How to read these numbers

Where the data comes from

Each state's package in xKDR/india-public-finance-portal holds the printed detailed expenditure volumes (PDFs), the extracted line-level CSVs, and a full set of arithmetic checks. This site is a static rendering of those CSVs; nothing is computed server-side, and every number can be traced back to a printed page.

What gets summed

Only the additive leaf rows are ever added up: Tamil Nadu's sub-detailed heads and Karnataka's object heads (the same level under two names). The volumes' own printed subtotals are deliberately not used as data; they serve as an independent check on the leaf sums. Where a printed total and the sum of the rows beneath it disagree, the head is flagged with a chip rather than silently corrected.

One fiscal year, three vintages

Each budget volume prints four money columns about three fiscal years. The portal files every amount under the year it is about:

• Budget estimate (BE) for year Y: from year Y's volume;
• Revised estimate (RE) for Y: from the following volume;
• Actuals for Y: from two volumes later;
• BE restated: the same year's budget as reprinted a year later. It genuinely differs from the original in places, so both are carried and offered side by side.

That is why the measures cover different spans: actuals stop where the accounts stop, and the newest years have only estimates. A dash is a figure not printed yet, never a zero.

Units

Tamil Nadu prints INR thousand, Karnataka INR lakh. Everything on this site is converted to ₹ crore, nominal; no deflation, no per capita scaling.

Error chips

The extraction pipeline tests every printed subtotal against the rows beneath it (check ids W01–W13 in the package READMEs). A grey ± chip marks a small gap; an amber verify against PDF chip marks heads where checks disagree enough that the printed page should be consulted (the chip links to it); a red chip marks a large extraction gap where the PDF should be trusted over the portal.

Extraction quality by year

Caveats

• This is a research preview, not a certified statistical release. The build manifest records the exact input files and any skipped malformed amount values; check it before citing a figure.
• State totals here are the sum of all expenditure heads, including debt servicing and loans; they are not the "budget size" quoted in speeches, which nets differently.
• Negative lines (deduct-recoveries) are summed as printed, so branch totals are net of recoveries.
• Head names are normalised across years for matching; codes, not names, are the stable identity of a head.
• The build reads locally generated *_cleaned.csv files; build_manifest.json in this site's root records the exact inputs behind every deploy.

Licences

Data and code under the repository's licence. India map © DataMeet contributors, CC-BY 4.0, simplified; boundaries follow the Survey of India depiction.