One honest picture of how India votes.
Every constituency, every election, both chambers — from the first general election of 1951–52 to today, on maps drawn the way the country was actually divided at the time.
In my own words
I’ve been interested in political and data analysis since I was a teenager. Every election season I read the newspapers closely: the constituencies and their issues, the voting patterns, why people voted the way they did, how the campaigns were run. What always bothered me was how hard it was to see the record of modern India’s own democracy – who won where, by how much, and how that has changed.
The idea for a site like this came to me in 2015, but I didn’t have the skills to build it. Years later, AI changed that: I built this site with Claude, and how that worked is explained further down this page. My first plan was a scenario builder based on the latest results, but I wanted the complete record behind it, so I set out to map every election since India became a republic.
On the tenth day of pursuing this project, I learned that there was no official, correct digital map of India’s constituencies from before the 2008 delimitation. The only correct versions I could find were commercial – licensed to some American universities for their students, but out of reach of an ordinary Indian citizen. Then I found Prof. Sandip Sukhtankar of the University of Virginia, who had hit the same wall years earlier. He was outraged by private companies charging $2000 for correctly mapped files, so he corrected the files himself and shared them for free, with Uttarakhand added by Manasa Patnam. I felt exactly the same that day. Starting from their map, I made my own corrections, then worked backwards through the older delimitation orders to rebuild the maps for the elections before 1977, as closely as the records allow.
Somewhere along the way, it stopped feeling like I was just building a website. It felt like work of national importance. This is the history of how India has chosen its governments since 1951, and a country and its citizens should be able to view the correct maps of all its elections.
I’m not a political scientist, a cartographer or a working programmer. I’m a management graduate who wanted something that didn’t exist: a way to look at any Indian election, in any year, on the map as it actually was. Not a spreadsheet or a static table, not a news-night dashboard that vanishes after the count, and not history flattened onto today’s boundaries.
India has a billion voters and one of the richest electoral records anywhere in the world, and most of it sits in PDFs. The people best placed to read that record – journalists, students, researchers, candidates, the merely curious – mostly can’t, because the tools are paywalled, frozen, or drawn on the wrong map. Billion Ballots is my attempt to change that: to make serious political analysis of India something anyone can do from a browser, for free.
What I brought to it wasn’t code. It was a set of questions, an insistence that every number be traceable to a primary source, and a refusal to trust anything I hadn’t seen rendered on screen.
If you find something wrong, please tell me. Every correction makes the record better for the next person who looks. Ideas are just as welcome, and so is a plain hello, even if it’s only to tell me which seat you looked up first.
– Nikunj Daga
- Follow Billion Ballots
- X (Twitter) @billionballotsInstagram @billionballots
What this is
Billion Ballots is a solo-built, open-data atlas of Indian elections: interactive constituency maps for every Lok Sabha and Vidhan Sabha election, seat profiles, swing analysis between any two elections, party histories, a scenario builder, a written story for every election — and a playable timeline of 73 years of results, 781 assembly by-elections included.
Most election archives flatten history onto today’s map. This one doesn’t: results from 1957 appear under the states, seats and boundaries of 1957. That decision — era-true identity — runs through the whole dataset, from dissolved states like Bombay and Madras to seats renumbered by each delimitation.
Everything is built from the Election Commission of India’s own published record, cross-checked against the delimitation orders in the government gazette.
Every figure here is computed from the data the site ships, not typed — the same numbers the homepage quotes.
Built with Claude
This site was built by one person directing an AI. Nearly every line of code, every parsing script and every audit here was written by Claude, Anthropic’s AI model – first in the Claude app, later in Claude Code. I set the goals, made the judgment calls, reviewed what came back, and refused to ship what I couldn’t verify.
At its best it did what I never could have alone. It turned long government delimitation orders into clean tables, and even spotted what looked like typos in the official orders themselves – flagging them rather than quietly fixing them. When I corrected its work by hand, it checked my corrections against the source and caught my mistakes too.
It also got things wrong, often confidently. It gave me a cleanup command that deleted a folder the homepage depended on. It once placed every Keralam seat on Karnataka’s map – which, to its credit, it caught and reported itself. And it proposed deleting a block of Andhra Pradesh records that turned out to be the pre-2014 history of Telangana’s seats. When I pointed that out, it replied, “your read is right where my last two weren’t.”
It can be cautious to the point of pedantry – it once argued that even freely shared data failed its test for openness, because “even attribution is a precondition”. Its biggest misjudgement was about what was possible. For weeks it insisted the pre-2008 maps would have to be traced by hand – “There’s no Option C hiding here.” I refused to accept that, and with that push, and a newer model, it realised the answer was already in a file it had set aside and built the whole map layer itself. Then it admitted, “I was over-cautious earlier.” The same persistence recovered Surat North, the one Gujarat seat missing from every national dataset, from the Surat Municipal Corporation’s own mapping server.
Newer models kept improving on older work, going back over it and finding mistakes nobody had noticed – maps from the wrong period joined to the right seat numbers, a state missing most of its ground. That is why the real defence here isn’t my expertise or the model’s – it’s the automated checks. Before every release, independent audits run over the whole dataset – does each boundary’s own name agree with the result joined to it? – and the release is blocked if any of them finds something new. Where a check can’t be written, we render the page and look at it. In the story card for each election, every number is computed from the data, never typed. We say this plainly because people should know how the thing they’re reading was made.
Era-true identities
Every seat lives under the state, name and number of its own year. Madras stays Madras until 1969; Bihar keeps Jharkhand's seats until 2000. Nothing is back-projected onto modern lines.
Boundaries have a vintage
A constituency map is only right for the delimitation it was drawn under. Every year is served geometry of the correct vintage, and every layer's confidence grade is disclosed.
Flag, don't corrupt
Anomalies in the record — uncontested returns, deferred polls, disputed entries — are kept and labelled with provenance, never silently patched to look tidy.
How it’s made
Ingest the primary record
Election Commission of India statistical reports — every general and assembly election, plus bye-elections back to 1952 — parsed, cross-checked and loaded into a relational store.
Rebuild identities era-true
Results are keyed to the state and seat numbering of their own era, verified against the delimitation orders themselves — including reservation status, multi-member seats and renames.
Match geometry to vintage
Each of the six delimitation eras gets its own boundary layer, joined by stable ids and audited by an independent name-agreement oracle after every change.
Serve it static
Everything you see is pre-computed JSON and TopoJSON on a CDN — no database on any page load, which is why the whole atlas is fast and shareable by URL.
A gate that blocks the release
Before any change reaches the site, independent audits run over the whole dataset, and the release is blocked if any of them finds a problem that is not already logged. The rule is “zero new”, never “zero”: known debt is listed openly and retired over time, not hidden behind a green light.
A name oracle the map never uses
Maps join results to boundaries by seat id only. Each boundary's own name is therefore free to act as an independent witness: after every boundary change, an audit checks that the name on the ground agrees with the name in the record for that id. Low agreement means a file of the wrong vintage — the defect that clicking around a map can never reveal.
Render it, then look
Where a check cannot be written, a real browser renders the page and a person reads the screenshot. A map can pass every data test and still paint one seat over the whole country.
Flag, don't corrupt
Anomalies — uncontested returns, deferred polls, disputed scans of a sixty-year-old order — are kept and labelled with their provenance rather than silently tidied away.
Every general election since 1951 has a map of its own vintage, including the 13 states that no longer exist. The same grades appear on every map and every seat-history drawer across the site, so you always know how much to trust the ground under a result.
Election results come from primary Government of India sources. Boundary geometry is assembled from the openly published datasets below, each used for the delimitation era it is correct for. The place behind each seat — its people, economy and heritage — comes from the public statistical and heritage registers listed last. Each entry names what it is used for and the terms it is used under.
- Election Commission of India — Statistical Reports — eci.gov.in; every general and assembly election since 1951 and bye-elections since 1952 — the source of every result, candidate, vote count, elector count and turnout on this site (© Election Commission of India, reproduced with acknowledgement)
- ECI results portal — results.eci.gov.in; 2019–2024 results and current constituency names (© Election Commission of India)
- OpenCity CKAN — data.opencity.in; machine-readable mirror of the ECI statistical reports (Public Domain)
- OpenCity — assembly-segment results of the Lok Sabha elections, 2019 and 2024 — data.opencity.in; the Election Commission’s votes for each assembly segment of each parliamentary seat, republished by OpenCity and used only for “if the 2024 votes were counted on these seats” on the Delimitation seat simulator (2024: License: Other (Public Domain); 2019: Public Domain, ECI results republished by OpenCity)
- Delimitation Orders, 1951 · 1956 · 1962 · 1967 · 1976 · 2008 — Gazette of India notifications as published by the Election Commission of India; seat identity, numbering and reservation status for each era (Government of India)
- MyNeta, Association for Democratic Reforms — myneta.info; candidate affidavits from 2004 onward — assets, liabilities, criminal cases, education. Every candidate panel links to MyNeta’s own page for that candidate; no affidavit PDF is republished here (public affidavits filed with the ECI, compiled by ADR — credited on every use)
- ECI affidavit portal — affidavit.eci.gov.in; the original affidavits (© Election Commission of India)
- Election Commission of India — Atlas of the 2024 General Election, Annexure III — eci.gov.in; electors per state and territory, used by the Delimitation seat simulator (© Election Commission of India, reproduced with acknowledgement)
- Constituency boundaries — NIC Mapserver via @planemad on GitHub (Public Domain)
- Original boundary digitisation — DataMeet community (CC-BY-SA 2.5 India)
- Assembly boundary detail — datta07/INDIAN-SHAPEFILES (MIT)
- Pre-2008 assembly boundaries — Sandip Sukhtankar, AC_All_Final (April 2012), corrected from Election Commission of India shapefiles; Uttarakhand data added by Manasa Patnam — used for the 1976-delimitation layers and for Goa and Delhi before the 2008 redraw (reproduced with the author’s written permission, September 2026)
- Mumbai, Thane and Palghar boundaries — © OpenStreetMap contributors, ODbL: place and taluka positions used to place Lok Sabha seat boundaries for Mumbai, Thane and the Palghar belt (and as test points for Hyderabad and Pune)
- Madhya Pradesh assembly boundaries — NIC BharatMaps, National Informatics Centre, Government of India (base data Survey of India 1:50,000)
- Gujarat assembly boundaries — Local Government Directory constituency layer, Government of India, via ramSeraph/indian_admin_boundaries (CC0 1.0); Ahmedabad’s city seats rebuilt, as the 2008 Delimitation Order defines them, from the ward map in Ahmedabad Municipal Corporation’s City Development Plan 2006–2012 and the Census of India 2001 District Census Handbook, Ahmadabad
- Surat assembly boundaries — Surat Municipal Corporation GIS, assembly constituency layer (Surat Municipal Corporation, Government of Gujarat)
- Historical princely-state boundaries — India State Stories, FLAME Center for Legislative Education and Research, FLAME University (1941–2011 decadal digitised shapefiles, adapted here; licensed under CC BY 4.0 with written permission, September 2026)
- Jammu & Kashmir 2022 boundaries — shijithpk/2024_maps_supplement (Unlicense)
- Delimitation seat simulator: simulated seats — drawn on this site by grouping today’s assembly seats from the layers credited above: an adaptation of those layers, simulated, not a boundary that any authority has drawn
- Census of India 2011 — censusindia.gov.in; Primary Census Abstract and House-listing tables at district level — population, literacy, religion, households, amenities (Government of India)
- Registrar General of India — Population Projections for India and States 2011–2036 — Report of the Technical Group of the National Commission on Population (July 2020), Table 21: population on 1 March 2026 and 2031, used for the seats each state would have by population on the Delimitation seat simulator (Government of India)
- NFHS-5 (2019–21) district factsheets — via data.gov.in; health and household indicators per district (Open Government Data Licence – India)
- NITI Aayog — Aspirational Districts — niti.gov.in; the 112-district list (2023) (Government of India)
- Reserve Bank of India — Handbook of Statistics on Indian States 2024-25 — rbi.org.in; about 320 state-level indicators across 15 domains, on the state pages and seat profiles (Reserve Bank of India)
- National Crime Records Bureau — Crime in India 2024 — ncrb.gov.in; state-level figures, shown with NCRB’s own caveat that they count cases registered, not incidence (Government of India)
- Archaeological Survey of India — asi.nic.in, via data.gov.in; the list of centrally protected monuments (Government of India)
- Wikidata — wikidata.org; heritage sites and landmarks — coordinates and images. The data is CC0; each image carries its own Wikimedia Commons licence, shown beside it
- One District One Product (DPIIT) — odop.in; district specialities (Government of India)
- Geographical Indications Registry, IP India — ipindia.gov.in; registered GI products and their districts (Government of India)
- OpenStreetMap — “open in map” links, and the place and taluka positions credited under Constituency boundaries above (© OpenStreetMap contributors, ODbL)
Party symbols and flags on this site are original illustrations we drew ourselves, not the parties’ or the Election Commission’s own artwork. Whether Indian law actually permits a third party reproducing an official symbol’s specific artwork — as opposed to the generic idea of what it depicts — is a genuinely open question with no court ruling on point either way, so rather than resolve that ourselves we draw a likeness of each symbol from its public description or published form. If a party would like to confirm or provide its own artwork, we’d welcome hearing from them.
How the Development pages work
Every figure on a seat’s development page (the ones “Look up any seat” opens) is a district statistic — Census 2011 for population, religion and homes, NFHS-5 (2019–21) for health — shown for the district the seat sits in. No government agency publishes these figures by constituency, so a seat that spans more than one district, or shares a district with other seats, shows its main district’s figures, not figures unique to it. Each of those pages names that district and lists any other seats that share it.
The state records and the by-subject pages are different: they describe each whole state, from the Reserve Bank of India and the National Crime Records Bureau.
Most subjects put every state side by side, for the year most states report. Those with at least 25 states reporting that year get a map and a ranking. Some get a map but no ranking: on subjects like poverty, deaths and crime a league table says little about why places differ, and on a parity index 1 is the aim, not one end of a scale. Those with too few states reporting the same year get a table. And some are totals, not rates, so states aren’t compared on them at all — a bigger state simply has a bigger total.
How the seat rating works
Every seat page rates the seat Safe, Lean, Competitive or Toss-up from its winning margin: the winner’s lead over the runner-up, as a share of the votes. Safe: a margin of 15% or more. Lean: 8% up to 15%. Competitive: 3% up to 8%. Toss-up: under 3%.
Each seat is rated twice, on its latest margin and on its average over the last three elections; the lower rating is shown. A seat with two contested elections is rated on its latest margin and on the average of the two; a seat with one is rated on that margin alone. By-elections and unopposed returns are not counted. A seat now rated Competitive or Toss-up that was rated Safe before its latest election, or whose record of six or more elections averages Safe, says so on its page.
The rating describes past results only, not a forecast.
“Parties in the race” is the effective number of parties: one divided by the sum of each party’s squared share of the vote (NOTA counted), so two parties with half the vote each read 2.0 and a crowded field reads higher.
Known limitations
- Candidate affidavit details exist only from 2004, when affidavits became mandatory; earlier candidates carry name, party and votes only.
- Multi-member seats (1951, 1957 and a few later cases) list every member; margins are computed against the true runner-up.
- Deferred and countermanded polls are recorded separately from the main election they belong to.
- Party names are shown as the Election Commission recorded them for that election; mergers, splits and renames are not projected backwards.
- A handful of state-years with no usable boundary source render as a results table, with the reason stated on the page.
- The forecast tool applies swings you choose to the last result.
What this site is not
Billion Ballots is not affiliated with, endorsed by or connected to the Election Commission of India, any government body, or any political party or candidate.
The party symbols shown are our own drawings, not official artwork. The Election Commission’s own publications remain the authoritative record; where this site and the Commission disagree, the Commission is right and we would like to hear about it.
Corrections & citation
Found an error?
Write to hello@billionballots.com. The page URL is enough — every seat has a permanent one — plus what you think is wrong; a source helps.
Citing this site
Billion Ballots, “Varanasi — Lok Sabha 2024”, billionballots.com/lok-sabha/2024/UP-77/, accessed <date>.
Reusing the data
Results come from public government sources and may be used with acknowledgement to the Election Commission of India. The boundary maps combine sources under different licences (listed above), some of which require credit or sharing changes under the same terms, so there are no bulk downloads yet — ask before reusing the maps. The analysis built on the results — swing, seat lineage, ratings, forecasts and story cards — is this site’s own work: quote it with credit, but ask before copying it in bulk.
A peacock whose fanned tail is a seat chart: India’s national bird, every feather a block of seats. When the tail opens — on the homepage, or when you hover any mark on this page or in the header — it plays a results night: seats declared one by one, blocs forming, the largest bloc taking the house, then everything settling back to the brand.
Built to be read, corrected and reused.
Fraunces · Libre Franklin · IBM Plex Mono — Next.js, D3 and a very large pile of Election Commission PDFs.
Built with Claude Code