Finding — DMK bled almost uniformly; their "held" seats were only 3 pp better defended

DMK held 49 of the 163 ACs they contested both years. In held seats, DMK's swing was −12.6 pp. In lost seats, −15.9 pp. The DMK had no real strongholds — they bled essentially uniformly across the state.

The numbers

	DMK held (n=49)	DMK lost (n=114)	Gap
Median 2021 vote share	49.78%	46.12%	+3.66 pp
Median 2026 vote share	36.49%	30.05%	+6.44 pp
Median TRUE swing (2026−2021)	−12.6	−15.9	3.3 pp

DMK won 49 of 163 seats where they contested both years (30% retention). Where they kept the seat, the 2021 share was 3.7 pp higher and the swing was 3.3 pp gentler. That's it. DMK had no real fortress effect.

Why "no fortress" matters

You'd expect a major incumbent party to have:

• High-margin strongholds that survive any wave because the base is overwhelming.
• Marginal seats that flip easily.
• A clear bimodal pattern in performance.

In 2026 TN, DMK shows almost no such structure. The seats they retained were modestly stronger in 2021 (~50% vs ~46%) and faced a slightly smaller swing (−12.6 vs −15.9). The difference is real but small.

The implication: DMK didn't have a base that was structurally insulated from the TVK wave. Every DMK seat was within ~3 pp of being competitive. There was no "no matter what happens, we'll keep these 30 seats" floor.

Where DMK's 49 surviving seats are

Of the 59 DMK 2026 wins (some from new ACs they hadn't contested in 2021, so n=49 in the both-years analysis):

• Concentrated in southern TN — Madurai, Trichy, Kanniyakumari districts.
• Some Cauvery delta — Thanjavur, Nagapattinam pockets.
• Reserved seats — DMK won 9 of 44 SC/ST reserved.

The geographic concentration matters because it tells us where TVK's wave was weakest — rural + southern + Cauvery delta. Consistent with the urban-rural divide finding.

The harder question — what kept DMK's 49 alive?

We can hypothesise — but the data alone can't fully answer:

• Strong incumbent candidates with personal brand (e.g., Udhayanidhi Stalin held Chepauk-Triplicane with a comfortable margin).
• Lower TVK penetration in older, more rural ACs (matches the postal-EVM gap finding).
• Alliance allies' efforts — DMK rolled out CPI/VCK/IUML candidates in some of these areas, which may have helped overall front share even if DMK proper bled.

Without booth-level or candidate-quality data, we can't fully decompose. The cleanest claim from this data alone: DMK's "strongholds" weren't structurally different from their losses — they just had slightly less wave hit them.

DMK's losing seats — by how much they fell

In the 114 ACs DMK lost (both years contested):

• Median 2021: 46.12%, Median 2026: 30.05% — drop of 16 pp
• Worst drop (top 5 losing): Thiruvottiyur (−44 pp), some Salem/Erode ACs (−38 pp), several Chennai metro losses.
• Lightest drop in losses: a handful of close-margin losses where DMK fell to ~38% from ~46%.

What we'd need to go deeper

• Per-candidate seniority (years in DMK, prior MLA terms) — does incumbency help?
• Welfare-delivery data — were "delivered" seats less likely to swing?
• Alliance partner co-contest — did the presence of CPI/VCK in nearby seats help?
• Booth-level vote variance within DMK-held vs lost ACs — were holds uniform high or pockets of strength?

None of these are in the current open data we've ingested.

Cross-check

• DMK held vs lost via inner-join of 2021 + 2026 results filtered to DMK + winner table.
• See pipelines/deep_dive.py (dmk_held_vs_lost block).