Karnataka budget data guide

Introduction to the documents

Every year the Government of Karnataka publishes its Detailed Estimates of Expenditure as a set of seven volumes (EXPVOL1–EXPVOL7) — the per-department backing detail behind the Demand for Grants. Each volume is a long PDF of accounting tables: for every department (a “demand”), money is broken down through a fixed hierarchy of account heads, with four money columns — prior-year Accounts, current Budget Estimate, current Revised Estimate, and next-year Budget Estimate.

These PDFs are built for printing, not analysis. This package extracts them to CSV and JSON while keeping the document’s shape, so any number can be traced back to the page it came from.

The CSV is document-shaped by design. It keeps headers, totals, and summary rows so you can line the table back up against the PDF — and that same shape is the one thing you must respect when you total a column (see Don’t double-count).

A snapshot of the budget document

Here is one real page and the rows we extracted from it. Every printed line becomes a row; the extractor also records what kind of line it is.

The table below is the machine-readable version of that page (the current-year Budget Estimate column shown):

Sum?	row_type	row_level	row_header	object_ head_code	BE 2024-25 (₹ lakh)
×	Header	Detailed-Head	Commissioner for Excise and Other Establishments
✓	Data	Object-Head	Pay-Officers	002	1,733.00
✓	Data	Object-Head	Pay-Staff	003	13,001.00
✓	Data	Object-Head	Interim Relief	004	2,501.00
✓	Data	Object-Head	Dearness Allowance	011	7,312.00
✓	… 13 more Object-Head leaf rows (all visible on the page image) …				6,788.31
×	Total	Detailed-Head	HOA Total (Head of Account)		31,335.31

Tinted rows are the additive leaves (row_type = Data and row_level = Object-Head) — the only rows you add up. The 17 leaf rows sum to 31,335.31. The grey Total row repeats that same 31,335.31: add up everything and you count it twice.

The columns you’ll use first

You don’t need all 30-odd columns to start. These few answer most first questions. (The complete reference is at the end of this guide.)

The accounting ladder

The DEA Budget Manual describes six tiers in the Detailed Demands for Grants. In this package they appear as code/name columns from Major Head down to Object Head.

Don’t double-count

You just saw it: the object-head leaf rows under “Commissioner for Excise” sum to 31,335.31, and the HOA Total row says 31,335.31 again. Do not sum the whole table. Totals and summary rows repeat amounts already present in leaf rows.

To total safely, keep only the additive leaf rows — all three conditions are required:

WHERE type_of_table = 'object_head'
  AND row_type      = 'Data'
  AND row_level     = 'Object-Head'

About 37% of rows are headers, totals, or summary rows — all excluded by this predicate. The examples/ tidy table (karnataka_budget_tidy.csv) has this filter already applied, so you can sum it directly with no predicate.

Voted and charged

vote_charge_marker captures the source distinction where it is printed. Charged expenditure is not submitted for vote under the Constitution; voted expenditure is subject to legislative vote.

Validation is an internal consistency signal, not a guarantee that every PDF character was extracted correctly.

The three measures: plan, revision, actual

Each year carries three measures: BE (Budget Estimate, the plan), RE (Revised Estimate, the mid-year correction), and Actuals (what was actually spent). The same fiscal year is reported across several yearly documents; the package’s canonical source rule takes each measure from one document — BE from the document’s own year, RE from the next year’s document, Actuals from the year-after’s document — so a value is never counted from two sources. The examples/ tidy table applies this rule for you.

Validation drilldown

Example key:

KA.2021-22.expvol_1.in_schema.object_head.minor.03_2043_00_101

In checks_2021-22.csv, this key has four financial-column comparisons. Two pass exactly; two differ by INR lakh 1 and fail the 0.01 threshold. A 0.01 difference is one-hundredth of the package amount unit; if INR_lakh is correct, that is Rs 1,000.

Files to use

Full column reference ↑ top

The complete budget table columns, grouped. Controlled values are shown as code chips. The machine-readable source, including the checks and validation_summary tables, is karnataka-state-finance/data_dictionary.csv.