How we make these reports

The method behind the Reports section — published so you can check our work, or do your own. Everything here is CC-BY-SA: take it, use it, improve it.

Why we do this

India publishes an enormous amount of official data — on corporate social spending, schools, health, poverty. Most of it sits unread. Yet inside those spreadsheets are system-level facts that change decisions: where money actually flows, who it misses, what a committee could fix this session. We read the data, verify what we find, and publish it in plain language while the decision is still open. That's the whole idea.

The five moves

1 · Frame — start only when it matters now

We pick a public dataset only when there's a live lever attached: a Bill in committee, a rule being written, a budget being set. Analysis without a decision window is filing, not contributing. Before touching the data, we write down the plan and the question: what would this change, and for whom, this week — in six months — in five years — by 2047?

2 · Analyse — find the one honest number

We look at the data's shape first: where it's concentrated, how it's trending, what it contradicts in itself. (Government portals sometimes disagree with their own reports — when they do, that's a finding, and we say so.) Then we join the spending data to a measure of need — poverty, population, learning — and build one number that carries the story. In the CSR series, that number is ₹61 versus ₹64,308: what a poor person's presence attracts in corporate social spending in Bihar versus Goa.

3 · Verify — before anything publishes

Every headline figure is recomputed from the source files in a separate pass, and cross-checked against independent reporting. Where our number matches the press, we say so; where nothing external exists to check against, we say that too. Every report names its own weaknesses — data vintages, proxies, exclusions — in the report itself, not in a footnote somewhere else. A contribution that hides its limits isn't one.

4 · Deliver — one finding, many doors

The same verified finding leaves the lab through several doors, each built from the last: a full report for readers who want the whole argument · a data deep-dive for people who want to check the maths · an interactive dashboard for people who think in charts · a short opinion piece for the public conversation · a two-page brief for the committee that can act. You saw all five on the reports page — they're one analysis, dressed for five readers.

5 · Publish — open by default

Everything lands here under CC-BY-SA with its sources linked and its method visible. The spark goes to social media; the escalation goes to opinion pages; the lever gets a formal submission. And the decisions that matter — what to send, when, to whom — are made by a person, not an automation.

The rules we hold ourselves to

• A live lever, or we don't start. The work must be able to change something.
• One honest number over forty tables. If the finding can't survive being one sentence, it isn't ready.
• Verify before publish. Recomputed, cross-checked, caveats named in the artifact itself.
• Defects are findings. When official data contradicts itself, we report that too — kindly, precisely, with the fix.
• Open by default. CC-BY-SA, sources linked, method public — this page included.
• Humans press the marks. AI helps us read and build; people decide what's claimed and what's sent.

Do it yourself

The datasets are public: csr.gov.in, data.gov.in, NITI Aayog's dashboards, UDISE+, NFHS. The method is this page. If you run it on a dataset we haven't touched — a district, a sector, a scheme — we'd love to see what you find, and we'll link work that shows its sources and its limits. Write to us.

Reference run: the CSR series above — from raw portal exports to a parliamentary submission. First published July 2026.