Evidence page · chronology and archaeology
150 mentions of pur. Ninety-nine of them broken for one patron, repeatedly, across separate hymns by separate poets. And the chronology rules out the answer everyone reaches for first.
Books ordered with the family books first. Red: named enemies. Green: categories and the fort word itself.
Śambara concentrates in Book 6 (6) and Book 1 (6) — Bharadvāja's book and the late compilation. Cumuri is almost entirely Book 6 and Book 7. Pipru spreads thinly. Book 3 — Viśvāmitra's — has almost none of them at all, which matters: the fort-breaking material is not evenly distributed among the priestly families. It belongs mostly to Bharadvāja's clan and to the later compilers.
Śuṣṇa runs the other way: 11 in Book 1 and 7 in Book 10 — the two latest books — against 5 in Book 6. He grows over time. That is what mythologisation looks like in a count: a figure who becomes more prominent as the events recede.
And rakṣas peaks in Books 9 and 10. The generic demon arrives as the named ones fade.
Correction to an earlier page here: by strata, pur looked like a declining Archaic concern. By book it is heaviest in Books 1 and 10. Those two measures disagree because Books 1 and 10 are large and internally mixed. The honest statement is that fort-breaking is present throughout and the named fort-holders are not.
| Event | Date |
|---|---|
| Mature Harappan urbanism — Mohenjo-daro, Harappa, Dholavira at full extent | c. 2600–1900 BCE |
| Indus deurbanisation — cities emptied, no destruction layers found | from c. 1900 BCE |
| Steppe-related ancestry enters South Asia | c. 2000–1500 BCE |
| Rigveda family books composed | c. 1500–1200 BCE |
By the time the oldest Rigvedic hymns were composed, the great Indus cities had been abandoned for three to four centuries. There is no destruction layer at Mohenjo-daro or Harappa. The collapse is associated with hydrological change — the Ghaggar-Hakra system drying, monsoon shifts — not with an attack.
Whatever the ninety-nine forts were, they were not Mohenjo-daro. That city fell to a river, and it fell before anyone was writing about it.
what the rivers place them
The hymns will not name the forts' location, but they name their world — and this platform has counted it. Every river named in the Rigveda, tallied by book, gives a geography that is not ambiguous:
The world of Śambara, Divodāsa and the ninety-nine forts is the Sarasvatī–Sindhu region and the northwest — the same territory the Indus civilisation occupied, and emphatically not the Ganges valley. So the instinct that these forts sit in Indus country is geographically right. What it cannot survive is the chronology above: the great cities were abandoned three to four centuries before these hymns were composed, and they show no destruction layer.
The defensible statement is therefore narrower and stranger than either simple answer: the hymns describe fortified settlements with stored wealth, in the territory the Indus civilisation had occupied, at a period after its great cities were already ruins. Whatever was being stormed was in Indus country — but it was not the Indus cities. Late Harappan and post-urban settlements persisted in this region, and they are the only candidates the chronology permits. The right neighbourhood, the wrong century. No specific site has been identified with any named fort, and this platform does not name one.
River counts: machine-verified on this platform's Rigveda corpus (see the map in the text). Abandonment chronology: as cited above. corpus + archaeology
This is the single most important chronological constraint on the whole subject, and it cuts against the reading this platform would otherwise prefer.
| Candidate | For | Against |
|---|---|---|
| BMAC fortresses Gonur Depe, Togolok 21, Dashly 3, Sapalli Tepe — Bactria and Margiana, c. 2200–1700 BCE |
These are actual massive fortified structures — monumental mud-brick, concentric walls, square and circular plans. Right date, right geography, on the route between the steppe and the subcontinent. Parpola argued explicitly that the Dāsa forts are BMAC. And the Rigveda's Archaic layer knows the Kabul, Kurram and Gomal rivers | Narasimhan et al. 2019 found BMAC individuals carry little steppe ancestry, and steppe groups appear to have largely bypassed the BMAC heartland. And the chronology does not close — see below |
| Post-urban Harappan settlements Cemetery H, Late Harappan sites in Punjab and Haryana |
Contemporary with the family books. Settled, walled, and in the right place | Smaller and less monumental than "ninety-nine forts of metal" implies |
| Local fortified settlements of no known archaeological identity | Requires nothing to be proved | Explains nothing either |
| Mature Harappan cities | The size fits | Ruled out by three to four centuries and by the absence of destruction layers |
Mohenjo-daro is estimated at 30,000–40,000 people at its height. The Indus zone as a whole may have held one to five million. Thirty thousand or a hundred thousand are not impossible figures for that world.
What still makes the Rigvedic numbers formulaic is their repetition — ninety-nine forts recurs across separate hymns for Divodāsa, for Pūru, for different poets. A number that appears as a constant is a convention. And the same genre reports sixty thousand cattle paid to one poet.
So: the scale was demographically possible and the specific figures are literary. Both are true, and I previously stated only the second.
You are right and my previous reading was too neat. Vṛtra "the obstructor" and Namuci "not letting go" describe someone who will not release something. That is the language of a dispute over goods, not necessarily of drought.
And the Paṇi collocations point the same way: hoard 40× · hide 19× · wealthy 11× · steal 70× · cattle 4×. An entire enemy category defined by withholding property.
Only Kuyava — "bad grain" — and Śuṣṇa — "the drier" — are unambiguously agricultural. The rest of the transparent names describe refusal, and refusal is something people do.
So the split is better drawn as: a small drought group, a large withholding group, and an opaque group who hold forts. The withholding group and the fort group may be the same people described two ways.
What is established: these three have no accepted Indo-Aryan etymology. In a corpus where Kuyava, Śuṣṇa, Vṛtra, Vala, Namuci and Vyaṃsa all parse transparently, opacity is a signal.
What is not established: what language they come from. Candidates include Dravidian, a Munda-related family, a BMAC substrate, and lost languages with no descendants. No regular sound correspondences have been demonstrated for any of these to a Dravidian source, and a resemblance-based derivation would be the exact error this platform documents elsewhere.
The honest position: three foreign-sounding names of fort-holders, source unidentified. A decipherment of the Indus script would be the thing most likely to settle it — and it is worth noting that if these names are Indus-region names, they are among the very few Indus words that survive anywhere, recorded by their enemies.
Not "nobody has shown it" — an actual test, with a control group
Proto-Dravidian has hard phonotactic constraints (Krishnamurti, ch. 4): no aspirates at all · no word-initial voiced stops · no initial clusters · no initial r-, l- or retroflex · no medial stop-plus-liquid clusters. Nasal-plus-stop clusters are permitted and characteristic. Sanskrit ś is the regular rendering of Dravidian c.
Run every relevant name through those constraints. Indo-Aryan names are included as a control.
| Name | As Dravidian would be | Result |
|---|---|---|
| Śambara — fort-holder | cambara | COMPATIBLE — and the medial -mb- is a nasal+stop cluster, characteristic of Dravidian |
| Cumuri — fort-holder | cumuri | COMPATIBLE — clean CVCVCV, the canonical Dravidian shape |
| Pipru — fort-holder | pipru | INCOMPATIBLE — medial -pr-, a stop+liquid cluster Proto-Dravidian does not permit |
| Vṛtra | vṛtra | INCOMPATIBLE — same -tr- problem |
| Śimyu · Śṛbinda · Anarśani · Kīkaṭa · Kulitara | cimyu · cṛbinda · anarcani · kīkaṭa · kulitara | COMPATIBLE — the minor enemy names largely pass |
| CONTROL — Indo-Aryan patrons and clans | ||
| Divodāsa · Dabhīti · Sudās | — | INCOMPATIBLE — initial voiced stop or initial s- |
| Bṛbu · Balbūtha — the two praised outsiders | — | INCOMPATIBLE — initial b-, and Balbūtha has an aspirate |
| Ṛjiśvan | — | INCOMPATIBLE — initial ṛ- |
| Turvaśa · Yadu · Kutsa · Mṛgaya | turvaca · yadu · kutsa | COMPATIBLE — and these are Indo-Aryan clan names. The test's weakness, shown |
Compatibility is weak evidence. Turvaśa, Yadu and Kutsa all pass and they are Indo-Aryan. Most CVCVCV shapes pass, so "could be Dravidian" is close to worthless on its own.
Incompatibility is strong evidence, and that is where the result lies. Pipru fails. The medial -pr- cluster does not occur in Proto-Dravidian. Whatever Pipru is, it is not a Dravidian name transmitted intact — it could only be a heavily corrupted one, and corruption that creates a forbidden cluster is not a normal borrowing outcome.
So of the three opaque fort-holders: Śambara and Cumuri are phonotactically compatible with Dravidian; Pipru is ruled out.
Compatible is not evidence of origin. Incompatible is evidence against it. Only Pipru gives a clean answer, and the answer is no.
And the control result is informative in its own right: the Indo-Aryan patron names — Divodāsa, Dabhīti, Sudās, Ṛjiśvan — all fail, on initial voiced stops, initial s-, and initial ṛ-. The enemy names pass the filter that the patron names fail. That is a real asymmetry in a small sample, and it is the strongest thing this test produces.
What would settle it: a proposed Dravidian etymon for cambara or cumuri with regular correspondences to attested Dravidian forms. I have not found one and am not going to construct one — inventing an etymology to fit is the exact failure mode this platform documents.
If the family books are conventionally dated 1500–1200 BCE, and BMAC's fortified phase ends around 1700 BCE, then the BMAC forts were already gone when the hymns were composed. The two claims do not sit together, and I presented them as though they did.
Either the family books are older than the conventional dating, or the forts are not BMAC. It cannot be both.
And there is real support for the earlier option. The Mitanni Indo-Aryan material — Mitra, Varuṇa, Indra and the Nāsatyas invoked in a treaty in northern Syria around 1380 BCE, with Indo-Aryan numerals in a horse-training manual — requires that Indo-Aryan had already separated from Iranian and travelled a very long way well before 1500 BCE. A language cannot be in Syria at 1400 and undifferentiated at 1500.
That pushes Proto-Indo-Aryan back toward 2000 BCE or earlier, and with it the plausible start of the Rigvedic tradition. Which would put the Archaic layer closer to the BMAC window — but at the cost of the dating everyone cites.
This is not resolved by picking whichever date suits the argument. The honest position is: the fort material and the BMAC fortresses fit each other in kind and in geography, and the conventional chronology puts them a few centuries apart. That gap is unresolved, and anyone who tells you otherwise is choosing a date to fit a conclusion.
They were not at the Ganges by 1500 BCE. Gaṅgā occurs once in the entire Rigveda, in Book 10, the youngest stratum. Yamunā three times. Himavant once, also Book 10.
Meanwhile Sindhu occurs 204 times, and the Kabul, Kurram and Gomal rivers are in the Archaic layer.
The corpus knows Afghanistan before it knows the Ganges. Whatever the dating, the geography of the family books is the Indus system and the northwest — which keeps the BMAC region in play as a setting even if the fortress chronology does not work.
RV 1.33.7: the Dasyu burned from heaven.
RV 1.175.3: burn, like a vessel with the flame, the lawless Dasyu.
RV 4.28.3: Indra smote down, Agni consumed the Dasyus before noon —
and of those who gladly sought a hard-won dwelling he cast down many a thousand.
That last verse is the one to sit with. Agni burns them; the victims are described as people who had sought a dwelling; and thousands are killed. Fire, a settlement, and a time of day.
But no verse joins burning to sleep. The five sleep-verses (Dhuni and Cumuri, all for Dabhīti) mention magic and weapons — 4.30.21: "with magic power and weapons sent to slumber" — not fire. Combining them would be my invention, and this page does not supply a detail the text does not.
On sleep-slaying across cultures: it appears in Norse saga and Irish material, and the Greek Rhesus episode in Iliad 10 is a night raid on a sleeping camp. Whether these constitute a shared Indo-European motif or simply reflect that night attacks work is not something I can settle, and the parallel has been asserted more often than demonstrated.
The Battle of the Ten Kings (RV 7.18) is Ārya against Ārya. Sudās of the Bharatas fights a coalition that includes the Pūru, Yadu, Turvaśa, Anu and Druhyu — all Indo-Aryan clans, several of them named elsewhere as recipients of Indra's help.
Pūru appears 21 times, Turvaśa 20, Yadu 15 — and they are on both sides depending on the hymn. Divodāsa's forts are won for him; Pūru gets ninety forts of his own at 1.130.7.
So the Rigveda records at least three kinds of conflict — against opaque-named fort-holders, against withholders of cattle, and against other Indo-Aryan clans — and the vocabulary of dāsa and dasyu is not reserved for the first kind. Treating every conflict in the corpus as invaders against natives flattens a record that is messier and more interesting than that.