Index

Evidence, and what it will not support.

An investigation of South Asian history built on primary sources, corpus data and published genetics — with the limits of each stated on the page rather than in a footnote.

Claim

Stated in one sentence, at the strength the evidence supports.

Evidence

Cited to text, corpus count, or dataset. Versions pinned.

Whose record

Who transmitted this, and what they gained from what it says.

Counter-evidence

The strongest case against, made properly.

What would change it

Named in advance, so the claim is falsifiable.

Those five fields appear on every entry. The third is the one that does the most work: almost every text in this canon was kept by people with an interest in what it says.

Instruments

Things you operate

Interactive. The data is real, the gaps are visible, and the controls let you check the claim yourself rather than take it.

Arguments

Things you read

Each takes one mechanism and follows it through the sources.

Sources

Things you read for themselves

Primary material presented without an argument wrapped around it.

Working notes

Findings that failed their own tests

Not a public page. An internal record, kept so that claims which did not survive checking are not quietly recycled — and so that anyone picking this up later knows which attractive-sounding results have already been tried and abandoned.

FailedUnconditioned retroflexion increases across the Rigveda's strata

Significant at token level (z = 4.63); flat at type level; absent entirely on surface forms (z = 0.32). An artefact of dictionary normalisation. Do not revive.

ReversedMixing ended earlier in northern and upper-status groups

The opposite. IE speakers average 72 generations, Dravidian 108 — and the northern dates are younger because those groups reject the single-pulse model.

FailedMeluhha and Ur named each other

Only Mesopotamia named. The Indus script is undeciphered; there is no Indus text naming anywhere.

FailedTutankhamun's lapis is chemically traced to Badakhshan

Attributed by elimination. Much of the mask's blue is coloured glass.

OpenSerendib and Ceylon derive from Sīhaḷa

Contested. Tamil Īḻam may be independent or prior. Direction unresolved.

FailedThe colour-as-people usage is absent from the Rigveda's oldest layer

It is present. kṛṣṇáyoni, "black-wombed," RV 2.20.7, Archaic stratum.

FailedRudra is a marginal Vedic deity

135 occurrences, 69 in the family books. More frequent than Viṣṇu.

FailedThe teaching-kings are an eastern phenomenon

Pravāhaṇa is Pañcāla; Aśvapati is northwestern Kekaya.

FailedThe śrāddha / karma contradiction was never resolved

Resolved scholastically — the twelve-day sapiṇḍīkaraṇa and the transit model.

FailedTherīgāthā is the earliest women's literature

Enheduanna, c. 2300 BCE, is earlier. The defensible claim is earliest anthology by multiple women.

FailedProto-Dravidian broke up c. 1500–1000 BCE

Krishnamurti places it in the early third millennium. Telugu's branch may predate 2000 BCE.

WeakIndus-Periphery ancestry was distributed subcontinent-wide before the steppe arrived

A model output from modern genomes plus ancient outliers found outside India. One low-coverage Harappan genome exists; no ancient DNA from South India of that period.

SupersededANI % as an analytic variable

Merges Iranian-farmer-related and Steppe ancestry. Rebuild on the three-way model; the sharper published finding is that priestly-status groups carry more Steppe ancestry than their Indus-Periphery proportion predicts.

FailedPāṇini mentions Greek script

yavanānī is in Kātyāyana's vārttika on 4.1.49, c. 3rd c. BCE — not the sūtra.

Standing rules: earliest-attested is the only claim a documentary record supports · undeciphered is not negative · a developmental sequence beats a first attestation · version-pin every citation · state what would change the conclusion.