Where our data comes from

Anonymous product measurement

We count model-page views, comparison-page views and use of the EMI calculator to improve the site. Our event store keeps only the date, event type, page path and aggregate count. It does not store names, contact details, IP addresses, identifiers, calculator amounts or search terms. This measurement uses no tracking cookies and respects browser Do Not Track and Global Privacy Control signals.

The sources we use

WikidataCC0 — public domain, no attribution required

Model identity, nameplate history, production years, manufacturer relationships and stable entity IDs. Structured data in the main, property and lexeme namespaces carries no downstream obligation at all, which makes it the cleanest source available to us. It is a discovery and lineage source, not a specification source — we deliberately ignore its dimension and power figures, which are usually the global-market number rather than the Indian one.

Manufacturer brochures and official specification pagesFacts are unprotected; layout, photography and prose are not

The best specification source: authoritative, variant-level, dated and citable. India rejected the 'sweat of the brow' doctrine in Eastern Book Company v D.B. Modak (2008) and has no database right, so a table of variant specifications is not protected by the effort of compiling it. The photographs and the copy in the same document are protected, and we do not use them. We extract the facts, rebuild the presentation ourselves, and cite the brochure and its date.

Government open data (data.gov.in)GODL-India — commercial use expressly permitted, attribution required

Gazetted in February 2017, granting a worldwide, royalty-free licence to use, adapt and publish for lawful commercial purposes, conditional on an attribution statement naming provider, source, licence and URL. This is the strongest licence position available to us, and the attribution obligation is enforced at render time rather than remembered.

Bharat NCAP and Global NCAPGovernment-published crash test results

Per-variant safety ratings. A rating applies to the specific body and safety equipment tested, which is why we attach it to variants rather than to a nameplate.

State Motor Vehicles Taxation Acts and gazette notificationsPublic law

The road-tax table. There is no single national source for this, so it has to be built from 36 state and union territory schedules with an Act section, notification number and effective date on every row. It is the most expensive thing we build and the most valuable, because it is our own computation from published law rather than someone else's data.

The sources we refuse, and why

We do not scrape competitor car portals. Not for specifications, not for images, not “just to bootstrap”, not for comparison. Ten of them are on a permanent deny list held both in the database and as a compile-time constant, so a mis-configured database cannot open one.

The commercial reason is decisive on its own. Google's spam policies name scraped content and scaled content abuse as violations, and list “scraping feeds, search results, or other content to generate many pages” verbatim. Penalties run to complete removal from Search. This is a business that lives on organic search, and spam updates now ship roughly quarterly — a scraped catalog gets tested against enforcement every three months. Even in a world where scraping carried no legal risk, it would still be a Google-penalty business model.

The legal reasons reinforce it rather than lead. Photographs are artistic works protected for 60 years from publication; editorial text is protected with no de minimis word count; India's fair-dealing exceptions in section 52 are a closed list that a commercial catalog republishing someone else's pages does not fit. Section 43 of the IT Act creates liability for downloading or extracting data from a computer resource without the permission of the person in charge, with no intent requirement.

And three operational lines we will not cross: we never authenticate to a third-party site in order to extract data, we never bypass an anti-bot measure, and we never ingest personal data — no named user reviews, no dealer staff details, no seller phone numbers.

How the policy is enforced in code

• There is exactly one function in the codebase that can make an outbound request. Everything else goes through it.
• It refuses any domain without an explicit allow-list row. The default is deny, and adding a row requires a named human and a terms-of-use review date — enforced by a database constraint, not a convention.
• It honours robots.txt. Where an operator publishes API terms that govern a documented endpoint instead, that exception is recorded with a reason, a URL, a reviewer and a date, and every request taken under it is counted separately in the audit log.
• It enforces a per-domain rate limit and politeness delay from a token bucket held in the database, so adding workers cannot increase the request rate against a source.
• It identifies itself with a real user agent and a contact URL.
• It caches every raw response by content hash, so re-extracting after a parser fix costs zero requests to the source.
• Every attempt — allowed, refused by policy, refused by robots, or rate-limited — is written to a log we would be comfortable handing to a lawyer.

Images

This is the genuinely hard problem, and we would rather tell you how we are handling it than pretend it is solved. Specifications can be re-keyed; photographs cannot. A press-kit download is not a licence — manufacturer press assets are near-universally marked for editorial use only, and a commercial catalog is not obviously editorial.

So for now every car on this site is drawn, not photographed: a profile generated from that variant's own measured length, height, wheelbase, ground clearance and tyre size. It is first-party by construction, it costs about 1.5 KB instead of a 45 KB photograph, and — because two cars can be drawn on one shared millimetre scale — it answers a question no photograph on any Indian car site answers, which is whether one SUV is actually bigger than another.

Real photography arrives through written permission from manufacturers, our own shoots, and owner submissions under a licence they grant us knowingly. An asset cannot be published without a recorded licence, and where that licence is a manufacturer permission, the permission itself has to be stored against the asset. That is a database constraint too.