What was Soma? Scoring seven candidates against the Rig Veda

The question

The ninth book of the Rig Veda, Mandala 9, is devoted to Soma, a ritual drink, and people have long argued about which plant or mushroom it was. This test asks one thing: does the hymns' own description pick out a single candidate, when each is scored by rules fixed in advance?

It was run by an AI model (Claude) on a public digital edition, at the direction of the author Paul L. G. Harris. It is a test of what an ancient text says, a history and botany exercise, and not guidance on any substance.

The test

The rules were written into a pre-registration file before any per-corpus counts. The text came from the VedaWeb edition (Cologne Center for eHumanities), which tags every Sanskrit word with its dictionary headword, so the study worked from the Sanskrit and not from a translation.

Three sets of verses were counted. The Soma book has 114 hymns and 1,108 verses. As controls, 1,691 verses to Agni (the fire god) show what is ordinary Vedic style, and 2,772 verses to Indra, Soma's chief drinker, are contaminated by design.

There were 19 clues, each a fixed list of Sanskrit words: the colour word hári, the stalk, mountains, sweetness, wakefulness, urination, the mad- family for exhilaration, and others. A verse either has a clue or it does not.

Seven candidates were scored: fly agaric, Ephedra, Syrian rue (Peganum harmala), psilocybin mushrooms, cannabis, opium poppy and a non-psychoactive ritual substitute (Cynanchum acidum). Each cell was +1, -1 or 0 from cited botanical facts. Clues about how the drink was handled in ritual score 0 for everyone, because any material could be handled that way. In the main scheme a clue counts only as far as it is commoner in the Soma hymns than in the Agni hymns.

The decision rule was fixed too. A candidate "fits best" only if it comes first in at least 95% of the hymn-reshuffling runs and of the runs varying the uncertain scores, and survives dropping any single clue.

What came out

The Soma hymns really are different, about 13 times richer in specific vocabulary than the Agni hymns. But most of that is about how the drink was handled in ritual (6% to 15% of Soma verses), which fits any candidate. The clues that could tell candidates apart are rare:

Mountains are Indra's word more than Soma's. "No leaves or roots" is not evidence, because the fire hymns are just as silent. The urine clue rests on one verse, RV 9.74.4.

On the main scheme the scores, from -1 to +1, were cannabis 0.244, Psilocybe 0.195, fly agaric 0.170, Ephedra 0.152, Syrian rue 0.069, opium poppy -0.035 and the substitute -0.043. Is that winner robust? No.

• Reshuffle the hymns. Cannabis came first in practically every run, but this tests only sampling noise in the text.
• Vary the uncertain scores. Cannabis first 45%, Ephedra 24%, fly agaric 19%, Psilocybe 8%.
• Drop one clue. Without the colour word, fly agaric wins. Without "sweet", Psilocybe. Without the exhilaration words, Ephedra wins and the non-psychoactive substitute comes second.
• Change the translation. Griffith's English of 1896 gives cannabis. Geldner's German of 1951 gives Ephedra outright on equal weights. At RV 9.74.4 Griffith has "rain down" where Geldner has pissen, so a reader of Griffith alone never meets the verse the urine argument rests on.
• Run it on the wrong hymns. On the fire hymns the same machinery, on equal weights, picks fly agaric in practically every reshuffle. A method that finds a mushroom in poems about fire is not identifying anything.

No candidate met the rule. The verdict is that the text alone does not decide.

What this does not show

• It does not show which organism Soma was, or that it was none of these. It does not show that a psychoactive plant is required either: without the intoxication words, the substitute is competitive.
• The scoring matrix is the study's own. Other experts would score some cells differently, and the uncertainty runs show it matters. The botany came from encyclopaedia-level sources, so a specialist re-check, especially of taste and colour, is advisable.
• Word lists miss metaphor and some words are ambiguous: go means both cow and milk, and hári covers green, yellow, tawny and bay.
• The Indra control is contaminated. Griffith left 7 stanzas of Mandala 9 untranslated.
• Only Mandala 9 was read. The later ritual texts, the Avesta and the archaeology were not used.

Where it goes in the books

This is "Test two: what was Soma?" in the Interlude of The Oldest Medicine, Book 4 of Hidden in Plain Sight. Chapter 5, "We Have Drunk Soma", sets out the candidates and who proposed them.

You are invited to check it. The corpus is public (github.com/cceh/c-salt_vedaweb_tei), the botanical sources are listed with the study, and a Vedic scholar or botanist who would score the matrix differently is welcome to say so.