[This is a guest post by Rudraksh Lakra. The writer wishes to thank Pushpendra for his valuable research assistance and contribution to the piece.]
Introduction
In West Bengal, anti-incumbency sentiment, opposition fragmentation, and Hindu consolidation all played an important role in the BJP’s 207-seat victory in May 2026. Yet the sheer scale of this result, a party that had never governed the state winning two-thirds of its Assembly in a single election, flipping approximately 130 seats from the Trinamool Congress, generated intense controversy. Opposition parties raised allegations that the election had been manipulated, and the institutional independence of the Election Commission of India came under scrutiny. That scrutiny was shaped by the Special Intensive Revision (“SIR”) of electoral rolls, which resulted in more than 90 lakh gross deletions from West Bengal’s voter lists.
This is the second essay in the Programming Democracy series. The first examined how Assam’s 2023 delimitation exercise redrew constituency boundaries to structurally weaken Muslim political representation, deploying GIS mapping and algorithmic modelling. This essay examines the Bengal SIR through a similar lens: as an exercise in upstream electoral engineering that reshaped the conditions of democratic competition before polling began. It argues that the SIR reflected a broader pattern of digital structural authoritarianism, where political exclusion emerges through opaque technological systems, and administrative procedures marked by opacity and undue burdens that shape democratic participation through their routine operation. The third piece will draw parallels between Assam delimitation and the Bengal SIR to offer five broader lessons about the changing relationship between technology, electoral administration, and the conditions necessary for genuinely competitive and fair democratic contestation.
The structure and scale of the SIR process
The SIR undertaken in West Bengal before the 2026 Assembly elections was presented as an administrative exercise aimed at removing duplicate, deceased, shifted, and otherwise ineligible voters from the electoral rolls. Examining what the exercise actually produced requires moving beyond the question of whether any single actor intended to disenfranchise voters. I suggest that my concept of digital structural authoritarianism offers a useful lens for understanding the Bengal SIR.
Much of the existing scholarship on digital authoritarianism focuses on the deliberate use of digital technology by the state to suppress political opposition and preserve power. That account captures an important dimension of what unfolded in West Bengal. My argument, however, shifts attention away from individual intent and toward institutional design. It focuses on how the system itself behaves. Once established, technological systems can shape incentives, administrative behaviour, and political outcomes through their ordinary operation. As I shall demonstrate in this section and the next, the Bengal SIR reflects this dynamic clearly. We focus on the design of the process we shall see that large-scale exclusion foreseeable, while the absence of transparency made meaningful accountability extremely difficult. Together, these features allowed the exercise to shape electoral participation in ways that extended beyond any individual administrative decision.
One feature distinguished the Bengal SIR from comparable exercises conducted elsewhere: the introduction of the “logical discrepancy” category. The Election Commission initially identified 1.36 crore, which were then reduced to 91.46 lakh such cases through their algorithm. Of note is that this category was introduced in the Bengal after the revision process was already underway. As the following section demonstrates, the criteria governing it were highly arbitrary and functioned in an exclusionary manner. More than 90 lakh gross deletions were carried out from West Bengal’s electoral rolls, with approximately 66 lakh net deletions remaining after deceased electors were excluded from the count. A further 27 lakh voters categorised as “under adjudication” were omitted from the final rolls. While their appeals remained pending before tribunals constituted under the supervision of the Supreme Court. Polling on 23 and 29 April therefore proceeded while those appeals remained unresolved and those voters remained excluded from participation.
The significance of the exercise lies both in the scale of exclusion and in its demographic distribution. Gender-disaggregated data published by the Sabar Institute showed that 61.8 percent of deleted voters were women across 88 percent of constituencies, with the concentration particularly sharp in Scheduled Tribe constituencies. The correlation between logical discrepancy flags and Muslim population density across districts reportedly stood at 0.80. In Muslim-majority districts such as Murshidabad, Malda, and Uttar Dinajpur, discrepancy rates operated at nearly three times the state average. These patterns reflect the consequences of applying uniform documentary standards to populations whose relationship with documentation remains structurally unequal. Electoral participation increasingly depended upon satisfying systems built around assumptions of documentary consistency, stable records, and administrative legibility that substantial sections of the population, particularly those located at the social and economic margins, were never equally positioned to satisfy.
The political implications of these deletions remain difficult to conclusively evaluate because the underlying electoral data have not been released in analysable form. Nevertheless, the broader electoral context intensified concerns. The BJP secured 207 out of 294 Assembly seats despite never previously governing the state, flipping approximately 130 constituencies from the Trinamool Congress. In 150 constituencies, the reported number of deletions exceeded the eventual margin of victory. Although this does not establish direct causation, it underscores why the process has become a major political flashpoint and has generated serious concerns regarding both its electoral impact and the integrity of the exercise itself.
Technology and the design of exclusion
The SIR process relied upon ERONET 2.0, the Election Commission’s centralised electoral roll management system. Originally developed by C-DAC under the Ministry of Electronics and Information Technology, ERONET 2.0 has reportedly been operated by Tata Consultancy Services since 2023. The Commission did not publicly disclose this arrangement. A private corporation was therefore operating the digital infrastructure through which the voting rights of approximately 7.66 crore individuals were processed. The contractual framework governing this arrangement remained undisclosed, and public audit mechanisms were absent. The Commission also withheld the algorithmic criteria governing voter flagging and verification. The constitutional concern arising from this arrangement extends beyond privatisation alone. Decisions affecting democratic participation were increasingly transferred into technological systems functioning outside meaningful public scrutiny.
The structure of the exercise itself produced conditions where large scale exclusion became highly foreseeable because each stage of the process systematically increased the probability that ordinary documentary variation would be converted into grounds for suspicion and removal.
First, ERONET identified “logical discrepancies” through automated matching exercises comparing voter entries across databases and associated records. Any algorithmic matching exercise conducted at such scale inevitably carries an inherent margin of error. Although the Election Commission never disclosed the error rate or functioning of the algorithmic system, it was foreseeable that some voters would face exclusion solely by virtue of automated processing operating upon inconsistent underlying records. Further, software-generated flags then triggered summons and hearings requiring voters to establish their own eligibility. Through administrative design, the burden shifted from the state onto the citizen. That shift sat uneasily with the principle articulated by the Supreme Court in Lal Babu Hussein v. Electoral Registration Officer (1995), where the Court held that electoral authorities carry the obligation to establish disqualification before removing a voter from the rolls. Under the SIR, continued inclusion increasingly depended upon the voter disproving software-generated suspicion.
Second, the algorithm operated upon documentation systems already marked by inconsistency and fragmentation. Public records frequently contain spelling variations, transliteration differences, unstable addresses, incomplete birth documentation, fragmented family records, and informal naming conventions (see here, here, and here). These inconsistencies are especially prevalent among economically marginalised communities, migrant populations, women whose surnames change after marriage, and residents of informal settlements. Anand Teltumbde’s observation regarding the sociology of disenfranchisement is useful here. Administrative exercises never operate upon neutral social terrain. Who survives them depends upon who possesses documents, stable addresses, literacy to navigate forms, time to contest bureaucratic errors, and lives that leave paper trails the state recognises as legitimate. Processing records marked by documentary variation through automated systems at an enormous scale, hence, predictably generated large pools of voters categorised as “suspect”, “unmapped”, or “logically discrepant”. The structure of the system ensured that the burden of verification fell most heavily upon populations already situated at the margins of documentary stability.
Third, the algorithmic criteria also were highly arbitrary. Voters were reportedly flagged for minor mistakes like spelling variations between the 2002 Bengali-script rolls and the 2025 English-script rolls, for having more than six voters linked to a single ancestor, for parent-child age gaps outside fifteen to forty-five years, and for grandparent-grandchild age gaps below forty years. These assumptions treated ordinary features of social and documentary life as indicators of suspicion. Bengali names are frequently rendered differently across scripts, extended households often share common ancestors, and age gaps within families do not always conform to administrative expectations. Bengali surnames are often rendered differently across scripts and generations. The middle name “Kumar”, widely used across Bengali households, became a discrepancy when omitted from older records. Child marriage, historically prevalent in parts of rural Bengal, generated parent-child age gaps that the software classified as implausible. On 9 February 2026, the Supreme Court itself observed that the Commission’s tools were “eliminating natural differences” and that the assumptions built into the system failed to reflect ground realities. The software therefore converted common social and linguistic variation into categories requiring administrative scrutiny.
Finally, these concerns were compounded by the manner in which the Election Commission processed the underlying data itself. The Commission reportedly digitised the 2002 electoral rolls, most of which existed in Bengali script, translated them into English, and subsequently deployed further algorithms to compare those digitised records against current filings. The reliability of the entire exercise, therefore, depended upon multiple layers of automated processing operating upon inconsistent records. Errors introduced during digitisation and transliteration then became part of the very database used to generate discrepancy flags and verification proceedings. The structure of the process therefore embedded uncertainty into the foundation of the exercise itself while simultaneously treating those inconsistencies as grounds for heightened scrutiny.
The software environment itself remained unstable throughout the exercise. The Booth Level Officer application reportedly underwent 31 updates within two months, averaging a new version every two days, without accompanying manuals or formal guidance. Officials received contradictory instructions through WhatsApp groups and video conferences while discovering that the software’s behaviour changed repeatedly between updates. Devices froze, data disappeared, and work had to be repeated. Officials administering the exercise lacked access to stable rules, transparent criteria, or institutional mechanisms through which they could meaningfully challenge software outputs. Thus, the instability of the software environment further intensified the risk of arbitrary exclusion by subjecting officials and voters alike to constantly shifting procedures, inconsistent system behaviour, and technologically mediated decision making operating without stable rules or meaningful accountability.
Another concern relates to the effect of the software on human oversight. Under the Representation of the People Act, Electoral Registration Officers are the statutory authorities responsible for inclusion and exclusion from electoral rolls. ERONET, however, increasingly displaced that authority in practice. The Election Commission informed the Supreme Court that there was “no automated generation of notices” and that Electoral Registration Officers were “signing them off after application of mind.” However, reporting by The Reporters’ Collective suggested that notices were in fact generated at a mass scale through centralised and untested algorithms. Electoral Registration Officers, required to process thousands of cases each day, were effectively left with little option but to sign notices in a largely perfunctory manner within compressed timelines. One ERO in Kolkata reportedly described signing between 4,000 and 5,000 printed hearing notices daily. Human oversight therefore receded gradually through software-generated classifications and layered approval systems until algorithmic outputs substantially shaped which cases advanced through the process.
The exercise also demonstrated how technological systems organised around centralised design can distance the state from the people it governs. Many of the “logical discrepancy” cases reportedly involved simple spelling mistakes, transliteration differences, or minor documentary inconsistencies that officials with local knowledge could likely have resolved with relative ease. Instead, these cases became absorbed into complex procedural and technological workflows that neither officials nor citizens fully understood. Moreover, responsibility became diffused across software systems, approval chains, and administrative layers, making meaningful resolution more arduous.
The appellate process deepened these problems instead of remedying them. Of the 12.87 lakh voters whose cases remained pending before tribunals as polling approached, only 1,474 cases were reportedly heard before the Phase 2 election date. Of those heard, 1,468 voters were restored to the rolls, producing a 99.59 percent inclusion rate. That figure may indicate that a large proportion of the cases capable of substantive review involved wrongful exclusion at the initial stage. Yet the tribunal system was constituted too late and functioned at too limited a scale to process the volume of pending appeals before polling day. The restoration rate, therefore, applied to fewer than 0.12 percent of those seeking relief.
This disparity was built into the structure of the process itself. A system capable of generating exclusions at algorithmic speed was paired with an appellate framework incapable of resolving challenges within the electoral timeline. The consequence was that exclusions remained operative regardless of their merits because appeals could not be adjudicated in time. The concept of structural authoritarianism is useful here because the suppressive effect does not depend upon appeals being formally rejected. Delay itself becomes the mechanism through which exclusion acquires political force. The tribunals themselves generated additional concerns relating to the rule of law. Their mandate and operating procedures were never published, and they even lacked a publicly accessible website. Voters were therefore required to navigate institutions whose procedural framework remained largely inaccessible even as those institutions determined questions of democratic participation.
Opacity, Accountability, and Electoral Data
The opacity surrounding the SIR formed part of the structure of the exercise itself. The absence of transparency shaped how the process operated, how exclusions acquired institutional legitimacy, and how the political consequences of the exercise could be examined or contested.
The functioning of ERONET, including the logic governing discrepancy flags and verification thresholds, remained undisclosed throughout the exercise. Public explanations regarding classification methodologies, weighting systems, and data matching standards were absent. The Commission never publicly acknowledged TCS’s role in operating ERONET. The accuracy of the underlying 2002 digitisation process was never independently tested in public view. The accuracy of the underlying 2002 digitisation process, and translation was similarly left outside independent public scrutiny even though that database formed the foundation for later algorithmic verification.
At each stage where transparency could have enabled scrutiny, the system instead limited access. The publication of the final electoral rolls reflected this pattern clearly. Electoral data was released not as machine-readable datasets but as scanned PDF images resembling photographs of printed pages. These files could neither be searched nor computationally analysed. The Commission’s portal reportedly permitted downloads for only ten areas at a time, with CAPTCHA restrictions blocking automated analysis. The PDFs themselves were vastly larger than searchable equivalents while containing none of the underlying structured data. The PDFs themselves were significantly larger than searchable equivalents while containing none of the underlying structured information necessary for systematic scrutiny.
Moreover, according to reporting by Newslaundry, there was insufficient data to assess the impact of the SIR on the election. As booth-Level results, a complete accounting of additions alongside deletions, and detailed information regarding who was removed, the reasons for removal, and the demographic effects of those exclusions remained inaccessible. Independent researchers, journalists, political parties, and affected citizens therefore lacked the information necessary to meaningfully assess the operation and consequences of the exercise.
The legal challenge to the SIR reached the Supreme Court in July 2025. Petitioners argued that the SIR violated fundamental rights under Articles 14, 19 and 21 of the Constitution, operated arbitrarily and without adequate justification, and departed from principles of natural justice including the standard articulated in Lal Babu Hussein. These were foundational constitutional questions concerning the legality of the SIR itself.
On January 29, 2026, the Court reserved judgment on the batch of petitions challenging the exercise, placing itself in a position to deliver a merits ruling before the election. Instead, as Vasudev Devadasan has observed, the Court moved toward “dialogic review”: supervising aspects of the SIR’s implementation through orders directing acceptance of Aadhaar, publication of deletion lists, constitution of judicial officers, and establishment of appellate tribunals, while leaving the central constitutional challenge unresolved. The Court remained closely involved in managing the process without determining its legality. This approach effectively preserved the operational status quo in favour of the government while the exercise continued during an active electoral cycle in which the SIR itself had become politically consequential. Further, on April 13, 2026, a bench headed by Chief Justice Surya Kant declined the plea seeking permission for excluded voters to participate in polling scheduled for April 23 and 29.
The Court treated the pending appellate process as sufficient institutional safeguard, even though the tribunals had heard only 1,474 of the 12.87 lakh pending cases and functioned without publicly published mandates, or clear procedural rules such as regarding fresh evidence. Former Chief Election Commissioner S. Y. Quraishi highlighted the constitutional distinction at the centre of the dispute: the state may verify electoral eligibility, but constitutional entitlement to vote cannot depend upon whether administrative verification concludes before polling day. By allowing exclusions to remain operative while appeals remained unresolved, the Court effectively accepted the SIR’s shift away from the presumption of inclusion even as the legality of that shift remained undecided.
Structural authoritarianism directs attention toward the relationship between infrastructure and democratic participation. Political consequences emerged through algorithmic classification, institutional control over data, unequal access to systems, and limits placed upon meaningful scrutiny of the electoral process. In West Bengal, the result was a system whose effects could be widely experienced yet only partially examined because the underlying infrastructure remained inaccessible. The Court’s approach completed that institutional arrangement: exclusion emerged through algorithmic classification, acquired permanence through the limited capacity of the tribunals, and continued through judicial supervision of the process without a ruling on its constitutional validity.
Exclusion and the “Infiltrator” Narrative
The SIR unfolded within a broader political discourse centred on “infiltrators”, demographic anxiety, and documentary suspicion. Increasingly, certain populations were treated as inherently suspect unless they could continuously establish their legitimacy through documents. This altered the assumptions underlying democratic participation. Universal adult suffrage traditionally rests on the premise that adult residents are entitled to political participation unless disqualified through lawful procedure. During the SIR, however, large numbers of citizens were repeatedly required to prove their eligibility through forms, hearings, supporting records, and bureaucratic verification. Suspicion thus became embedded within the mechanics of participation itself. The political rhetoric surrounding the exercise deepened these concerns. Amit Shah’s statement in Parliament that the government’s policy was to “detect, delete and deport” situated electoral exclusion within a wider framework of administrative control. Removal from the electoral roll increasingly appeared linked to broader questions of citizenship, welfare access, and civic belonging. These anxieties manifested after West Bengal BJP Chief Minister Suvendu Adhikari on May 12 reportedly suggested that individuals excluded through the SIR process could lose access to government welfare schemes. On May 19, reports further indicated that administrative efforts could soon move toward “deportation” measures targeting alleged “illegal immigrants.” The precise contours of these policies remained unclear. However, the experience of Assam illustrates that once a person is deemed suspect for purposes of citizenship verification, that suspicion can persist for decades. Even within the best-case scenario, individuals may be referred to Foreigners Tribunals despite concerns regarding their independence and competence. As the alternative is illegal detention or deportation from the country.
The significance of the SIR lies not merely in the management of electoral rolls but in the emergence of a governance structure where documentary scrutiny, technological classification, welfare access, and political participation become closely interconnected.
Conclusion
The Bengal SIR produced its consequences through the ordinary operation of systems that were flawed by design, opaque by choice, and insulated from meaningful review by institutional arrangements that prioritised pace. An algorithm built on imprecisely digitised data flagged 1.36 crore voters as suspect. An undisclosed private contractor operated the infrastructure through which those flags became legal notices. A tribunal system constituted too late to function heard fewer than 0.12 percent of pending cases before polling day. A court that supervised the process extensively declined to rule on its legality before the election rendered that question academic. Each element reinforced the others, and together they produced an electorate that had been shaped before a single ballot was cast.
The concept of digital structural authoritarianism sheds light on what made this possible. The exercise required no centralised discriminatory directive because the architecture itself performed the discrimination. A documentation standard applied uniformly to a population whose records are structurally unequal will produce structurally unequal outcomes. Opacity shields those outcomes from verification. Delay converts unresolved appeals into effective disenfranchisement. The system, once in place, generates its political consequences through its routine operation, and those consequences outlast any individual decision made within it. What remains unresolved is the constitutional question the Supreme Court declined to answer before the election: whether the SIR was lawful at all. The judgment reserved in January 2026 will be delivered into a political landscape that the SIR has already shaped in ten States and three Union Territories, with a government already in office and an election already conducted on rolls whose legal basis remains contested. Whatever that judgment finds, it cannot restore the votes that were not cast. The more consequential question, for Indian democracy, is therefore whether the architecture that enabled the Bengal SIR will remain an exceptional mechanism or evolve into a standardised model for administering and shaping the electorate itself.
