TN 2026Election Analyser

Data-only findings. No narrative recap.

AC 185 decided by ONE vote

TVK 83,375 vs DMK 83,374. The closest race in TN 2026. 27 of 234 ACs decided by under 2,000 votes.

TVK lost postal votes by 10.6 pp

TVK won 35% of EVM votes but only 24% of postal. Institutional / elderly / armed-forces electorate skewed DMK — quantified.

🎢

AC 56 had 7 lead changes

121 ACs decided from round 1. 18 ACs saw 4+ lead changes. AC 56 saw the lead shift seven times across 23 rounds.

🏆

TVK 46% win rate (108 of 233)

Highest of any party. DMK 33%, AIADMK 27%, BJP 3% (1 of 33). NTK contested all 234 and won zero.

🌊

Third-front voters → TVK, r = +0.21

ACs with high MNM/AMMK/DMDK 2021 vote share saw higher TVK 2026 share. Real signal — Vijay inherited the Kamal Haasan / Vijayakanth / Dhinakaran pool.

👑

Stalin never led a single round in Kolathur

Sitting CM lost his own seat. Round-by-round shows Babu (TVK) ahead from round 1. The postal-vote effect just barely shows up.

🗺️

Where TVK lost — by caste belt

TVK won 0 of 6 in Kanniyakumari (47% Christian) and 0 in Cauvery delta + Vanniyar belt. The Christian-vote-for-TVK hypothesis fails. Caste-bloc holdouts explain the loss map.

💯

110+ specific findings (index)

One-page index of every specific number we found — closest races, blowouts, flip matrix, reserved-seat winners, vote efficiency, drama. Each line cites the parquet + code.

What's different about this site

Most TN 2026 commentary you'll find online tells you what happened: TVK won, DMK lost, here's the seat tally. This site digs into what the 234 ACs × ~18 candidates × 22 counting rounds of raw data show that headlines miss.

Six findings the news cycle didn't surface:

AC 185 was decided by 1 vote. A 1-vote margin in a multi-million-vote election. 27 ACs total were decided by fewer than 2,000 votes — 11.5% of the state. →
TVK underperformed by 10.6 pp on postal votes. The institutional/elderly electorate preferred DMK significantly. Quantifies the "youth wave" thesis. →
AC 56 had 7 lead changes during counting. 18 ACs were genuine nail-biters with 4+ lead changes. →
NTK contested all 234 ACs and won zero. TVK contested 233 and won 108. BJP contested 33 and won 1. Conversion rates vary 100× across parties. →
Third-front voters (MNM/AMMK/DMDK) really did move to TVK. AC-level correlation of +0.21 — small but real, identifies one mechanism for TVK's coalition-building. →
Stalin never led a single round in his own seat. Babu was ahead from round 1. Kolathur was a clean wire-to-wire loss. →

And the things we got wrong first

The first version of this site (commit 424c76c) had a confidently-asserted finding ("BJP gained 2× more in reserved seats") that was a methodology artefact, not a real signal. We retract it and explain how it happened. Full retraction log →

How to read this site

Findings drill into specific patterns from the data — each cites the parquet file and the computation.
Method & data shows the pipeline, sources, schema.
All numbers reproduce from pipelines/path_a_build.py + pipelines/deep_dive.py against the S3 parquets — they're listed at the bottom of each finding page.

What this site can't tell you

Booth-level patterns within ACs (Path B not run; image PDFs need OCR)
Caste at jati level (no open dataset)
Individual-voter cohort breakdowns (no demographic data)
Causation — only correlation across 234 data points

The honest gaps are documented at findings/limits.