same way, in the same branch, at the same period — and it happened to the entire vocabulary at once, including words nobody was thinking about.
Coincidence cannot do this. Borrowing cannot do this either, because borrowing is haphazard and this is total. A sound law is the closest thing the study of languages has to a law of physics.
The particular set of shifts that separates the Germanic languages from the rest was worked out by Jacob Grimm — the same Grimm who collected fairy tales. Once you have a rule like that, you can run it backwards. You can take an English word, undo the law, and predict what the Latin or Sanskrit form ought to be. Then go and look. It is there.
That predictive power is why linguists trust this method more than almost anything else in the historical toolkit.
4.4 — Building a language nobody spoke
If you can run the rules backwards for one language, you can run them backwards for all of them at once, and land on the forms they all came from.
The result is called Proto-Indo-European. Nobody ever wrote it down. Nobody has heard it for at least five thousand years. But several thousand of its words and a great deal of its grammar have been reconstructed, and the reconstruction is stable enough that when a genuinely new Indo-European language turns up — as happened when Hittite tablets were deciphered in the twentieth century — the predictions largely hold.
Cognate: two words in different languages that are descended from the same ancestor word. Father and pitar are cognates. Not to be confused with a loanword.
Proto-language: a language reconstructed by working backwards from its descendants. It is a hypothesis with a lot of evidence behind it, not a document.
Proto-Indo-European (usually shortened to PIE): the reconstructed ancestor of a family that now includes Punjabi, Hindi, Bengali, Sanskrit, Persian, Kurdish, Greek, Latin and all its children, the Slavic languages, the Celtic languages, German, English, Armenian and more. About half the world speaks a descendant of it.
Substrate: the traces an older, displaced language leaves behind in the language that replaced it — usually in place names, in words for local plants and animals, and sometimes in pronunciation.
4.5 — Words as a time machine
Here is where the reconstruction stops being a curiosity and starts being evidence about the world.
If a word can be reconstructed for the parent language — meaning the daughter languages all have matching forms that obey the sound laws — then the speakers of the parent language had the thing. You cannot inherit a word for something you have never met.
So what did they have?
They had wheels. The word survives as Sanskrit cakra, Greek kuklos, English wheel. They had an axle and a yoke and a word for going in a vehicle. They had horses — Sanskrit ashva, Latin equus, Greek hippos. They had wool, cattle, sheep, pigs, honey, and a fermented honey drink. They had snow and a word for cold. They had wolves and bears and birch trees.
And what do they not have? There is no reconstructable word for tiger, elephant, monkey, bamboo, rice, or lotus. The Indo-European languages of India have words for all of these, but each language made or borrowed its own; they do not match up across the family.
The wheel matters most of all. Wheeled vehicles do not appear anywhere in the archaeological record before roughly 3500 BCE. If the parent language already had a full wheel vocabulary that all its children inherited, then the parent language cannot have broken up much before then. That single argument puts a floor under the whole chronology, and it is the reason most linguists will not accept dates for PIE much earlier than about 4500 BCE.
4.6 — Where was the homeland?
Everyone in this argument agrees the family is real. The question is where the parent language was spoken and how its children spread. There are three serious answers.
The grasslands north of the Black and Caspian Seas, roughly 4500–3000 BCE. In favour: it fits the reconstructed vocabulary almost item for item — horses, wheels, wool, cattle, cold, birch, no tigers. Horses were first domesticated in that region. The archaeology shows a distinctive burial-mound culture spreading out of it in several directions at about the right time. And, since 2015, ancient DNA shows large-scale movement of people out of exactly that region into Europe — which is not proof for South Asia, but it does show the mechanism was real. This is where most specialists sit.
Modern Turkey and the highlands south of it, considerably earlier — spreading not by conquest but with farming itself, as farmers slowly outbred foragers. In favour: the earliest-splitting branch of the whole family, Hittite, was spoken in Anatolia; farming is a far better engine for spreading a language than raiding is; and several recent attempts to date the family tree by statistical methods borrowed from biology have produced dates far older than the Steppe model allows, pointing south of the Caucasus. A 2023 study of this kind put the split at roughly eight thousand years ago and proposed a hybrid: an older southern origin, with a later Steppe phase carrying the branch that reached both Europe and India.
That the homeland was in north-west India, and the family spread outward from here — so no one came in, and Sanskrit's ancestors were always here. In favour: Sanskrit preserves some strikingly archaic features; the Rigveda contains no memory whatever of a migration from outside, which is odd for a people who supposedly arrived recently and remembered everything else; the whole homeland question is an inference from word lists, with no physical object anywhere labelled "Proto-Indo-European"; and if the Sarasvati of the hymns is a great flowing river that dried before 1900 BCE, the Rigveda has to be older than any proposed migration.
What would settle it: nothing in linguistics alone. What has actually moved the field is ancient DNA — Chapter 6 — and what would settle it further is more of the same, from the right places and the right centuries. Specifically: genomes from Punjab and the north-west between 2500 and 1000 BCE. We have almost none.
Why people care so much: because the homeland has been treated, by everybody, as a deed of ownership. It is not one. See Chapter 7.
4.7 — What this does not prove
Now the part that gets skipped, and it is the most important paragraph in the chapter.
Everything above is about languages. Languages are not people. They are not genes. They are not culture. They travel by being learned, and human beings learn languages from whoever is useful, whoever is powerful, and whoever is nearby.
That a language moving means a people moving.
Every version of the Aryan argument leans on this, and it is false as a general rule. Look at the cases we can actually check, because they happened in recorded history.
Egypt speaks Arabic. Egyptians today are overwhelmingly descended from ancient Egyptians. The Arab conquest brought in a small ruling group. The language changed completely; the population did not.
Anatolia speaks Turkish. Modern Turks are largely descended from the peoples who were there before, with a modest genetic contribution from Central Asia. A language crossed a continent on the backs of a minority.
Ireland speaks English. Nobody replaced the Irish.
France, Spain, Portugal and Romania speak Latin's children. Rome did not resettle western Europe with Italians. It ran it.
In every one of these, a language spread through prestige, administration, trade and religion, carried by a group far too small to show up as a population replacement.
So what does this do to the argument? It cuts hard in both directions, which is why almost nobody states it plainly.
It undercuts the invasion story: you do not need an army, a conquest or a massacre to explain Sanskrit in Punjab. A small group with prestige is enough.
It equally undercuts the indigenous story: showing that the population of Punjab is mostly descended from the people who were already here proves nothing about where the language came from. Those are two different questions and the answer to one does not give you the answer to the other.
Hold on to this. It is the single most useful idea in this part of the book.
Keep the language question and the ancestry question in separate hands, then. Chapter 6 is about the second one, and it is a different kind of evidence entirely.
Sanskrit, Punjabi, Persian, Greek, Latin, English and about four hundred other languages are descended from one ancestor language that nobody wrote down. This is not a guess. It rests on regular sound correspondences that work like a machine.
The ancestor language can be partly reconstructed. Its speakers had wheels, horses, wool, snow and birch trees. They had no word for tiger, elephant or rice. The wheel puts a floor under the date at roughly 4000–3500 BCE.
Most specialists place the homeland on the Steppe north of the Black Sea. A minority argue for Anatolia or the lands south of the Caucasus. A much smaller group argues for India. The reasons the third position has few takers among linguists are technical, not political.
And the thing everyone forgets: languages move without people moving. Egypt speaks Arabic. Turkey speaks Turkish. Ireland speaks English. None of those needed a population replacement.
The language family is the strongest evidence here. It is also evidence about words, not about blood — and those two questions have to be kept apart.
5The Horse and the Wheel
Part 2 promised you that a missing animal would become the front line of the biggest argument in Indian history. Here it is.
5.1 — A book soaked in horses
You cannot read the Rigveda for ten minutes without meeting a horse.
Horses pull the chariots of the gods. Horses are the top gift a patron can give a priest, worth more than cattle. Two of the gods are themselves called the Horsemen. There are hymns to individual named horses, treated as divine. There are chariot races, and the poetry about them is some of the most excited in the collection. There is a whole ritual built around the sacrifice of a single stallion, described in loving detail down to the parts of its body.
Now put that beside Part 2.
The Harappans left behind more than four thousand inscribed seals. On them are bulls, buffaloes, elephants, rhinos, tigers, crocodiles, goats and an animal with one horn that nobody can identify. There is not one horse. Not a single agreed horse image from the entire mature Harappan civilisation.
The bones tell the same story. Harappan sites yield cattle bones by the tonne. Horse bones are, at best, a handful of disputed fragments.
Two societies in the same land, a few centuries apart. One is obsessed with an animal. The other appears not to have known it existed.
5.2 — Why a horse bone is hard to name
Before we get to what that means, we have to deal with a technical problem, because it is where most of the actual fighting happens.
India has its own wild member of the horse family: the khur, or Indian wild ass, which still lives in the salt flats of Gujarat. It is not a horse. It was never domesticated. It has been in India for a very long time.
Here is the difficulty. The leg bones, foot bones and vertebrae of a khur are very close to those of a true horse. Close enough that even an experienced specialist, handed a broken shin fragment, often cannot say which animal it came from.
What can be told apart reliably: the teeth and parts of the skull. Horse cheek teeth have a distinctive enamel pattern.
What usually turns up on a dig: broken bits of leg and foot, because those are what survive being cooked, chewed, trampled and buried for four thousand years.
Why this matters: almost every claimed horse find at a Harappan site is a fragment of exactly the ambiguous kind. This is not a case of one side being dishonest. It is a case of the evidence being physically incapable of answering the question on its own.
Modern DNA analysis of bone can settle individual cases, and where it has been applied the results have generally not supported the horse claims. But it has been applied to very few of the disputed pieces.
The best-known case is a set of bones from a Harappan site in Gujarat. In the early 1990s a Hungarian specialist in ancient animal bones examined them and concluded, with confidence, that some were true horse. He was a serious scholar and he was not making a political point. A few years later two American specialists re-examined the same material and concluded they were khur.
Neither side ever conceded. The first scholar died before the exchange finished. The bones are still there and the disagreement is still open.
5.3 — The spoked wheel
The horse argument has a partner that gets far less attention and is, in my view, stronger.
The Rigveda's chariot is not a cart. A cart is heavy, has solid wooden wheels, and is pulled by oxen at walking pace. The Rigvedic ratha is a light two-horse vehicle built for speed, and its wheels have spokes.
A spoked wheel is a serious piece of engineering. It has to be light enough to be worth building and strong enough not to fly apart under a galloping horse. Getting there took a long time and specific skills — steam-bending wood, precise joinery, a proper axle.
The difference between a solid wheel and a spoked wheel is roughly the difference between a bullock cart and a motorcycle.
A solid disc of wood is easy to make and enormously heavy. Fine for hauling grain at four kilometres an hour. Useless for anything else.
A spoked wheel might weigh a fifth as much. That is what lets two horses pull a vehicle at thirty or forty kilometres an hour with a man standing in it shooting a bow.
It is a weapons technology, and when it appears it changes warfare across half a continent within a few hundred years — the way gunpowder did later.
The earliest actual spoked-wheel chariots ever found are in burials in the southern Urals, on the edge of the Steppe, dating to around 2000 BCE. Whole vehicles, buried with horses and drivers, wheels set into slots cut in the grave floor. The impressions of the spokes are still in the soil.
The Harappans had wheels. Their toy carts show solid wheels. Nobody has found a spoked wheel, a chariot fitting, a horse bit or a chariot burial anywhere in the Harappan world.
So the chariot has a known birthplace, a known date, and an apparently clean absence in the civilisation that occupied Punjab before the Rigveda. That is a harder set of facts than the bones.
This is the argument that gets the most heat and, I think, deserves the least.
The domestic horse's origin has been traced by ancient DNA to the grasslands of the western Steppe, and the lineage that became every domestic horse in the world today expanded out of there from about 2000 BCE. The chariot appears in the same place at the same time. The Rigveda is full of both. The Harappans, occupying the same land earlier, show neither. The simplest reading is that horse and chariot arrived, from the north-west, with the people who brought the hymns.
Three answers, all reasonable. Horses do badly in hot, wet climates and have always been hard to breed in the plains of India — so even a society that used them heavily would leave few bones, because a valuable imported animal is not everyday meat. Several claimed horse finds at Harappan sites exist and have not been disproved, only disputed. And the absence of horses on Harappan seals is worth exactly as much as the absence of cows on Harappan seals, which is to say we do not know what the seal animals were for.
Both sides are running an argument from silence, and Chapter 2 showed how unstable that is. Even at its strongest, the horse tells you when a technology arrived. It does not tell you how many people came with it, whether they conquered anybody, or who anyone's ancestors were. Trade moves technology all the time without moving populations.
What would settle it: DNA testing of every disputed bone fragment from every Harappan site that has claimed one. This is technically possible now and comparatively cheap. It has not been done systematically, which is itself worth noticing.
Why it is fought over so hard: because both sides have decided the horse is a proxy for the whole argument, and it is not. It is one animal. It carries far more weight in this debate than it can bear.
5.4 — The seal that was doctored
Part 2 told this story briefly. It belongs here in full, because it is the clearest warning in the whole subject.
In 2000, a book was published claiming to have deciphered the Indus script. Among its illustrations was a seal impression showing what the authors said was a horse — which, if real, would have punched a hole in one of the main arguments against an indigenous origin for the Vedic people.
Two scholars checked it. The image turned out to be a computer-altered version of a known, published, broken seal. The original showed the usual one-horned animal. The break had been tidied, the horn removed, and the result presented as a horse.
It collapsed within weeks and did real damage to the credibility of the position it was meant to support.
I want to be careful about what lesson you take. The lesson is not "that side cheats". Every side of every historical argument has produced bad work, and the history of the invasion theory in Chapter 7 contains errors just as bad, made by people just as respectable, in the opposite direction.
The lesson is about wanting. The people who circulated that image were not, most of them, frauds. They were people who wanted something to be true, saw an image that showed it, and did not check — because checking is what you do to claims you dislike.
Which brings us back to the request I made at the start of this part. Notice what you are hoping for as you read. Everybody has a hope in this argument. The danger is not having one. The danger is not knowing you have one.
The Rigveda is full of horses — gods, gifts, races, sacrifices. The Harappan record has no agreed horse image on four thousand seals and only a handful of disputed bone fragments.
Horse bones are genuinely hard to tell apart from the bones of the Indian wild ass, unless you have teeth. Almost every disputed find is exactly the ambiguous kind. This is a limit of the evidence, not a failure of honesty.
The spoked-wheel chariot is the stronger case. It has a known birthplace on the edge of the Steppe around 2000 BCE, and no trace at all in the Harappan world.
The doctored seal of 2000 is a warning that cuts in every direction. People do not usually check things they want to be true.
The horse can tell you roughly when a technology arrived. It cannot tell you who anybody's ancestors were, and it is asked to do far more work than it can carry.
6What the Genes Say
In the last ten years a new kind of evidence arrived, and it has moved this argument more than anything since Jones. It is also the evidence most routinely misreported, in both directions.
6.1 — Reading a four-thousand-year-old bone
When a person dies, their DNA starts breaking into shorter and shorter pieces. Heat speeds it up. Damp speeds it up. After a few thousand years in a hot country there is usually nothing usable left.
Two things changed that. First, machines got good enough to read DNA in fragments a few dozen letters long and stitch the fragments back together by computer. Second, somebody worked out that one particular bone — the dense lump around the inner ear — protects DNA far better than the rest of the skeleton.
The result is a field that barely existed twenty years ago and now has thousands of ancient genomes in it. But South Asia is the hardest place on Earth to do it. The climate is exactly wrong. Attempts fail far more often than they succeed.
Ancient DNA: genetic material recovered from human remains rather than from living people. It tells you about a specific person on a specific date, which is why it is so much more powerful than working backwards from the living.
Ancestry component: when geneticists say a population "has 20% X ancestry", they mean that a fifth of that population's genome looks most like a reference group they have labelled X. It is a statistical statement about resemblance, not a passport.
Steppe ancestry: the genetic profile typical of Bronze Age herders of the grasslands north of the Black and Caspian Seas, and their eastward descendants in Central Asia.
Haplogroup: a branch on the family tree of either the Y chromosome (passed father to son) or mitochondrial DNA (passed mother to child). Because each is inherited down one line only, they let you track men and women separately.
6.2 — The two ancient ingredients
Start with what was here before any of this argument begins.
The population of South Asia, going far back, is built out of two very old ingredients.
The first is a deeply ancient lineage descended from the people who reached India tens of thousands of years ago, in the first spread of modern humans out of Africa. Their closest surviving relatives, distantly, are the indigenous people of the Andaman Islands. Everyone in South Asia carries a substantial share of this.
The second is a lineage related to the ancient populations of the Iranian plateau. Here is a detail that matters and is usually skipped: this lineage split off before farming was invented. So its presence does not mean farmers walked in from Iran. It means a related hunting population was in this region very early, long before any of the events in this book.
Mix those two in different proportions and you have the underlying population of the subcontinent. That mixture was in place, in various ratios, well before the Harappan cities.
6.3 — Rakhigarhi
Rakhigarhi in Haryana is one of the largest Harappan sites. In 2019, after years of failed attempts, a team published a genome from a woman buried there — the first and, so far, effectively the only genome we have from anyone inside the Harappan civilisation.
Her ancestry was the two ingredients above. Iranian-plateau-related and the deep indigenous lineage.
She had no Steppe ancestry at all.
That single result was reported around the world, by both sides, as proving the opposite things. One camp said: see, the Harappans were not Steppe people, so the Vedic people must have come later from outside. The other camp said: see, no foreign ancestry in the Harappans, so nobody came in.
Both readings run past what the data can support, and the reason is embarrassingly simple.
It is one woman. One individual, from one site, at the eastern edge of the Harappan zone, in one century.
Ask what a single genome from one person in Delhi in 1950 would tell you about India in 1950. It would tell you about that person. It would tell you a certain kind of person existed. It would tell you nothing about proportions, nothing about the rest of the country, and nothing about who arrived twenty years later.
The wider study that framed the Rakhigarhi result compared her with eleven individuals found at sites in Iran and Turkmenistan whose ancestry matched hers — people who appear to have been travellers or migrants from the Indus world, buried far from home. Those eleven, plus her, are the entire sample for the genetics of the Harappan civilisation.
Twelve people. For a civilisation of perhaps a million.
And here is the sentence that should be in every report of this and almost never is: we have no ancient DNA at all from Punjab itself in the relevant period. None from Harappa. None from the Ravi or the Beas or the Chenab. The land this entire book is about is a hole in the map.
6.4 — When Steppe ancestry arrives
Now the finding that actually carries weight, because it rests on hundreds of individuals rather than one.
A large study published in 2019 sequenced over five hundred ancient people from Central and South Asia across several thousand years. The pattern it found is this.
Before roughly 2000 BCE, in every sample that can be linked to the Indus world, there is no Steppe ancestry.
After roughly 1500 BCE, in samples from the Swat valley in northern Pakistan, Steppe ancestry is present. It goes on being present, and it is present in the living populations of South Asia today.
The particular Steppe profile that shows up is not the earliest Steppe population but their descendants who had moved east into Central Asia — the same broad group associated with the chariot burials of Chapter 5.
So: an ancestry component that is absent before about 2000 BCE and present after about 1500 BCE, in the north-west, arriving from the direction of Central Asia, at the same period as the horse and the chariot and the composition of the Rigveda.
That is a real result and it should be stated plainly rather than softened.
6.5 — Why it sits on the male line
There is a further detail, and it is the one that tells you most about what actually happened.
The Steppe contribution is much stronger on the Y chromosome — which passes from father to son — than in the genome overall. Mitochondrial DNA, which passes from mother to child, is overwhelmingly local and ancient in South Asia.
Translated out of genetics: the incoming group was heavily male. The women in the resulting population were, overwhelmingly, already here.
That is the genetic signature of a relatively small number of men moving in, marrying local women, and their sons doing better than average — which is what happens when a group has some kind of advantage, whether military, ritual, economic or just organisational.
It is emphatically not the signature of a population being replaced. In modern South Asians the Steppe share is a minority component — broadly in the range of a tenth to a fifth, higher in the north-west and in some communities, lower in the south, and never anywhere close to a majority.
Think of the ancestry of a Punjabi today as a jar with three things in it.
The largest share by far is the ancient population of this subcontinent — the deep indigenous lineage plus the early Iranian-plateau-related lineage. These are the people of Mehrgarh in Part 1 and the people of Harappa in Part 2. They are the bulk of you.
The smaller share, arriving around three and a half thousand years ago, is Steppe.
If you want a rough picture: for most people in Punjab, something like four parts in five is the ancient population of this land, and something like one part in five is the later arrival — and even that fifth is on the high side compared with most of India.
Whatever happened here, it was not a replacement. The overwhelming majority of your ancestry has been in this region since long before anybody composed a hymn.
All of which brings us to the fight over what any of this means.
The data above is not much disputed as data. What it means is disputed hard.
An ancestry component absent before 2000 BCE and present after 1500 BCE has to have arrived. It arrived from the direction the linguistics predicted, in the period the linguistics predicted, alongside the technology the linguistics predicted. Three independent lines of evidence — words, wheels, genomes — converging on the same window is about as good as ancient history gets.
The Harappan genetic record is twelve individuals, eleven of them found outside the Indus region. There is no ancient DNA from Punjab in the relevant centuries at all. The post-1500 BCE samples come from one mountain valley in the far north-west, which is exactly the sort of corridor you would expect outside influence in, and is not the plains. Dates of mixing calculated from living populations carry error bars of centuries. Change the reference populations and the numbers move. A pattern from thin sampling in the wrong places is a hypothesis, not a finding.
Grant everything. Some men came in around 1800–1500 BCE and their descendants did well. That is a demographic fact about a minority ancestry component. It does not tell you which language they spoke, whether they composed the Rigveda, whether they arrived peacefully or violently, or whether the culture that resulted was theirs or a fusion. Genes travel with bodies. Hymns travel with teachers.
What would settle it: ancient DNA from Punjab and Sindh, from between 2500 and 1000 BCE, in numbers. That is the missing piece, everybody knows it is the missing piece, and the difficulty is partly climate and partly that excavation permissions in the relevant countries are not simple.
What no version of this settles: who belongs here. See the box below.
And now the thing that is really being argued about, which is not in any of the papers.
That ancestry is identity.
This assumption is why people shout about the previous three pages. Everyone in the argument behaves as though a percentage in a genetics paper is a verdict on who they are.
It is not, and the reason is worth stating carefully.
Go back far enough and everybody's ancestry is a mixture, because populations have always mixed. Every European carries ancestry from at least three separate ancient groups. Every East Asian population is a compound. The people the Steppe herders themselves descended from were a mixture. There is no pure population anywhere on Earth and there never was. Purity is not a thing that got lost. It is a thing that was never there.
So when someone asks "are they us?", the honest answer is the one I gave at the start of this part: partly, along with several other groups, and this is true of nearly every population on the planet.
And notice what the assumption does to both sides of this argument. It makes one side need the migration to be a conquest, so that some Indians can be cast as foreign. It makes the other side need the migration not to have happened at all, so that nobody can be. Both are treating a fact about the second millennium BCE as a claim about the twenty-first century CE.
It is not one. A person born in Punjab is from Punjab. That was true of the child of an incoming herder in 1500 BCE and it is true now. Nothing a laboratory finds in a four-thousand-year-old ear bone changes it in either direction.
With that said, here is what the genetics has actually established.
The population of South Asia is built from two very old ingredients — a deep indigenous lineage and an Iranian-plateau-related lineage that split off before farming. Both were here long before this argument starts.
The Rakhigarhi Harappan genome has no Steppe ancestry. It is one woman. The entire Harappan genetic sample is twelve people, and there is no ancient DNA from Punjab in the relevant period at all.
Across a much larger sample, Steppe ancestry is absent before about 2000 BCE and present after about 1500 BCE in the north-west. That is a real result.
It arrives mostly through men. The maternal line stays local. That is the signature of a small incoming group marrying in, not of a population being replaced. For most Punjabis, roughly four parts in five of your ancestry has been in this land since before the hymns.
Ancestry is not identity. Everyone alive is a mixture, and a percentage in a genetics paper has never yet told anybody where they belong.
7Invasion, Migration, or Nothing at All
This is the chapter the whole part has been building towards. Three positions. Each gets its strongest case, and I do not choose between them for you.
7.1 — Three positions, not two
Most public arguments about this are between two sides shouting at each other. There are three positions, and confusing the first two is the single biggest reason the argument goes nowhere.
Invasion. A large body of Aryans came in as conquerors, destroyed the Harappan cities and subjugated the population. This is the nineteenth-century version.
Migration. Groups speaking an Indo-Aryan language moved in gradually over several centuries, in modest numbers, mixing with a much larger existing population, and their language and religion became dominant without their numbers ever doing so. This is what most specialists now hold.
Indigenous origin. Nobody came in. Indo-Aryan speakers had been in this region all along, and if anyone moved, they moved outwards from here.
These are not three points on one line. Invasion and migration differ about violence and scale. Migration and indigenous origin differ about whether anything arrived at all. When you see the three collapsed into two — usually as "invasion versus indigenous" — somebody is attacking a position almost no living scholar defends.
7.2 — The invasion theory and how it collapsed
Part 2 told this story as a case study in the discipline working properly. Here is the full version.
In 1947, Mortimer Wheeler was excavating Mohenjo-daro. In the upper levels of the city he found a number of human skeletons — a few dozen — some lying in what looked like sprawled positions rather than proper burials. He put that beside the Rigveda's Indra, the fort-breaker, smasher of the strongholds of the dasa. And he wrote the sentence that fixed the theory in the public mind for two generations: on the evidence, Indra stands accused.
It was vivid, it was quotable, and it was wrong.
The case was dismantled over the following decades, piece by piece, by archaeologists who went back to the material.
The skeletons are not one event. They come from different levels and different periods, separated by a long time. They are not a massacre lying where it fell. Several are in proper burials. Several show no injury at all.
There is no destruction layer. A sacked city leaves a burnt horizon — a band of ash and collapse running across the site. Mohenjo-daro does not have one. Neither does Harappa. Neither does any major Harappan city.
There are no weapons. No caches of arms, no arrowheads embedded in walls, no siege works, no defensive scramble.
There is no new population. Study of the skeletal material found no sign of an incoming physical type replacing the old one.
And the timing does not work. The Harappan decline had been under way for centuries before the earliest date anyone proposes for an arrival. The cities were emptying before there was anybody to blame.
What this proves: that the Harappan civilisation was not destroyed by an invasion. This is now agreed by essentially everyone, on all three sides.
One more thing about the invasion theory, and it is uncomfortable but true. It was proposed and popularised in the nineteenth and early twentieth centuries by scholars living inside an empire whose own story was that civilisation arrived in India from the north-west, carried by fair-skinned people who ruled the natives. The theory matched that story exactly.
That does not automatically make it false — theories are not wrong because of who thinks of them. But it should have made people check harder, and it did not, for about eighty years.
7.3 — What migration actually claims
The word "invasion" still gets attached to the mainstream position by its opponents, and the mainstream position is not that.
What it claims is undramatic. Over several centuries, roughly between 2000 and 1200 BCE, groups of herders speaking an early Indo-Aryan language moved through the passes into the north-west in small numbers. There was no single event and no date. They found a large, long-settled population whose cities had already come apart into villages. They mixed with it. Their language spread — the way languages spread, by prestige, by ritual authority, by intermarriage, and by the young learning it because it was useful.
The resulting culture was not theirs. It was a fusion, and the Rigveda itself shows the fusion happening: gods who are not Indo-European, ritual practices with no Iranian parallel, and — this is the technical bit — sounds in Sanskrit that no other Indo-European language has and that look very much like the fingerprint of an older local language underneath.
Elite dominance: a model of language change in which a small group with prestige, power or ritual authority causes a much larger population to adopt their language, without replacing them. It is how Latin took over Gaul and how Arabic took over Egypt.
Retroflex consonants: sounds made by curling the tongue back — the t, d and n of Punjabi in words like ਟਾਲ or ਡੰਡਾ. Every language of India has them and no other branch of Indo-European does. One reading is that Sanskrit picked them up from the languages already spoken here. Another is that they developed on their own. It is argued about, but the first reading is the standard one, and if it is right it is direct linguistic evidence of a fusion.
7.4 — The indigenous case at full strength
I said in Chapter 1 that if you cannot make a position sound reasonable, you have not understood it. So here is the indigenous case made as well as I can make it, without the political noise that usually surrounds it.
Where is the archaeology? This is the strongest point and it is not a small one. Migrations of the size being proposed usually leave traces — a new pottery style, a new burial custom, new house forms, new tools, a frontier that moves. In the plains of Punjab there is no clean archaeological signature of an Indo-Aryan arrival. The pottery sequence runs continuously. The settlement pattern shifts eastwards, but it shifts the way you would expect if rivers were changing and people were following the water. If a distinct people came in and their language took over completely, the silence in the dirt is remarkable.
The Rigveda does not remember coming from anywhere. Peoples who migrate usually remember it and put it in their sacred texts. There is no exodus in the Rigveda, no crossing, no lost homeland, no origin beyond the mountains. The composers write about Punjab as home and appear to know nowhere else as home. For a group supposedly arrived within living memory, that is strange.
The dating was never solid. The conventional date for the Rigveda goes back to a nineteenth-century estimate built by counting backwards from the Buddha at a guessed number of years per literary layer. The scholar who made that estimate later said in print that no power on earth could fix a date for the Vedic hymns. That guess hardened into a fact through repetition, and much of the chronology in this part still leans on it.
The Sarasvati. If the river the hymns describe as great and flowing to the sea is the Ghaggar-Hakra, and if that system had ceased to be a major river by around 1900 BCE, then the hymns describing it must be older than 1900 BCE — which is older than any proposed migration. Part 2 laid out why the geological consensus has moved away from a mighty glacier-fed Sarasvati, and also why the sampling behind that consensus is thinner than it is usually presented as being.
The elite left no trace. The migration model needs incomers with enough standing to impose a language on millions. Such groups normally leave something — a palace, a distinctive grave, a fortress, an art style. This one is invisible in the ground and visible only in words.
Those are real arguments. Anyone who tells you the indigenous position is nothing but nationalism has not read it.
7.5 — Where things stand
Everything in this part comes down to this. Here are the three, and then the plainest statement I can make of the state of play.
Now abandoned by essentially all specialists. No destruction layer, no massacre, no weapons, no replaced population, and the Harappan decline starts too early. Its historical interest is as a lesson: a dramatic theory that fitted the politics of its moment, was examined properly, and was dropped.
Held by the large majority of linguists, archaeologists and geneticists working on the period. Four independent lines converge: the language family's reconstructed vocabulary is northern and its earliest branches are western; the domestic horse and spoked chariot appear in the north-west after 2000 BCE and have a known origin outside India; Steppe-related ancestry is absent from the Indus samples and present in the north-west after about 1500 BCE, chiefly through men; and a language extremely close to Vedic Sanskrit is attested in Syria by 1400 BCE, which is very hard to explain if Sanskrit's family began in Punjab. It claims a slow, small-scale movement and a language shift, not a conquest.
Held by a small minority of specialists and a large number of people outside the field. Its strongest cards are the near-total absence of an archaeological signature for any arrival, the Rigveda's complete lack of a migration memory, the shaky foundations of the conventional dating, and the thinness of the ancient DNA sampling from the region that matters most. Its hardest problems are the reconstructed vocabulary of the parent language, which describes a cold northern ecology with no Indian animals in it; the westernmost branches of the family, which would have to have crossed all of Asia from India leaving no linguistic trail; and the horse.
What it does not show: that anyone was conquered. That the Harappans were destroyed or driven out. That the incomers were numerous. That they were racially distinct, whatever that would mean. That the culture which followed was theirs rather than a fusion. And it does not show that the indigenous position has been refuted — it has been outweighed, which is a different thing, and its best objection, the empty archaeology, has never been properly answered.
What would settle it: ancient DNA from Punjab and Sindh between 2500 and 1000 BCE. An unambiguous chariot or horse burial, either found or definitively ruled out. A decipherment of the Indus script — which would end this argument in an afternoon. All three are possible. None is close.
What you should do with this: hold it as most specialists do — as the best current reading of incomplete evidence, not as a settled fact, and certainly not as a statement about anybody alive.
7.6 — Why people care so much
That being first confers ownership.
Every participant in this argument, on every side, is operating on it. It is never stated, because stating it would expose how odd it is.
The reasoning goes: if the Vedic people came from outside, then the culture built on their texts is foreign, and its inheritors are guests here. If they did not, that culture is native, and its inheritors are the rightful owners. Either way, a question about the second millennium BCE is being used to hand out standing in the twenty-first century CE.
Look at what that assumption commits you to if you apply it consistently.
It means the English have no claim to England, since the Anglo-Saxons arrived. It means almost nobody in Europe belongs there, since nearly every European population has Steppe ancestry from the same broad movement. It means the Turks should leave Anatolia and the Arabs should leave Egypt. Applied evenly, the principle empties the world.
And it has a second failure, which is worse. Go back one more layer and the "original" people turn out to have arrived too, from somewhere else, displacing or absorbing someone earlier. There is no bottom to it. Every population everywhere is descended from people who came from somewhere else, if you go back far enough — and in this case "far enough" is not even very far.
The people of Punjab today are overwhelmingly descended from the people who farmed at Mehrgarh and built the cities at Harappa, with a later minority component from the north-west, and speak a language descended from the hymns. All three of those things are true at once. None of them is in tension with any other. And not one of them is a deed.
Belonging is not a genetics result. It never was, in either direction.
That is the end of the argument, and I have kept my promise not to settle it for you. What follows is the one event in this whole period we can actually name people in.
There are three positions, not two. Invasion is dead — no destruction layer, no massacre, no replaced population, wrong dates. Essentially nobody defends it.
Migration is what most specialists hold: small groups over centuries, mixing into a much larger existing population, language spreading by prestige rather than by force.
Indigenous origin has real arguments, and its best one — that a migration of this significance left almost no mark in the archaeology — has never been properly answered.
The evidence currently points towards migration, because four different kinds of evidence converge on the same window. It has not proved it, and the sampling in the place that matters most is close to nothing.
Whatever the answer turns out to be, it settles nothing about who belongs here. Being first is not a deed, and if it were, it would empty the world.
8The Battle of the Ten Kings
Part 1 promised you a fight on the Ravi. It is the only event in this entire period that we can point to, name people in, and put on a specific river.
8.1 — The only battle we have
Four hundred years of Punjab's history, and we have one story. Not a chronicle. Not an inscription. A hymn, and a few supporting verses scattered elsewhere.
It is called the Battle of the Ten Kings, and it is in the seventh book — the family book of the Vasishthas, and one of the older layers.
The bones of it are these. A king called Sudas, of a people called the Bharatas, is attacked by an alliance of ten other kings. The fight takes place on the Parushni, which is the Ravi. Sudas wins. The alliance is broken. Many of the attackers drown in the river.
That is the whole of it, told in perhaps twenty verses. Everything else is inference.
8.2 — Sudas and the Bharatas
Sudas is a chief of the Bharata people, a group based somewhere in the eastern part of the Punjab plain, around the Ravi and towards the Sarasvati. His family is named. His father and grandfather appear. He is a real political figure in a real landscape, which puts him ahead of almost everyone else in the Rigveda.
The Bharatas are not the biggest group in the hymns. They are one people among many in a crowded, quarrelsome plain, and this hymn catches them at the moment they stop being one among many.
Jana: the word for one of these groups. Usually translated "tribe" or "people". It is a body of related families under a chief, not a state — no bureaucracy, no capital, no fixed borders.
Rajan: the chief. Usually translated "king", which is misleading. He led in war, held things together by agreement, and could be removed. A rajan of this period commands something closer to a large extended clan than a kingdom.
Purohita: the chief's own priest, and the second most powerful man in a jana. He performed the rituals that were believed to secure victory. When you read this chapter, remember that the man who wrote it held this job.
8.3 — Two priests
Behind the battle is a quarrel between two men, and it is the most human thing in the Rigveda.
Sudas had a priest called Vishvamitra. Vishvamitra's family composed the third book of the Rigveda, which contains the single most recited verse in Hinduism to this day. He served Sudas, and there are hymns in which he leads Sudas's people successfully across rivers.
Then Sudas changed priests. He took on Vasishtha instead.
We are never told why. But the seventh book, composed by Vasishtha's family, is where the victory hymn sits — and it makes very sure you understand who was performing the ritual when Indra came through. Later tradition turned the two men into bitter enemies and built a large body of story on it.
So the one battle report we have from Bronze Age Punjab was written by a priest who had just taken his rival's job and had an interest in the outcome looking miraculous.
8.4 — The river that was turned
The core of the hymn is the crossing.
Sudas and the Bharatas are on one side of the Parushni. The alliance is on the other, or coming across. In the hymn, Indra makes the river fordable for Sudas, and the waters take his enemies. They are described as being swept away, as not knowing the crossing, as drowning in numbers.
Read plainly, without the god, it describes an army caught by water at a crossing.
Many scholars have gone one step further and read the verses as describing a deliberate breach — Sudas or his people cutting an embankment and putting the river across the enemy at the right moment. The hymn does not say so in as many words. It says Indra did it. But the language of the water being made to go where it did not go before has struck a lot of readers as more than weather.
I flag it as a reading, not a fact. If it is right, it is the earliest description of an army using a river as a weapon anywhere in South Asia, and it happened on the Ravi.
The hymn says the slain numbered sixty-six thousand, six hundred and sixty.
Do not believe it. Look at the shape of the number — six, six, six. It is a poetic figure, the way "a thousand thanks" is a poetic figure, and Vedic poets liked stacking digits.
Think about what would be needed for it to be real. Sixty-six thousand fighting men means a population behind them of perhaps a quarter of a million in one alliance, in a landscape of villages and herding camps, all fed and moved and kept together without carts of any scale, roads, or writing to organise it.
A realistic guess for a fight of this kind is a few hundred to a few thousand men on each side. That is not a criticism of the poem. It is a reminder that a number in a hymn is doing the job an adjective does.
8.5 — Who the ten kings were
The hymn names the alliance. Some of the names are the great peoples who appear throughout the Rigveda — the Purus, the Yadus, the Turvashas, the Anus and the Druhyus, who together make up a standing phrase, "the five peoples". Others are smaller and appear only here: the Alinas, the Pakthas, the Bhalanas, the Shivas, the Vishanins.
Two things about that list are worth your attention.
First — this is a war among insiders. The Purus, Yadus, Anus and Druhyus are not dasas. They are arya peoples, in the sense Chapter 3 described. They speak the same language, worship the same gods, and are the ancestors, in later tradition, of famous lineages. The one battle we have from this period is not Aryans against natives. It is Aryans against Aryans, ten of them against one.
Hold that against everything in Chapter 7. The world of the hymns is not two blocs. It is a crowded plain of quarrelling relatives.
Second — some of the smaller names look like places. Several scholars have suggested that a few of the minor tribal names in the list resemble names known later from the mountain country to the west and north-west. The resemblances are suggestive and are used, in both directions, as evidence about where people came from. They are also exactly the kind of resemblance that is easy to see and hard to prove, and I would not build anything on them.
8.6 — How much of this happened, and why it matters
Take the hymn at the discount Chapter 1 recommended and see what survives.
Probably solid: that a chief called Sudas existed, that the Bharatas existed, that the other peoples existed under those names, that there was a serious fight involving several of them, and that it happened on the Ravi. Praise poetry lies about scale and about causes. It does not usually invent the names of its patron's enemies, because its audience knew them.
Probably not solid: the numbers, the miracle, the completeness of the victory, and Vasishtha's indispensable role in it.
Unknowable: what the fight was about. Water and pasture are the obvious candidates, since that is what these societies fought over. The hymn does not say.
Now the reason a small river fight in about 1400 BCE is in this book at all.
After this battle the Bharatas are ascendant. Within a few generations they and the Purus — their enemies here — appear to have merged, and out of that merger comes the Kuru people, who dominate the next phase of North Indian history from the country east of the Sutlej. The Kurus are where the later Vedic texts are compiled, where ritual is systematised, and where, much later, the Mahabharata is set.
And the name of the winning side, Bharata, travelled. It became the name of the land in Sanskrit tradition — Bharatavarsha. It is the constitutional name of India today.
The Republic of India is named after the winners of a fight on the Ravi, recorded by a priest who had just been promoted.
The Battle of the Ten Kings is the only event in this whole period we can name people in. It was fought on the Parushni — the Ravi. Sudas of the Bharatas was attacked by an alliance of ten and won, with many of the attackers drowned at the crossing.
The hymn was written by Vasishtha, Sudas's newly appointed priest, who had just replaced Vishvamitra and had every reason to make the victory look like a miracle he arranged.
The number of dead — sixty-six thousand, six hundred and sixty — is poetry, not counting. A few thousand is realistic.
It was a war among insiders. The ten kings were arya peoples, not outsiders. The world of the Rigveda is quarrelling relatives, not two blocs.
The winners were the Bharatas. India's name in its own constitution is theirs.
9How Varna Began
The four-fold order appears exactly once in the Rigveda, in the latest layer of the book. What happens to it over the next thousand years is one of the most consequential things in Indian history.
9.1 — One hymn, late in the book
In the tenth book there is a hymn about a cosmic man. The gods sacrifice him, and from his body they make the world — the moon from his mind, the sun from his eye, the sky from his head.
And from four parts of him they make four kinds of people. From his mouth, the priest. From his arms, the warrior. From his thighs, the ordinary producer. From his feet, the servant.
That is the four-fold division. It is the founding statement of one of the most durable social systems ever built, and it is the only place in the entire Rigveda where it appears.
Chapter 1 asked you to keep the layering in your pocket. Open it now. The tenth book is the latest part of the collection, by common agreement among scholars who agree on very little else. Its language is later. Its arrangement is looser. The word used here for the fourth group appears nowhere else in the Rigveda at all.
So the four-fold order is not the foundation of the Vedic world. It is a late arrival within the Vedic world's own oldest text.
9.2 — What the earlier books show instead
Go back into the older books and the picture is quite different.
There are priests, and priests are important and well paid. There are chiefs and fighters. There are people who herd and farm. Nobody would call that a surprise; almost every society has those roles.
What is missing is the fixing. There is no statement that these roles are inherited, sealed, ranked by birth, or divinely allocated. There is no rule against marrying across them, no rule against eating together, no penalty structure.
And there is one verse — a strange, funny, throwaway thing in a hymn about the ritual drink — where the speaker says plainly that he is a poet, his father is a healer, and his mother grinds corn. Three jobs in one household, in one generation.
You cannot build much on a single verse. But it is very hard to write that line in a society where occupation is sealed at birth.
Varna: literally "colour" or "covering". Used in the Rigveda for the contrast between groups, and later for the four-fold order. It is a category — there are four, and they are the same four everywhere.
Jati: literally "birth". A specific, local, marrying-within-itself group, usually tied to an occupation. There are thousands of them, and which ones exist depends on where you are.
Caste: a Portuguese word, casta, applied by Europeans to both varna and jati at once. The confusion this caused has never been fully cleaned up, in Indian or in English writing. When someone says "caste" they usually mean jati. When a Sanskrit text says varna, it means the four.
Janapada: "the place where the people set their foot". The word that replaces jana as groups stop moving and start holding territory. It is the beginning of the state in India.
9.3 — The move east
To see how the loose thing became the fixed thing, you have to follow the people out of Punjab.
Chapter 2 showed the Rigvedic world centred on the five rivers with the Ganga as a rumour at the edge. In the texts that come after the Rigveda — the ritual manuals and commentaries composed over the following centuries — the centre has moved. It sits around Kurukshetra and the land between the Yamuna and the Ganga.
One of those later texts even tells the story of the move, in mythological dress: a fire carried eastwards across the land, drying the marsh as it goes, and settlement following behind it. Strip the god out and it describes clearing wet forest for farming, which is exactly what would have had to happen.
Why east? Because the land east is wetter and, once you can clear it, far more productive. And clearing it required a tool.
9.4 — Iron, and the first kingdoms
Chapter 2 noted that the Rigveda has no iron. From about 1200 to 1000 BCE, iron appears in North India, and it changes what land is worth.
A bronze axe is soft. It will take down a small tree and then need reworking. Bronze also needs tin, and tin has to be traded from far away, which makes it expensive and keeps it in the hands of the few.
Iron ore is everywhere. Once you know how to smelt it, an ordinary farmer can own an axe and a ploughshare, and both will cut things bronze cannot.
Put an iron axe into the forest of the Ganga plain and you can clear it. Put an iron ploughshare into the heavy wet soil underneath and you can plant rice in it. Rice on that land yields far more per field than barley on dry land.
More food per field means more people who do not have to farm. More people who do not have to farm means specialists, soldiers, officials and priests — and somebody to collect from everyone else to pay for them.
That is how you get from a chief with a herd to a king with a treasury. It took roughly five hundred years.
The archaeological signature of this period in the doab is a distinctive grey pottery, and it maps closely onto the region the later Vedic texts talk about. Villages get bigger. Some become towns. The jana, a people who move, becomes the janapada, a people who hold a place.
And a settled kingdom needs something a herding camp does not: a permanent, justified, unarguable order of ranks.
9.5 — The law books harden it
Over the next several centuries the four-fold division is elaborated relentlessly. The ritual texts assign duties. The law books that follow assign penalties — and the penalties are graded, so that the same act carries different consequences depending on the varna of the person who did it and the varna of the person it was done to.
Birth replaces function. Marriage between groups is restricted, then prohibited. Eating together is regulated. A fifth position appears below the four, of people placed outside the system altogether, with the harshest treatment of all.
By the time of the most famous of these law books, the system on the page is rigid, hereditary, hierarchical and enforced.
That a text describes the society, rather than the society its authors wanted.
This is Chapter 1's assumption again, and it does more damage here than anywhere else in the series.
The law books were written by members of the group placed at the top of the system they describe. They state, in detail, that this group should receive deference, gifts, exemption from certain punishments and control of ritual. Read as a description, this is sociology. Read as what it actually is — a set of claims made by an interested party — it is a demand.
And there is hard evidence that the demand was not the reality for a long time. Studies of the genomes of living Indian communities show widespread mixing between groups continuing until roughly two thousand years ago, after which marrying strictly within the group becomes the pattern and stays that way.
Sit with that number. The texts prescribing rigid separation are considerably older than the point at which people actually started obeying them. For centuries, the rules existed on the page and were not the world.
Which cuts against two popular claims at once.
It cuts against the claim that caste as lived today is three and a half thousand years old and Vedic. It is not. The Rigveda has one late hymn.
It equally cuts against the claim that caste is a British invention and was fluid before colonialism. Two thousand years of genetic endogamy is not a colonial artefact, whatever the British later did to codify and freeze it.
The honest answer is the uncomfortable middle one: an old idea, hardened slowly, locked in far later than the texts claim, and locked harder still under later regimes.
That leaves the question of where the idea itself came from, which is a separate matter from when it bit.
Three explanations, and they are not fully exclusive.
A ranked division into priests, warriors and producers appears in several Indo-European societies — in ancient Iran, and arguably in Rome and among the Celts. If those societies inherited a three-way scheme from a common ancestor, then the Indian version is that scheme with a fourth level added underneath for the people already here. The word varna meaning "colour" is then read as marking that addition.
Every complex agrarian society invents a hierarchy, because states need one — someone to collect, someone to fight, someone to produce, someone to do the work nobody wants. The four-fold scheme appears in the Rigveda only in the latest layer and hardens exactly as agriculture intensifies and kingdoms form. That correlation is very tight, and it does not require anyone to have imported anything.
An inherited three-way idea about priests and fighters and producers may well have supplied the vocabulary. But an idea does not become a system that governs marriage for two thousand years unless something in the material world is holding it in place. What did the holding was land, surplus, kingship and the interests of the people who wrote the rules.
What would settle it: better dating of the later Vedic texts, and more genetic work on when endogamy began in different regions and communities. The second is being done now and is one of the few parts of this whole subject where evidence is arriving quickly.
What is not in dispute: that the Rigveda contains one late hymn on the subject and nothing else, and that the rigid system of the law books is many centuries younger than the hymns.
9.6 — Varna is not jati
One last distinction, because almost every public argument about caste runs the two together and becomes unusable.
Varna is the four-part scheme in the texts. It is neat, it is the same everywhere, and it is largely theoretical.
Jati is the thing people actually live in. Thousands of local birth-groups, each marrying within itself, usually tied to an occupation, ranked differently in different regions, and constantly arguing about their own position. A jati that ranks high in one district may rank differently a hundred kilometres away.
The relationship between the two is that the four-part scheme provides the justification and the thousands of local groups provide the reality. The texts explain why there should be a hierarchy. The jatis are what the hierarchy is made of.
Punjab's own history with this is distinctive, and we will come to it. The region has had, for centuries, a comparatively weak Brahmin presence, a strong land-holding peasantry, and religious traditions — Sufi, Sant and Sikh — that rejected the whole apparatus explicitly. The Gurus of Parts 7 and 8 attacked it directly, and the common kitchen where everyone sits in one line and eats the same food was designed to make a point about exactly this. Whether the point was fully carried is a question Part 13 has to face honestly.
The four-fold division appears in the Rigveda once, in the latest book, in a hymn about a cosmic man. The word for the fourth group appears nowhere else in the collection.
The older books show priests, fighters and producers, but no rule fixing them by birth — and one verse of three jobs in one family.
The system hardened as the centre of gravity moved east, as iron made the Ganga forest farmable, and as chiefs became kings who needed a permanent order of ranks. That took roughly five hundred years, and the law books that followed made it rigid.
Genetic evidence says strict marrying-within-the-group becomes the rule only about two thousand years ago — long after the texts demanded it. The rules were written before they were obeyed.
Varna is the four in the books. Jati is the thousands people actually live in. The books supplied the excuse; land, surplus and kingship supplied the force.
10An Honest List of What We Do Not Know
Every part of this series ends here. Two lists, in plain words: what is genuinely open, and what is firm enough to build on. This one has the longest first list in the series so far.
Genuinely unknown
These are open. Not "probably this but let us be careful" — open. Anyone who tells you otherwise is telling you what they hope.
- Whether Indo-Aryan speakers came from outside, and if so how many. The evidence points to migration. It has not proved it, and the number of people involved is a guess in every version.
- Why there is no archaeology of an arrival. This is the strongest objection to the mainstream position and it has never been answered properly. Either the incomers were so few and so similar in material culture that they are invisible, or something is wrong with the model. Nobody knows which.
- When the Rigveda was composed. Roughly 1500–1200 BCE is a reasoned estimate from four indirect arguments. It is not a measurement. The oldest material could be considerably older.
- What the composers of the Rigveda called their own history. They left no origin story, no migration memory, no account of where they had been. We do not know whether that is because there was nothing to remember or because they did not think it worth saying.
- Who the dasa and dasyu actually were. Rival groups, earlier inhabitants, both, neither. The text supports several readings and settles none.
- Whether the "colour" verses refer to skin at all. One ambiguous word, occurring once, and one word with half a dozen meanings.
- What a pur was. The stronghold Indra keeps breaking. Made of stone, of metal, with a hundred walls — the descriptions are not consistent and no one has securely identified one on the ground.
- Whether any Harappan spoke an Indo-Aryan language. Until the script is read, this cannot be answered, and it is the hinge the whole argument turns on.
- What language the Harappans spoke. Still the same wall as Part 2. Dravidian is the most argued-for candidate. Nobody knows.
- Whether the disputed horse bones are horse. Testable with current technology. Largely untested.
- Whether the Parushni was deliberately breached at the Battle of the Ten Kings. A widely held reading of verses that do not quite say it.
- What happened to the Harappan population. Genetically they are largely still here. Culturally, almost nothing of theirs is visible in the Vedic material. How a population survives and its culture does not is not understood.
- How the four-fold order actually spread. We have texts demanding it and genetic evidence for when it was obeyed. The centuries in between are dark.
- Almost everything about ordinary life. What most people ate, how they were governed, what women's lives were like, what anyone who was not a priest or a chief thought about any of it. The Rigveda does not say and there is nothing else to ask.
Solid
Now the other list. These are not certainties in a mathematical sense — nothing in ancient history is — but they are firm enough that a new discovery would have to be extraordinary to move them.
- The Rigveda was composed in and around Punjab. Its rivers are our rivers, named in order, and its ecology is this region's ecology. This is not seriously disputed by anybody.
- It was transmitted orally with extraordinary accuracy for a very long time before being written, and the mechanism that made that possible is understood.
- Sanskrit is related to Greek, Latin, Persian, English and Punjabi, by descent from a common ancestor. The sound correspondences are regular and predictive. This is as firm as anything in the humanities.
- That ancestor language had wheels, horses, wool, snow and birch, and had no word for tiger, elephant or rice.
- "Aryan" is not a race. The Rigveda never defines arya by body. The racial theory was a nineteenth-century European construction laid on top of a linguistic term.
- The Harappan cities were not destroyed by invaders. No destruction layer, no massacre, no weapons, no replaced population, and the decline starts too early.
- Steppe-related ancestry is absent from the Indus-related genetic samples and present in the north-west after roughly 1500 BCE, arriving disproportionately through men.
- Most of the ancestry of people in Punjab today has been in this region since long before the hymns. The later component is a minority everywhere and never close to a majority.
- The domestic horse and the light spoked-wheel chariot originate outside South Asia and appear here in the second millennium BCE.
- A language extremely close to Vedic Sanskrit was in use in northern Syria by about 1400 BCE. Whatever explains that, it has to be explained.
- Books 2 to 7 of the Rigveda are older than Books 1 and 10. Agreed across scholars who disagree about everything else.
- The four-fold division appears once, in the latest layer. The rigid, birth-based system is many centuries younger than the hymns.
- The Battle of the Ten Kings was fought between arya peoples, not between Aryans and natives.
One more thing, and then we move on
I want to end this part by saying something about the shape of the argument rather than its content.
You have just read seventy-odd pages in which the honest answer to the biggest question was "the evidence points this way, and it is not settled". That is not a failure of the writing. It is what the record actually supports, and I would rather hand you an open question than a confident answer I cannot back.
But notice what happens when a question stays open for a long time in public. It fills up. People need an answer, so the gap gets filled with whatever the person filling it already believed. That is true of the nineteenth-century scholars who filled it with race, and it is true of every partisan since, in every direction.
The discipline of leaving a gap empty is unnatural and it is the whole of the method. If there is one thing I would like you to take out of this part and use somewhere else entirely, it is that.
The biggest question in this part is open. The evidence points towards a gradual, small-scale migration in the second millennium BCE, because four independent kinds of evidence converge there. It has not been proved, and the ancient DNA from the one region that matters most does not exist.
What is firm: the Rigveda is a Punjabi book, orally transmitted with great accuracy; Sanskrit is related to Greek, Latin and English; "Aryan" is not a race; the Harappan cities were not sacked; most Punjabi ancestry is ancient and local; and the caste system in the texts is much younger than the hymns.
The one who tells you this is all settled, in either direction, is telling you about themselves.
An open question left open is a result. Filling it in is how the last two centuries of this argument went wrong.
Sources & further reading — Part 3
Timeline of Part 3
Dates before about 600 BCE are estimates, some of them contested. Where a date is argued over, the table says so.
| When | What |
|---|---|
| c. 4500–3000 BCE | Proto-Indo-European spoken. Most specialists place it on the Steppe north of the Black and Caspian Seas; a minority argue for Anatolia or the lands south of the Caucasus. Disputed. |
| c. 3500 BCE | First wheeled vehicles anywhere. Solid wheels, ox-drawn. This puts a floor under the date of the parent language. |
| c. 3300–1900 BCE | The mature Harappan civilisation. Part 2. |
| c. 2200–2000 BCE | The lineage that became every modern domestic horse expands out of the western Steppe. |
| c. 2000 BCE | Earliest known spoked-wheel chariots, buried with horses, in the southern Urals. |
| c. 1900 BCE | Harappan cities in serious decline. The Ghaggar-Hakra system weakening. Population shifting east. |
| c. 2000–1500 BCE | Steppe-related ancestry appears across Central Asia and reaches the South Asian north-west. Absent from all Indus-related samples before this. |
| c. 1700–1500 BCE | On the migration model, Indo-Aryan speakers present in the north-west in small numbers. Disputed — the indigenous position holds that no arrival occurred. |
| c. 1500–1200 BCE | The bulk of the Rigveda composed, in the land of seven rivers. Books 2–7 are the older core. Date is a reasoned estimate, not a measurement. |
| c. 1400 BCE | The Battle of the Ten Kings, fought on the Parushni — the Ravi. Sudas of the Bharatas defeats an alliance of ten. Date approximate. |
| c. 1380 BCE | A treaty in Mitanni, northern Syria, calls Mitra, Varuna, Indra and the Nasatyas to witness. Indo-Aryan horse-training vocabulary in use in the same region. |
| c. 1200 BCE | Books 1 and 10 of the Rigveda composed, including the hymn of the cosmic man — the only appearance of the four-fold order in the collection. |
| c. 1200–1000 BCE | Iron appears in North India. The Ganga forest becomes clearable and its heavy soil becomes farmable. |
| c. 1200–600 BCE | The later Vedic period. Painted Grey Ware in the Ganga-Yamuna doab. The centre of gravity leaves Punjab and does not return for a thousand years. |
| c. 1100–900 BCE | The Bharatas and Purus merge. The Kuru people emerge around Kurukshetra and dominate the north. |
| c. 900–500 BCE | The Brahmanas and early Upanishads. Ritual systematised. The four-fold order elaborated and hardened. |
| c. 700–600 BCE | Kingdoms proper. The jana becomes the janapada. In the north-west, Gandhara becomes a substantial power with its centre at Taxila. |
| c. 550–518 BCE | The Achaemenid Persian empire expands east and reaches the Indus. Punjab enters written history — somebody else's. |
| c. 518 BCE | Part 4 starts here. |
Glossary
Every hard word used in Part 3, in plain English.
| Word | What it means |
|---|---|
| AASI | Short for a deeply ancient lineage in South Asia, descended from the first modern humans to reach the subcontinent. Everyone here carries a share of it. |
| Ancient DNA | Genetic material recovered from old human remains rather than from living people. Tells you about one specific person on one specific date. |
| Arya | The word the composers of the Rigveda use for themselves. Marks the in-group. Later softens to "noble". Never defined by body anywhere in the text. |
| Ashva | Horse. Cognate with Latin equus and Greek hippos. |
| Avesta | The oldest Iranian sacred text. Its language is so close to Vedic Sanskrit that lines convert between the two by regular sound swaps. |
| Ayas | The Rigveda's one metal word. Copper or bronze. Not iron. |
| Bharatas | The people of Sudas, who won the Battle of the Ten Kings. India's name in its own constitution comes from them. |
| Brahmana | Two different things, confusingly. A priest; and a class of later prose texts explaining ritual. |
| Cognate | Two words in different languages descended from the same ancestor word. Father and pitar. |
| Dasa / Dasyu | The Rigveda's words for enemies and outsiders. Dasa later drifts down to mean "servant". |
| Elite dominance | A small group with prestige causing a much larger population to adopt its language, without replacing it. How Latin took Gaul and Arabic took Egypt. |
| Haplogroup | A branch on the family tree of the Y chromosome (father to son) or mitochondrial DNA (mother to child). Lets men and women be tracked separately. |
| Indo-Aryan | A modern label for the branch of the language family containing Sanskrit, Punjabi, Hindi, Bengali and about two hundred others. A statement about grammar, not about people. |
| Indo-European | The whole family — from Punjabi and Persian to Greek, Latin, Russian and English. |
| Jana | A people or tribe. A body of related families under a chief. Not a state. |
| Janapada | "Where the people set their foot." The territory a settled jana holds. The beginning of the state in India. |
| Jati | "Birth." A specific local group that marries within itself, usually tied to an occupation. There are thousands. This is what people usually mean by "caste". |
| Khur | The Indian wild ass. Never domesticated. Its bones are hard to tell from a horse's unless you have teeth. |
| Mandala | One of the ten books of the Rigveda. Literally "circle". |
| Mitanni | A Bronze Age kingdom in northern Syria whose rulers used Indo-Aryan god names and horse vocabulary around 1400 BCE. |
| Painted Grey Ware | A distinctive pottery of the Ganga-Yamuna doab, roughly 1200–600 BCE. The archaeological marker of the later Vedic period. |
| Parushni | The Rigvedic name for the Ravi. |
| Proto-Indo-European | The reconstructed ancestor of the whole family. Nobody wrote it down. It is reached by working backwards from its children. |
| Pur | A stronghold. Indra's title is "fort-breaker". What a pur physically was is not known. |
| Purohita | A chief's own priest. The second most powerful man in a jana. |
| Rajan | A chief. Usually translated "king", which oversells it. |
| Ratha | A light two-horse chariot with spoked wheels. A weapons technology, not a cart. |
| Retroflex | Sounds made by curling the tongue back. Every Indian language has them; no other branch of Indo-European does. |
| Rigveda | "The knowledge of praise verses." 1,028 hymns in ten books, a little over ten thousand verses. The oldest surviving text in any Indo-European language. |
| Sapta Sindhavah | "The seven rivers." The Rigveda's own name for this region. |
| Sindhu | River, and the Indus. The source of the words Indus, India, Hindu and Hindi. |
| Steppe ancestry | The genetic profile of the Bronze Age herders of the grasslands north of the Black and Caspian Seas, and their eastward descendants. |
| Substrate | What an older, displaced language leaves behind in the one that replaced it — usually place names, local words and pronunciation. |
| Sudas | Chief of the Bharatas. Won the Battle of the Ten Kings on the Ravi. |
| Sukta | A single hymn. "Rigveda 7.18" means Book 7, hymn 18. |
| Varna | Literally "colour" or "covering". The four-part scheme of the texts: priest, warrior, producer, servant. There are four, and they are the same four everywhere. |
| Vasishtha | Sudas's priest, whose family composed Book 7 — including the hymn about the battle they had a stake in. |
| Veda | "Knowledge." There are four Vedas. The Rigveda is much the oldest. |
| Vedic Sanskrit | The language of the Rigveda. Older than classical Sanskrit, and different enough to need learning separately. |
| Vishvamitra | Sudas's earlier priest, replaced by Vasishtha. His family composed Book 3. |