"Detect, delete, deport." West Bengal, 2026. 91 lakh names removed from the electoral roll. BJP wins by 32 lakh votes. The Election Commission calls it routine. The Bureau calls it arithmetic. The Bureau has been calling things arithmetic for some time now.
The Bureau begins with the denominator. In electoral arithmetic, the denominator is who gets to be counted. Change the denominator, and the percentage changes. Change enough denominators, and the result changes. The Bureau will document the denominator.
West Bengal held its 2026 Legislative Assembly election in two phases: 23 April and 29 April 2026. Results were declared 4 May 2026. In the months before polling, the Election Commission of India conducted a Special Intensive Revision (SIR) of the electoral rolls. The SIR removed 91,02,577 names — 91 lakh, 9.1 million — from the rolls. The pre-SIR roll: 76,637,529 registered voters. The post-SIR roll: 68,125,496. Net reduction: approximately 12% of the electorate. The Bureau records the numerator and the denominator separately, because they tell different stories.
Pre-SIR voter roll: 76,637,529. Post-SIR registered voters: 68,125,496. Net reduction: ~12%. Names removed: 9,102,577 (91 lakh). Appeals filed against deletions: over 34 lakh (3.4 million). Total votes cast: over 63 lakh across two phases. Official turnout: 92.47% (of post-SIR roll). Turnout as percentage of pre-SIR roll: ~82.2% — consistent with West Bengal's historical average. BJP-led coalition result: 207 seats out of 294. Vote margin between BJP coalition and TMC coalition: approximately 32 lakh votes. The Bureau records the 32 lakh margin and the 34 lakh appeals in the same sentence.
Observation 1 · The mechanism: what "logical discrepancy" means
The Bureau does not assign intent. The Bureau assigns mechanism. The mechanism here has a name: "logical discrepancy." The Bureau will document what that name covers.
- 01 The "logical discrepancy" category was invented for this SIR. It did not exist in the Bihar SIR (Phase I, June–September 2025), the Gujarat SIR, or the Uttar Pradesh SIR. It was created specifically for West Bengal. The ERONET software — developed originally by C-DAC in 2018, operated by TCS since 2023 — was programmed to flag voters when their name appeared differently across documents. Examples from the record: "Maqbul Sheikh" vs. "Shekh Mokbul." "Anil Mukherjee" vs. "Anil Mukhopadhyay" (the same name in Bengali and its Anglicised transliteration). "Shikha Sen" vs. "Shikha Dasgupta" (post-marriage surname change). Multiple siblings sharing a father's name. The system flagged 1.36 crore (13.6 million) voters for "logical discrepancies" at its peak. The Bureau notes that the Bengali language has multiple valid transliterations into English. The software did not.
- 02 The ECI told the Supreme Court that each notice was individually reviewed. ECI's statement to the Supreme Court: "There is no automated generation of notices. EROs are signing them off after application of mind." The Reporters' Collective documented the operational reality: over 1.5 crore notices were generated by centralised algorithm. EROs (Electoral Registration Officers) signed off 4,000 to 5,000 notices daily. A senior ECI official acknowledged: "It is a near impossible exercise to selectively discard notices, given the sheer volumes." Another official described the process: "ERO is opening ERONET, seeing that these number of notices on logical discrepancies have been generated. So he will give the command of printing that part's notices." The Bureau records the ECI's Supreme Court statement and the senior ECI official's operational description in the same observation.
- 03 The procedural record. Instructions were issued via WhatsApp groups and oral commands during video conferences, not in written orders. The original revision manual was abandoned in favour of a generic overarching order. The ERONET system prevented uploading documents outside 12 specified categories, while officials required additional document types. Software features were toggled on and off overnight without notice; a "verification button" disappeared from ERO interfaces without explanation. A District Magistrate wrote formally to the ECI requesting: "An SOP may kindly be provided of logical discrepancy cases and the role of BLO, ERO/AERO and DEO." The SOP was not provided before the rolls were published. One data entry operator: "We don't get written checklists. Instructions are updated in meetings, and my colleagues and I discuss and make our script or standard operating procedure, which I write in my notes."
Observation 2 · The demographic distribution
The Bureau records the community-wise breakdown of the 90,62,188 deletions as documented by The Week and Alt News, using ECI data obtained under RTI and through constituency-level analysis.
- 01 Overall deletions: 34% Muslim, 63% Hindu. Muslims constitute approximately 27% of West Bengal's population (2011 Census). Their share of total deletions: 34.32% — seven percentage points above their population share. The Bureau records the gap.
- 02 The adjudication sub-category: the gap widens to 65.5%. The 91 lakh deletions comprised two sub-categories. The first: 58.2 lakh flagged as "Absent, Shifted, Duplicate, or Dead" (ASDD). The second: 26.95 lakh placed in the "Under Adjudication" / "Logical Discrepancy" category requiring tribunal hearings — the category invented for this SIR. Within the ASDD category: Muslim share was 22.9%. Within the Adjudication category: Muslim share was 65.5%. The Bureau notes: the category where the barrier to restoration was highest — requiring a tribunal hearing, in a process where 19 judges deciding 100 appeals daily would require over 526 working days to clear 10 lakh pending cases — was the category where Muslim voters were most concentrated.
- 03 Constituency-level: the ratio in six measured constituencies. Alt News obtained and analysed data from six constituencies. Across those six: Muslims comprised 51.7% of the combined electorate. Muslims comprised 92.6% of the voters placed under adjudication. The Muslim adjudication rate: 42.2%. The Hindu adjudication rate: 3.5%. Ratio: 12:1. Individual constituency data: Samserganj (82% Muslim electorate) — 98.8% of adjudicated voters were Muslim. Mothabari (69.5% Muslim) — 97.4% Muslim adjudication. Manikchak (49.4% Muslim) — 97.4% Muslim adjudication. The Bureau notes that 49.4% of the electorate produced 97.4% of the adjudication cases.
- 04 Women and other affected groups. Women face structural disadvantage in the process: shifting residences after marriage creates surname-change documentation gaps that the algorithm reads as "logical discrepancy." In SC-reserved constituencies, women comprised 52.4% of adjudication cases. In ST-reserved constituencies, 53.4%. The Matua community — a Hindu refugee community historically associated with BJP support — was also affected, particularly in North 24 Parganas district, where over 12 lakh names were removed across eight Matua-majority constituencies. A past BJP campaigner from the region: "Everyone whose name has been deleted in this part of the state is a Matua Namasudra. These are people who have been voting for decades."
Observation 3 · The election arithmetic
- 01 The record turnout, and what produced it. Official reported turnout: 92.47% — described as the highest ever recorded for a state or national election in India. The Bureau records the arithmetic of this record. Pre-SIR roll: 7.66 crore. Post-SIR roll: 6.81 crore. Votes cast: over 6.3 crore. Turnout as percentage of post-SIR roll: 92.47%. Turnout as percentage of pre-SIR roll (the roll that would have existed without the deletions): approximately 82.2% — consistent with West Bengal's historical average of 80–84%. The "record" exists in the denominator.
- 02 The 31 seats. After the results were declared, TMC's senior advocate placed before the Supreme Court (11 May 2026) the following submission: in 31 constituencies, the number of voters removed during the SIR adjudication process was higher than the margin by which BJP defeated TMC candidates. Specific example placed on record: one constituency, TMC candidate lost by 862 votes; over 5,000 names deleted. Two further examples from published data: Satgachhia — BJP margin 401 votes; deletions 26,000+. Rajarhat New Town — BJP margin 316 votes; deletions 63,000+. The Bureau records these as figures placed before the Supreme Court of India, in an ongoing proceeding. The Bureau does not characterise the outcome. The Bureau records the margin and the deletion count in the same sentence.
- 03 The tribunal backlog that will clear in 526 working days. Over 34 lakh voters filed appeals against their deletion. The appellate tribunals, as of May 2026, have a backlog that — at 19 judges processing 100 appeals per working day — would require over 526 working days to clear: more than two years. Senior Advocate Menaka Guruswamy, before the Supreme Court on 11 May 2026: appellate tribunals "could take several years to clear the backlog." The Supreme Court directed the Chief Justice of the Calcutta High Court to submit a report on pending appeals. The election was held on 23 and 29 April 2026. The appeals will be cleared in approximately 2028.
Observation 4 · The post-election consequence
- 01 Welfare benefits: the deletion extends beyond the roll. West Bengal BJP's Minister Agnimitra Paul announced that deleted voters would not receive the Lakshmir Bhandar benefit — a direct cash transfer scheme — from 1 June 2026. In Bihar, the Chief Minister stated that deleted voters "will not be entitled to any government benefits, including ration" and that bank passbooks would be cancelled. Former IAS officer and Rajya Sabha MP Jawhar Sircar: "You cannot deprive a person of welfare, livelihood, and state-sponsored benefits based on 'logical discrepancy.'" He also noted: "If the ECI is saying SIR is not to decide citizenship, how can state-level leaders say it is?" The Bureau records that the process was described as a voter roll revision. The Bureau records what the roll revision produces downstream.
- 02 The Assam precedent, for longitudinal reference. Assam's D-voter ("Doubtful voter") mechanism, introduced in 1997, created a category of voters placed under suspicion. As of 2019: 113,738 D-voters remained in the system, 70,723 of them women. The case of Dulubi Bibi: declared a foreigner in 2017 on the basis of a name variation (Dulubi Bibi vs. Dulabjan Begum), reinstated as a citizen six years later by the same tribunal. The West Bengal "logical discrepancy" category is structurally parallel to Assam's D-voter mechanism, applied at scale. The new Immigration and Foreigners Order, 2025 (rules 16–20) enables the Foreigners Tribunal model to be extended nationwide. The Bureau notes this is the longitudinal trajectory.
The SIR removed 91 lakh names. The BJP won by 32 lakh votes. Appeals exceeded 34 lakh. In 31 constituencies, the deletion count exceeded the BJP margin of victory. The official turnout of 92.47% is arithmetically produced by a 12% reduction in the denominator: the same votes cast would be 82% turnout on the pre-SIR roll — consistent with historical norms. The "logical discrepancy" category, invented specifically for West Bengal, placed Muslims at 65.5% of the hardest-to-restore deletion sub-category despite their 27% population share. The Supreme Court Justice who presided over the pre-election hearing asked: "Why this agitation in Bengal?" The same Justice, the Bureau notes, had compared unemployed youth to cockroaches and parasites eight days earlier, which is why the Cockroach Janta Party exists, which is why the ECWP is filing a despatch about it. The Bureau names behaviours, not names. The behaviour documented here is: reduce the denominator; announce a record turnout; proceed.
Habitat: electoral rolls subject to Special Intensive Revision; ERONET software interpreting transliteration variation as identity fraud; tribunal backlogs measurable in years; constituency margins smaller than deletion counts.
Sting mechanism: Stage one — the pre-envenomation audit: the roll is revised; categories are created; the algorithm flags. Stage two — the adjudication bottleneck: tribunal capacity is set below the volume of appeals; the election is held before clearance; "mere pendency of appeals shall not entitle them to exercise their right to vote" (Supreme Court, 13 April 2026). Stage three — the denominator adjustment: 91 lakh removed; 92.47% turnout announced; the record belongs to the reduced roll. Stage four — the downstream extension: welfare benefits withheld; bank passbooks cancelled; the deletion follows the person past election day into their access to the state.
Diagnostic indicator: 31 constituencies where BJP margin of victory was smaller than deletion count. The Bureau notes that this is TMC's submission before the Supreme Court, not the Bureau's. The Bureau records it as a figure placed on the record of a court proceeding.
Conservation status: expanding. Phase I covered Bihar. Phase II covered 9 states. The Immigration and Foreigners Order, 2025, extends the Foreigners Tribunal model nationally. The specimen is scaling.
The larva is doing well. The larva ran in all 294 constituencies. The samosas at the Lexicon Committee's post-election meeting were procured from a vendor whose name appears correctly in all documents. The Committee has noted this. The three voters named Nabijan Mondal, Sohidul Islam, and Muhammad Ali — a 73-year-old who voted for fifty years, a 49-year-old who asked who he could approach, and a 65-year-old who served in the Indian Army — are not members of the Lexicon Committee. The Committee has taken no formal action on their behalf. The Committee has no legal standing. The tribunal has a two-year backlog. The larva is doing well.