Forecasting Tamil Nadu’s 234 seats.
A four-model ensemble. Twenty-five years of data. One transparent methodology, published before the count.
Publish the methodology now. Reveal the predictions on May 4 at noon IST. Compare against the result. Either the model called it, or it didn’t.
Election Commission begins counting May 4.
Live now
The full architecture, in plain language
Four models, three backtests, one freeze list. How vote shares map to seats. What we can and can't predict.
Every seat, twenty-five years of history
Click any of the 234 ACs to see how it voted across seven cycles, who lives there, and who's contesting in 2026.
Nine scenarios, ranges of possibility
Vote shares and seat counts under different assumptions about TVK strength and DMK scheme delivery.
The model’s call — live now.
The 4-model ensemble forecast. Confidence per constituency. The full Monte Carlo distribution. Predicted versus actual, seat-by-seat, as the count unfolds.
Why this election is different.
Tamil Nadu 2026 is not a wave election. It is a contest of arithmetic between two close-matched alliances, fragmented by two standalone parties whose vote share is real but whose seat impact is structural — they redistribute votes more than they win them.
The model treats each of the 234 constituencies as its own micro-election: a coefficient assembled from seven cycles of historical voting, modulated by demographic structure, sitting-MLA personal vote, candidate quality, and the specific spoiler effects of TVK, NTK, AIPTMMK and AJPK.
Then those 234 micro-forecasts get aggregated four different ways — top-down, bottom-up structural, bottom-up with mood overlay, and Bayesian hierarchical — to triangulate uncertainty. Where the four models agree, confidence is high. Where they disagree, the seat is genuinely unpredictable.
Read the full methodology → How the model works