Plinth

Source Trust Charter

Plinth is a political archival infrastructure: a searchable memory of India's parliamentary record, built only from official primary documents. This charter states, in plain language, what we include, what we exclude, and how we use AI — so anyone can hold us to it.

What Plinth includes

Only official primary documents from the parliamentary record. Every record we store carries a link to its official source.

  • Parliamentary debates and proceedings
  • Bills and legislative texts
  • Voting records
  • Committee reports
  • Official manifestos and budget documents (added in a later phase)

Our launch focus is the Lok Sabha record from 2004 to the present. Rajya Sabha, elections, state legislatures, and earlier years are deferred, not cancelled.

Official government records

Alongside the parliamentary record, we include official government records that carry the same primary-document standard.

  • MPLADS scheme records and eSAKSHI fund-disbursal data
  • CAG audit reports
  • ECI election results and candidate affidavit filings
  • Census and government scheme statistics (data.gov.in, MoSPI)
  • Union Budget documents
  • PIB press releases
  • ECI-filed party manifestos

What Plinth does not include

We deliberately exclude anything that is not an official primary document — even as supplementary material.

  • Opinions and editorial judgement
  • News articles and press commentary
  • Social-media posts and reactions
  • Television coverage and broadcast commentary

How AI is used

AI helps organise and connect the official record. Every AI-assisted field is traceable to the source documents it came from, and no AI output is published without validation against those records.

  • Classifying debates and records by topic
  • Summarising long debates and documents
  • Improving search and ranking over the record
  • Linking mentions of members, parties, and bills to the right records
  • Retrieving the official records that support or contradict a stated claim

How AI is not used

AI never judges a person. We do not build, and will not build, any of the following:

  • Truth or honesty judgements about a person
  • Bias detection or ideology labelling
  • Sentiment scoring
  • Intent or motive attribution
  • Rating or ranking one politician against another
  • Election prediction
  • Promise-fulfilment scoring

The Evidence Layer

Plinth makes exactly one kind of assessment, and only under strict limits: it can describe how a stated claim aligns with the official record. This is record-against-record alignment, always shown alongside the specific official records that drive it. It is never a verdict on a person, and it never ranks one politician against another.

  • Supported by official records
  • Partially supported by official records
  • Contradicted by official records
  • Insufficient public evidence
  • Requires contextual interpretation

Every status cites the records behind it and is fully reversible from those sources. The label describes how the evidence aligns — never honesty, intent, or truth.

Extensions to the Evidence Layer

Four further forms of record-vs-record alignment are permitted, each under the same strict limits as the Evidence Layer above: every status cites its records, describes alignment rather than honesty or intent, is fully reversible, and never ranks one politician against another.

  • Promise evidence-alignment: a public promise is labelled with the same statuses as any other claim — Supported by official records, Partially supported, Contradicted by official records, Insufficient public evidence, or Requires contextual interpretation — never a binary fulfilment verdict and never a percentage headline.
  • Transparent mechanical network metrics: any structural graph metric shown on Plinth always displays its formula and raw inputs beside the number, and is never presented as a top list.
  • Aggregate-only affect: patterns in tone or civility are described only at the level of the House, a session, or a party bloc — never as a label attached to one member.
  • Machine-generated cluster labels: where statements are grouped by similarity, the group label is marked as AI-generated and describes only mechanical features such as shared terms, date range, or number of speakers.

Every one of these forms reuses the same evidence-alignment vocabulary and the same reversibility and non-ranking limits already stated above — nothing here widens what Plinth is permitted to say about a person.

The Plinth Citation

Every statement, vote, and record on Plinth is independently citable, like an entry in a legal database. Each citation carries the speaker, the date, the session, the debate, the official source, and a permanent link — so any claim can be traced back to the primary document it came from.