About & Data
An analytical layer over election data — not a newsroom, not a portal.
ElectionLens India turns questions about Indian elections into structured analysis: comparisons, swings, close races, and hypothetical alliances. The interface is built for every state and for Lok Sabha; this early build is scoped to the Andhra Pradesh Assembly elections of 2019 and 2024 so the experience can be judged end to end.
Principles
- Official election data is the source of truth for every number.
- A language model never invents seat counts, vote totals, vote shares or margins.
- Actual results and simulations are always visually distinct.
- Every scenario exposes its assumptions and its per-seat arithmetic.
- A simulation is not a prediction, and is never presented as one.
- The platform stays politically neutral; party colours appear only in data visualisation.
What’s real
- 414 real elections, 1951–2026 — 18 Lok Sabha general elections and 396 state Assembly elections across 31 states, all mirrored from ECI results. The Assembly backbone is the TCPD / Lok Dhaba all-states export (Ashoka University); DataMeet's india-election-data covers 1951–2009; recent polls carry turnout and electors. Every dataset is tagged "community data" with its source shown on the page.
- Andhra Pradesh 2014, 2019 and 2024 are real: 2014/2019 from Lok Dhaba, and 2024 from a compiled ECI results workbook (winner + runner-up per seat, so the field below the top two is not listed). The homepage story and AskVoter's answers run on this data.
- Explore any of them: house composition, vote share, seat change, two-party swings, closest races, constituency grid — with the previous election of the same house for comparison.
- Scenario Lab on any election: re-run it with a different alliance, or project a future one from the latest result plus an assumed vote swing. Alliance choices are the strongest parties of that election.
- The deterministic scenario engine (anchor + per-party vote transfer), shared by the Lab and by AskVoter.
- Responsive editorial design system, light and dark.
What’s partial
- Andhra Pradesh 2024 lists only the winner and runner-up per seat — the workbook doesn't carry the rest of the field, so third-place-and-below shares are absent (not estimated).
- Candidate names are thin in the older imports (the source often carries only party + votes). A few seats deferred by the ECI (Patkura 2019, Ramgarh 2018) are missing from the Lok Dhaba export.
- The constituency map is a space-filling grid, not a geographic choropleth, and 16 Andhra Pradesh shapes don't name-match the dataset.
No figure shown anywhere in this build is an official Election Commission of India result. Replacing data/elections/ap.ts with an ingested dataset of the same shape is the entire migration.
Next
- Match constituencies across delimitation properly (2008 boundary redraw) instead of by name
- Fill the deferred-poll seats and thicken candidate coverage in the pre-2010 imports
- Replace the grid with a GeoJSON / MapLibre choropleth
- Swap the local AskVoter classifier for a real query planner that emits the same AskAnswer contract