The 4.76 Million Question

In February 2025, 70 constituencies elected Delhi's assembly; the Bharatiya Janata Party (BJP) won 48 seats and the Aam Aadmi Party (AAP) 22, and the Election Commission of India (ECI) published, as it does after every election, a full statistical breakdown of electors, turnout and results down to the constituency level. In 2026, ECI ran a Special Intensive Revision (SIR) of the electoral rolls - a periodic, legally mandated re-verification in which every elector is sent an enumeration form to confirm they still live where the roll says they do. Forms that cannot be delivered, collected, or verified are marked "uncollectable" and removed outright; electors who survive that first cut can still be issued a formal notice to produce further documents, and lose their place on the final roll if they don't (Exhibit 06).

This analysis compares ECI's own 2025 election-day numbers against ECI's own 2026 SIR numbers, constituency by constituency, to see what changed. It draws on the Commission's published statistical reports, its SIR summary tables, its constituency-level uncollectable-reason codes, and its notice-service records - no outside or unofficial data was used. All figures behind this piece are reproducible from these public records; the full working dataset is in the ledger table below.

ECI's 2025 statistical report puts Delhi's electorate at 15,626,736. By the time the 2026 SIR drew its base list, that had already fallen to 14,510,299 - a gap of over a million that predates any SIR deletion at all. The SIR then marked a further 4,756,722 names "uncollectable" (32.8% of its own base), leaving a draft roll of 9,753,577. Statewide, that is barely above the 9,494,607 people ECI's own Report 12 says voted in February 2025 - a citywide buffer of just 259,000 people, or 2.7%, across all of Delhi.

Across all 70 seats, the correlation between a constituency's 2025 winning margin and its 2026 deletion rate is r = −0.32: tighter races, more deletion. Seats decided by under 5% lost a mean of 36.1% of their SIR base; seats decided by 20% or more lost a mean of only 29.4%. The strongest single predictor we found, however, was not margin but 2025 turnout itself (r = −0.65): constituencies where fewer people voted in 2025 saw far higher deletion in 2026 - a pattern equally consistent with a demographic explanation or an administrative one.

The four closest contests in Delhi - Sangam Vihar (344 votes), Trilokpuri (392), Jangpura (675) and Timarpur (1,168) - were decided by a combined 2,579 votes. Those same four constituencies recorded a combined 268,727 "uncollectable" names in the 2026 SIR: 104 times the margin that decided all four races.

Deletion rate and deletion count are two different stories. A constituency can post a large absolute number of deletions simply because it has a large electorate, even if its rate is unremarkable - and a smaller seat with a middling electorate can post a disproportionately high rate. The chart below keeps both in view at once.

That does not, by itself, indicate an error. Some voters will legitimately leave the roll between elections - they die, move, or are found duplicated elsewhere. But it creates a straightforward reconciliation question: what happened to the people who demonstrably voted in 2025?

Ranking all 70 seats by deletion rate, all seven South East Delhi constituencies - Jangpura, Kasturba Nagar, Sangam Vihar, Kalkaji, Tughlakabad, Badarpur and Okhla - land in the top 11 of 70 citywide. The district's mean deletion rate is 43.5%, nearly eleven points above the citywide average of 32.8%, and it holds both the single highest deletion rate in Delhi (Tughlakabad, 47.8%) and the second highest (Okhla, 45.2%).

We tested how likely that clustering is by chance: the probability that seven randomly chosen seats would all land in the top eleven of seventy is about 1 in 3,635,000 (a straightforward combinatorial calculation, not a model). Whatever is driving South East Delhi's deletion rate, it is not statistical noise.

Notably, South East is not a landslide district - its mean 2025 margin was just 7.7%, the second-tightest of any district in the city. Two of its seven seats were among Delhi's four closest races of any kind.

SIR is a two-stage filter. Stage one marks unreachable forms "uncollectable" and removes them from the draft roll - that is everything shown in Exhibits 01-05. Stage two runs on whoever is left: field verification can still flag a surviving elector's entry for a document discrepancy (an age gap with a listed parent, a name mismatch, a duplicate-looking record) and issue a formal notice demanding proof. An elector who does not respond is removed from the final roll when it is published, even though they were never marked "uncollectable" at all.

Delhi's SIR database records 3,311,746 such notices - 34.0% of everyone who survived the initial cull, or 22.8% of the original 2026 SIR base. Add that to the 32.8% already marked uncollectable, and 8,068,468 of Delhi's 14,510,299-name 2026 base - 55.6% - is currently either struck off outright or fighting to stay on. Only 44.4% of the base sits unchallenged right now.

Layering the two mechanisms together sharpens the margin pattern from Exhibit 03, rather than softening it. The correlation between winning margin and total jeopardy (uncollectable plus noticed) is r = −0.43 - stronger than deletion alone (−0.32). Grouped by margin band, the trend is close to a clean staircase:

BJP won 48 of Delhi's 70 seats in 2025, AAP 22 (this analysis excludes no seats, but two-party vote share means a handful of these are close three-way finishes). Splitting the SIR figures by 2025 winner is a direct check on a question the margin analysis alone can't answer: is the deletion pattern tracking how close a race was, or which party actually won it? These are reported as plain distributions, not as a claim that either party benefited.