What did the Buddha actually say — before his words became a religion?

For 2,500 years, the Buddha's teachings have been carried inside one of the largest bodies of religious literature in human history: the Pali Canon, 36,434 passages spanning millions of words. Monks copied it by hand. Councils recited it from memory. Commentators added layers of explanation. Systematizers turned living wisdom into doctrine. And somewhere inside all of that — buried under centuries of transmission, elaboration, and institutional devotion — are the oldest words. The ones closest to a historical human being who sat under a tree in northern India and figured something out.

The question has always been: which ones?

In 2026, we asked a computer to find them.

Not by searching for meaning. Not by applying doctrinal criteria. By listening — statistically — to the sound of the language itself. The rhythms, the vocabulary distributions, the grammatical fingerprints that mark one era of composition from another. Five independent computational methods, each using a different mathematical framework, each knowing nothing about the others' results.

They converged on the same answer.

The oldest recoverable stratum of the Buddhist canon is the Aṭṭhakavagga — sixteen ancient poems that predate the great prose collections by centuries. In them, the Buddha sounds nothing like the systematic teacher of later tradition. He is urgent, poetic, paradoxical. He is not building a religion. He is trying to get someone to put something down.

And when the computer went looking for everything else in the canon that sounds like these poems, it found surprises: the Kālāma Sutta, the oldest charter of free inquiry in recorded history. The Tevijja Sutta, where the Buddha asks Brahmin priests which of them has actually seen the god they're arguing about. A discourse in the Majjhima Nikāya where a man blind from birth insists he knows exactly what white looks like. A creation story that begins with beings made of light, eating the earth, and forgetting how to glow.

This book is two things at once.

For general readers: a translation of the oldest Buddhist texts that reads the way great literature should — as if the Buddha is speaking now, to you, about things that have not stopped being true. The translations are in plain English. The commentary assumes nothing. The scholarship is in the footnotes where it belongs.

For scholars and the technically curious: a complete account of the computational methods — FastText semantic embeddings, formulaic density analysis, Cosine Delta stylometry, metrical profiling, morphological archaism scoring, and Bayesian phylogenetic analysis using BEAST 2.7.7 — with full technical appendices, data tables, and reproducible methodology.

What you will find inside:

• Complete translation of the Aṭṭhakavagga: sixteen poems on desire, argument, old age, and the silence at the end of all positions
• Complete translation of the Pārāyanavagga: sixteen students who walked hundreds of miles to ask one question each
• Computer-retrieved passages from the prose Nikāyas — texts the algorithm ranked highest for archaic signature — with full commentary
• Nine technical appendices covering every computational method in detail
• The story of how five independent methods, designed for everything from biological evolution to demographic linguistics, found the same ancient layer

The religion came later. The words came first.

This is what they sound like.