Skip to content
ArivuஅறிவுArivu

How a figure gets on this platform

Methods

The reconciliation wedge only works if our own definitions are stated, versioned and stable. This is the page the whole platform rests on.

The pipeline

01

Ingest

GODL bulk from data.gov.in, DES PDFs, announcements and public registers — affirmatively licensed layers only, never a commercial compilation.

02

Normalise

Entity resolution: dedup on CIN → website hostname → normalised name and city. CIN held as a dated attribute, not the key.

03

Verify

Human curation, claim-and-correct, a last-verified stamp on every record. Curation is permanent, not a phase.

04

Publish

Report surface, data surface, API and bulk download — every figure carrying its tag, its as-at date and its source.

The tag standard

Three tags. No fourth.

An estimate never becomes a fact by repetition.

Verified against a primary source.
An organisation's own claim, reported as such.
An estimate, including our own arithmetic.

Nothing tagged [C] leaves this building as though it were [A].

The announced-to-deployed discount

A headline investment figure for any Indian state should be read at 20–40% of face value. The discount is 60–80% C, derived from public filings, and applied to every headline figure on this platform and stated wherever it is used.

Why CIN is not a permanent identifier

A Corporate Identity Number changes when a company converts its type, relocates between states, or lists or delists. We therefore hold CIN as a dated attribute against a stable internal identifier, not as the key itself. Platforms that key on CIN silently lose companies at exactly the moments those companies become most interesting.

Claim and correct

The subject of a record is the cheapest verifier available. Every firm page invites its own subject to correct it, and every correction is logged publicly with the date it was made and who made it. We do not silently amend.

The corrections log

We publish what it costs us to say.

A year with none would mean nobody was recording them.

  • 19 Aug 2026Material

    Commitments tracker v1.3 → v1.4

    Was: Two commitments from the same counterparty, announced eleven months apart, were counted separately.

    Now: The superseded commitment is excluded. The traced total falls.

    Found by External reader, with consent

  • 30 Jul 2026Material

    Hosur cluster output v2.0 → v2.1

    Was: FY output was described as verified. It was provisional at the time of publication.

    Now: Marked provisional and the tag is corrected from [A] to [C].

    Found by Internal review

  • 2 Jul 2026Minor

    Sriperumbudur employment v1.1 → v1.2

    Was: An establishment count double-counted a shared-services entity.

    Now: The entity is counted once; employment falls.

    Found by Anonymous, by email