How this site is built — and what it won’t claim.

Swaveda is a small site about Indian history. It is written by a large language model under hard editorial constraints, fact-checked by a second model, and reviewed by a human before contested or flagged drafts publish. This page describes that process plainly so you can decide how much to trust what you read.

The source corpus

The model does not write from its training data. It writes from a curated library of sources we maintain — peer-reviewed papers, archaeological reports, primary-text translations, and a small number of accessible synthesis books. Each source has a trust tier:

• Tier 1. Peer-reviewed papers in venues like Science, Cell, Nature, and Antiquity.
• Tier 2. Reputable but contested or older work, or non-peer-reviewed reports from established institutions.
• Tier 3. Useful background — accessible synthesis, popular history — cited cautiously.

The full library is browsable on the sources page. The library is small at launch and grows over time.

The daily pipeline

Each day, the system drafts five short stories — one each on genetics, archaeology, linguistics or texts, daily life, and one piece on a contested or commonly-misframed topic. For each story:

1. Two or three sources are pulled from the library that haven’t been used in the last month.
2. A first model writes the story using only those sources. Inline citations are required.
3. A second model fact-checks the draft against the sources, looking for unsupported claims, fabricated quotes, overconfident dating, and nationalist framing in either direction.
4. Drafts that pass the check on a non-contested topic publish. Drafts that touch a contested topic, or that the fact-checker flagged, queue for a human to publish, edit, or reject.

What “contested” means here

Some questions in Indian history are still actively debated by serious scholars. When a story touches one of those questions, it gets a ⚖Scholarly debate badge and presents at least two scholarly positions in their own terms, without picking a winner. Topics this currently includes:

• The Indo-Aryan migration question — including the date, route, and demographic scale of any movement, and the relationship between archaeological cultures and language families.
• The dating of the Vedic corpus and the Mahabharata and Ramayana traditions.
• Whether the Indus Valley script is a writing system encoding language, and what language family it might encode if so.
• The identification of the Saraswati river with the Ghaggar-Hakra paleochannel, and what that does or does not imply.
• The origins of caste — its relationship to Vedic social categories, regional variation, and when endogamous practices became fixed.

What we won’t do

We do not state any of the following as fact, in either direction: that the Indus Valley script has been deciphered, that the Aryan question is settled, that the Mahabharata war happened in 3102 BCE (or in any other specific year), or that any modern political claim about ancient ancestry is supported by current evidence.

We do not use the registers of grievance or triumph. No “greatest civilization,” no “the West refuses to accept,” no “lost golden age.” Tradition is described as tradition; evidence is described as evidence; the two are not silently merged.

Corrections policy

When we get something wrong — a misread source, a date that shouldn’t have been stated as fact, a claim that the cited source doesn’t actually support — we fix it visibly. The original text is struck through or replaced and a dated correction note appears at the bottom of the article. Email corrections@swaveda.com if you spot something. Please include the article URL and the source you think we should be reading.

Limits we want you to know

The corpus is incomplete. The pipeline can make mistakes the fact-checker doesn’t catch. The reviewer is one person. None of this replaces reading the underlying papers and books, and the point of citing everything is precisely so you can.