On 19 May 1982, in fifty polling stations of the Parur assembly constituency in Kerala, Indian voters pressed buttons on an electronic voting machine for the first time, and the EVM entered the republic’s electoral architecture through a by-election that the Supreme Court would partly undo two years later. The four decades after Parur carried the machine from that contested trial to the centre of the largest voting operation on earth, and then attached a printed paper slip to it. This article defends one claim about that trajectory. The EVM and VVPAT story is neither the simple march of electoral modernisation that the technological-progress framing describes, nor the sequence of integrity scandals that the controversy framing assembles. It is a three-phase architectural transition, from paper ballots identified by symbol, to a machine that recorded votes with no paper trail, to a machine paired with a voter-verified paper audit trail, and at every phase the same question organised the politics: how can a vote recorded inside a device be checked against something outside the device? That verification question, raised by a losing candidate in 1982, pressed by litigants in 2009 and 2013, priced by statisticians in 2019 and answered with two fresh directions in April 2024, is the substantive content of the story. The machines changed. The question did not, and each court, committee and Commission that handled the machines left the question standing in a new form.

The Paper Inheritance: What the EVM Was Built to Replace

The machine arrived as a remedy for documented failures of paper. Under the founding paper-ballot-symbol-based architecture, designed by Chief Election Commissioner Sukumar Sen for the general election of 1951 and 1952, each candidate had a separate ballot box in the polling station, marked with the candidate’s election symbol, and the voter dropped a blank ballot paper into the box of the chosen candidate. The design solved the literacy problem of a largely non-literate electorate by converting choice into recognition: a voter who could not read a name could identify a symbol and a box. It also built the weaknesses of paper into the count. Ballot papers could be marked wrongly and invalidated, boxes could be stuffed, booths could be captured by local strongmen who voted on behalf of a village, and counting by hand across more than a million ballot boxes took days and invited dispute at every table. The post-1951 first general election context therefore bequeathed two inheritances at once: a symbol-centred method of voting that every later technology preserved, and a paper process whose physical vulnerabilities the Election Commission of India spent three decades trying to engineer away.

The Commission first mooted electronic voting in December 1977, when the idea of replacing the ballot paper with a machine was formally placed on its agenda, and a prototype followed in 1979. On 6 August 1980 the Commission demonstrated the prototype to leaders of the political parties, seeking the broad agreement that a change in the act of voting required. The design work went to the Electronics Corporation of India Limited at Hyderabad, a public sector electronics firm, and Bharat Electronics Limited at Bangalore was later brought in as a second manufacturer, a dual-source arrangement that persists. The engineer M. B. Haneefa is credited in the Commission’s own institutional histories with the early design of an Indian voting machine around 1980. Three features of this founding moment matter for everything that followed. The machine was conceived inside the state, by the Commission and two government-owned firms, not bought from a commercial vendor market. It was designed as a deliberately simple device, a single-purpose recorder rather than a computer in the ordinary sense. And it was introduced experimentally, constituency by constituency, under the Commission’s general powers in Article 324 of the Constitution, before Parliament had written a single line authorising it. The third feature produced the first crisis of the EVM story, and the crisis was legal before it was technical: a court had to decide whether the Commission’s constitutional superintendence could carry a technology that the Representation of the People Act 1951 and the Conduct of Elections Rules 1961 did not mention.

The paper era also fixed the benchmark against which every later verification debate was argued. Under paper, a disputed count could be reopened: the ballots existed as physical objects, a returning officer could recount them, and an election court could inspect them. Verification was slow, corruptible at the margins and expensive, but it was external to the act of recording, because the record and the recording instrument were different things. The paper ballot was both the instrument and the evidence. When the machine replaced it, the record migrated inside a sealed electronic device, and the evidence for the result became the device’s own memory read back by the device itself. That single architectural fact, invisible in the celebration of faster counting in 1982, is the seed of the verification question that this article traces through four decades of amendments, committee reports, demonstrations and judgments. The defects were not abstractions, and the scale of the first machine-free decades explains the constituency the machine later commanded among election officials themselves. The general election of 1951 and 1952 took more than four months to poll, from 25 October 1951 to 21 February 1952, across an electorate of about 173 million, and counting and re-polling stretched the uncertainty further. In later decades the pathologies acquired names that entered the political vocabulary: booth capturing, at which armed supporters seized a polling station and voted its ballot papers in a block; ballot stuffing during transport and storage; and the slow violence of counting halls in which result sheets were contested table by table for days. Bihar’s elections of the 1970s, 1980s and 1990s supplied the standard examples cited in the Commission’s own institutional histories, and returning officers who had presided over such elections became the machine programme’s most committed internal advocates, a fact the institutional literature on the Commission records and the security literature rarely weighs. A presiding officer who has watched a booth being captured does not experience the replacement of paper by a sealed recorder as a loss of verifiability. The loss is real, but it is the second term in his calculation, and the first term is the crowd at the door. Any honest adjudication of the EVM story has to hold both terms in view at once, which is precisely what a trajectory organised by the verification question, rather than by either camp’s founding memory, is built to do.

The technological-progress framing treats the paper era as a prologue of defects overcome. The trajectory reading treats it as the source of the standard, independent checkability, that the machine era would spend forty years trying to rebuild in electronic and then in paper-assisted form.

Parur, 19 May 1982: The First EVM Trial and Its Undoing

The first deployment was modest in scale and exact in its limits. For the by-election to the Parur assembly constituency, numbered constituency 70 in Kerala and also rendered Paravur in later records, held on 19 May 1982, the Commission directed that voting machines be used in fifty of the constituency’s eighty-four polling stations, with the remaining thirty-four stations voting on paper ballots in the ordinary way. Six candidates contested. The two principals were Sivan Pillai of the Communist Party of India and A. C. Jose of the Indian National Congress. When the result was declared on 20 May 1982, Sivan Pillai had secured 30,450 votes against 30,327 for Jose, a margin of 123 votes, of which 19,182 had been cast by machine and the rest on paper. The narrowness of that margin converted an experiment in voting technology into a question of electoral legality, because the losing candidate’s petition did not allege that the machines had counted wrongly. Jose’s case was more fundamental: the Representation of the People Act 1951 prescribed voting by ballot, the Conduct of Elections Rules 1961 supplied ballot papers and ballot boxes, and neither text gave the Commission power to substitute a machine, so every vote recorded electronically was, on his argument, received without authority of law.

Which polling stations at Parur used machines on 19 May 1982?

Fifty of the eighty-four polling stations in the Parur constituency used electronic voting machines on 19 May 1982, and the remaining thirty-four stations voted with paper ballots and ballot boxes under the Conduct of Elections Rules 1961. The fifty machine stations recorded 19,182 votes in the by-election.

The division of one constituency between two voting technologies became the factual foundation of the litigation that followed, because it allowed the Supreme Court, two years later, to sever the machine stations from the paper stations when it came to fashion a remedy, annulling the half of the poll that had been conducted without statutory authority and leaving the lawful half standing.

The Kerala High Court dismissed Jose’s election petition on 12 October 1982, holding that the machine system was not vitiated by non-compliance with the Act and the Rules and that the votes had been rightly counted. Jose appealed. On 5 March 1984 a three judge bench of the Supreme Court, in A. C. Jose v Sivan Pillai, reported at AIR 1984 SC 921, allowed the appeal and reversed the institutional logic of the experiment. The Court held that Article 324 gives the Commission plenary superintendence over elections, but that the power is to be exercised in conformity with the laws Parliament has made, supplementing them where they are silent and never supplanting them where they speak. The Act and the Rules spoke in detail about ballot papers, ballot boxes and the manner of marking a vote. A direction to vote by machine in fifty stations was therefore not gap-filling but law-changing, and it was beyond the Commission’s authority. The election was set aside for the fifty machine stations, their results were annulled, and a fresh poll was ordered in those stations by ballot paper; the results in the thirty-four paper stations were expressly left untouched. The fresh poll followed on 21 May 1984.

Why did the Court order a fresh poll at only fifty Parur stations in 1984?

The Court confined its remedy to the fifty machine stations because its objection was jurisdictional, not arithmetical: the defect was the absence of legal authority for machine voting, not any proven error in the count. The thirty-four paper stations had complied with the Conduct of Elections Rules 1961, so their results stood undisturbed.

Severing the two sets of stations let the Court annul the unauthorised part of the poll, order it held again by ballot paper, and preserve the lawful part without forcing the constituency to vote twice over ground that no party had shown to be tainted.

The composition of the bench underlines how seriously the Court took a dispute about fifty polling stations in one Kerala constituency. The judgment of 5 March 1984 was delivered by a three judge bench of Justices S. Murtaza Fazal Ali, Ranganath Misra and A. Varadarajan, with Justice Fazal Ali writing, and its authority has never been disturbed: the proposition that Article 324 empowers the Commission to supplement electoral law but never to supplant it remains the frame inside which every later machine case was argued, including the cases the Commission won. The 2013 and 2024 benches deferred to the Commission’s expertise on design and protocol precisely because Parliament had, after 1984, supplied the statutory warrant whose absence the Parur bench had condemned, and the deference of the later judgments is therefore unintelligible without the strictness of the first one. A Commission that had been indulged in 1984 might have built the machine era on directions alone, and every protocol within it would have carried the Parur defect in its foundations. Instead the era was built on Section 61A, and the section’s existence is the 1984 judgment’s most durable product, more consequential for the architecture than the annulment of the fifty stations that made the headlines of the day.

Read as a technical episode, Parur is a footnote: an experiment run ahead of its statute. Read architecturally, it is the template of the entire EVM story. The Court did not find that electronic recording was inaccurate, and it did not hold that the votes cast by machine were false. It held that a change in the instrument of voting is a change in the law of elections, requiring Parliament’s text and not the Commission’s direction, because the instrument determines what evidence an election produces and what a disappointed candidate can later examine. The verification question was present in embryo inside a case argued entirely in the vocabulary of statutory authority. A paper ballot could be produced before the Court. A machine’s count could only be asserted by the institution that had deployed the machine. The 1984 bench did not use the language of audit trails, but its holding rests on the same intuition that would drive the litigation of 2013: the form in which a vote is recorded decides who can check it, and a democracy may not leave that decision to administrative convenience. The technological-progress framing has no place for Parur except as delay. The trajectory reading treats it as the first statement of the story’s governing problem, delivered before the machine had a statute, a committee or a paper trail.

Section 61A and the Legal Architecture of the Machine

Parliament answered Parur with a short amendment and a long silence. The Representation of the People (Amendment) Act, Act 1 of 1989, passed in December 1988 and brought into force on 15 March 1989, inserted Section 61A into the Representation of the People Act 1951. The section empowers the Commission to specify the constituencies in which votes shall be given and recorded by voting machines, and adapts the counting provisions of the Act to machine recording. Its brevity is its significance. Parliament authorised the machine as a method, leaving the machine’s design, custody, testing and counting protocol to the Commission’s directions and to the amended Conduct of Elections Rules. The legal architecture of Indian electronic voting therefore rests on a one-section statutory foundation and a large body of administrative protocol built on top of it, a distribution of authority that later litigants would attack from both directions: critics argued that the Commission’s protocols were too thin a basis for displacing paper, and the Commission argued that Section 61A plus Article 324 gave it room to refine those protocols as the technology matured. The 1989 amendment also closed the Parur defect precisely as the 1984 bench had framed it, by supplying the statutory text whose absence had annulled fifty polling stations. Between 1989 and 1998 the machines were manufactured, stockpiled and refined, and the Commission used the interval to constitute the expert scrutiny that the original experiment had lacked, referring designs and upgrades to a Technical Expert Committee of engineers and academics drawn principally from the Indian Institutes of Technology, whose approval became a standing stage in the machine’s development.

The Long Interregnum: Design Work and Consensus from 1982 to 1998

Between the Supreme Court’s annulment of March 1984 and the statutory machine’s return in November 1998 lie fourteen years in which the EVM existed as an object under development rather than as an instrument in use, and the interregnum shaped the machine’s later politics as deeply as any courtroom episode. The Commission shelved field deployment after Parur, but it did not shelve the programme. The internal correspondence of the Commission through the 1980s, preserved in its archives and still the least-cited primary record of the whole story, documents the design choices being argued out inside the institution: how simple the device should be, what a presiding officer should and should not be able to do to it, how the count should be stored, and what would persuade political parties that a button could replace a ballot without replacing their right to object. The archival record matters because it shows the verification question being posed by the institution to itself, in engineering vocabulary, years before litigants posed it in constitutional vocabulary. A machine designed by administrators who had watched paper elections being stolen physically was always going to privilege custody and sealing over inspectability, because the thefts its designers remembered were thefts of objects, boxes stuffed and booths captured, and not manipulations of code that no one in the India of 1982 had yet imagined as an electoral crime.

The political work of the interregnum ran beside the engineering. The 1980 demonstration to party leaders had established the pattern the Commission would follow for four decades: no expansion of the machine without a staged performance of consultation, at which parties could register consent or dissent on the record. Parliament’s insertion of Section 61A in 1989 removed the legal bar, but the Commission waited a further nine years before the 1998 deployment, building stock, refining the design through its expert committee processes, and letting the idea of the button settle into the routines of returning officers and the expectations of the press. The delay is usually read as bureaucratic caution. Read architecturally, it was constituency building for a technology: by 1998 the machine arrived in its sixteen constituencies not as a novelty imposed by direction, the error of 1982, but as the endpoint of a statute, an expert process and a consultative record, each layer of which would later be cited in court as part of the system’s defence. When the Delhi High Court and then the Supreme Court examined the machine decades later, the interregnum supplied much of the evidentiary record on which deference rested, the affidavits of committees, trials and consultations that allowed judges to describe the system as considered rather than improvised. The trajectory reading therefore assigns the quiet years a causal weight that both rival framings deny them. The progress framing treats 1984 to 1998 as lost time before the real story starts. The controversy framing treats it as irrelevant, since its disputes all postdate 2004. In the record, the interregnum is where the Indian answer to the verification question was frozen into its durable form: trust organised procedurally, by committees and consultations and seals, with the device itself kept deliberately opaque even to most of the officials who operate it. Every later conflict in the story is a conflict between that procedural answer, formed between 1982 and 1998, and claimants demanding a substantive one.

Anatomy of the EVM: Ballot Unit, Control Unit and the Design Defence

An Indian EVM is a pair of units joined by a cable five metres long. The ballot unit sits inside the voting compartment and carries the names and symbols of the candidates, each against a button, with a lamp that lights when a vote registers; up to four ballot units can be linked, accommodating as many as sixty-four candidates, and from 2013 the final position carried the NOTA option, none of the above, after the Supreme Court’s judgment of 27 September 2013 in PUCL v Union of India required a negative vote facility. The control unit stays with the presiding officer outside the compartment, arms the ballot unit for one vote at a time, and stores the count in electronic memory. Under the Commission’s published specification a machine can record a maximum of 3,840 votes. Nothing in the pair connects to any network: there is no radio receiver, no port for an external device in ordinary use, and no operating system of the kind that runs a general purpose computer. The programme is burnt into a one-time programmable microcontroller at manufacture, in masked form, inside secure facilities of the two manufacturers, and the Commission’s case, recited in detail by the Supreme Court on 26 April 2024, is that the burnt code cannot be altered after manufacture, that each key press is dynamically coded between the units so that signals cannot be intercepted and re-read, and that the firmware is agnostic to candidates and parties because symbols and names are loaded onto the ballot paper of the ballot unit and, later, into the VVPAT, without touching the counting programme.

Which two public sector firms build the machines, and in which cities are they based?

Bharat Electronics Limited, based at Bangalore, and the Electronics Corporation of India Limited, based at Hyderabad, build India’s electronic voting machines and VVPAT units. The Electronics Corporation received the original design assignment after the prototype demonstration of August 1980, and Bharat Electronics joined as second manufacturer, keeping supply and engineering knowledge inside two government-owned firms.

How many votes can one control unit record at most under the published specification?

Under the Election Commission’s published specification, an electronic voting machine can record a maximum of 3,840 votes, and its linked ballot units can carry sixty-four candidates plus the NOTA position added after September 2013. Polling stations are assigned far fewer electors, so the ceiling works as an engineering margin.

The figure still matters to the architecture of the system, because it fixes the largest pool of votes that a single sealed memory can ever hold, and therefore the largest loss that one compromised or defective unit could in principle inflict on a count before the paper trail, where one exists, could expose it.

Around the hardware the Commission built an administrative architecture that it defends with equal insistence: manufacture under expert committee scrutiny, a first-level check of every machine before use at which political parties’ representatives may be present, two-stage randomisation of machines to constituencies and then to polling stations so that no one can know in advance which physical unit will serve which station, a mock poll on polling morning in the presence of polling agents, sealing of the control unit with the signatures of candidates’ agents, storage in guarded strong rooms, and counting conducted under observation with the result sheets signed at each table. Two generations of design, designated M2 and M3 after the original M1 fleet, added tamper detection that renders a machine inoperative on unauthorised opening, internal self-diagnostics, and stronger cryptographic signing of the code. The M3 generation, vetted by the Technical Expert Committee, is the fleet described in the 2024 judgment. Sandeep Shastri, in essays on electoral reform written across the 2000s and 2010s, read this combined apparatus through the Commission’s survey evidence on voter confidence, treating trust in the count as a measurable quantity that the institution was entitled to defend. E. Sridharan, in his essays of the post-2014 period on institutions under strain, treated the same apparatus as a case of an institution answering a legitimacy question with procedural depth. Both readings accept the Commission’s premise that the defence of the machine is a defence of a system, hardware plus protocol, and both therefore locate the verification question exactly where the courts would later find it: not inside the microcontroller alone, but in the joints between the device and the procedures surrounding the device.

Counting Day: Where the Architecture Meets Its Public

Polling day produces the record; counting day produces the result, and the distance between the two is where the architecture’s promises are tested in front of the candidates who lost. On the morning of the count, control units and VVPAT units are brought from the strong room to the counting hall under the observation of candidates and their agents, in halls arranged so that each assembly segment’s machines pass across tables in sequence. At each table the returning officer’s staff unseal the control unit, and the result is obtained by pressing the result key, which displays and permits the recording of the candidate-wise totals stored in the unit; the counting supervisor records the figures in the result sheet, and the votes recorded for the station are reconciled against the presiding officer’s account in Form 17C, the statutory record of votes cast at the station, copies of which polling agents received at the close of poll. Since 2019 the draw of lots for the five VVPAT stations of each segment is held in the presence of candidates or their agents, and the slip count proceeds at a dedicated table, the printed archive opened, sorted and tallied by hand while the electronic count from the same stations waits for comparison. The choreography is slow by design, and its slowness is the point: every stage at which a figure changes hands is performed before witnesses whose principals have an interest in finding error, and every figure that enters the final compilation has passed through at least one pair of hostile eyes.

After counting, the machines and the sealed slip compartments return to storage, where they remain through the period within which an election petition may be filed under the Representation of the People Act 1951, forty-five days from the declaration of the result, the same forty-five day horizon the Supreme Court wrote into its 2024 directions for the sealed symbol loading units. The retention period exposes a quiet asymmetry in the architecture that litigants have noticed and that the framed dispute rarely names. The paper record of the election is fullest at the moment of counting and wastes thereafter: slips are a working archive for the petition window, not a permanent public record on the model of the ballot papers that election courts once inspected decades later, and the machines themselves are eventually cleared and returned to the fleet for later elections. Verification in the Indian system is therefore an event with a season, concentrated in the weeks after polling, rather than a standing property of a preserved record. Defenders answer that a democracy needs finality more than it needs a perpetual audit, and that the petition window plus the courts’ powers to order inspection give any genuinely wronged candidate a forum. Critics answer that a check which expires before suspicion matures is a check designed for the institution’s calendar rather than the citizen’s. The trajectory reading registers the asymmetry as one more relocation of the governing question, this time in the dimension of time: not only how much of the record may be checked, and by whom, but for how long the record survives in checkable form at all.

The Rollout of the EVM, from 1998 to the Full Election of 2004

The statutory machine returned to polling stations in November 1998, in sixteen assembly constituencies spread across Madhya Pradesh (five), Rajasthan (five) and the National Capital Territory of Delhi (six), covering nearly three thousand polling stations. The Commission’s press note of 25 June 1999 called the experiment an unqualified success and cited a study conducted for it by the Centre for the Study of Developing Societies reporting that more than ninety-five percent of electors had welcomed the machines. The expansion then proceeded by steps that the electoral calendar supplied: forty-five parliamentary constituencies in the general election of 1999, the full Haryana assembly in February 2000, and in May 2001 the complete assembly elections of Tamil Nadu, Kerala, Puducherry and West Bengal conducted on machines. The assertive-commissioner-architectural-shift of the 1990s had rebuilt the Commission’s public authority under T. N. Seshan and his successors, and the machine programme drew on that authority: a Commission that had enforced the Model Code of Conduct against sitting governments could present a change in the instrument of voting as one more act of institutional housekeeping. The political setting helped as well, since the Rao government’s institutional-political context had normalised technocratic reform carried out by administrative decision between elections rather than by parliamentary debate during them.

In May 2003 the Commission directed that subsequent general elections be held with machines throughout, and the fourteenth Lok Sabha election of April and May 2004 became the first fully electronic general election in India. Chief Election Commissioner T. S. Krishnamurthy announced a deployment of about 1.08 million machines for an electorate of roughly 675 million, voting in four phases between 20 April and 10 May 2004. The operational gains were immediate and were the core of the technological-progress case: no invalid votes caused by wrongly marked papers, counting completed in hours rather than days, an end to the late-stage booth capturing that had disfigured polling in parts of Bihar and Uttar Pradesh, and a drastic reduction in the paper, printing and storage burden of a national election. The 2004 verdict itself, which ejected the incumbent National Democratic Alliance government that exit polls had expected to win, generated the first national tampering allegation of the machine era, when the advocate Pran Nath Lekhi petitioned the Delhi High Court arguing that the machines had been doctored to favour the victorious United Progressive Alliance; the petition rested on the gap between the result and the projections, the Court found no merit in it, and it was withdrawn. The pattern of that episode became a fixture of the next two decades and must be recorded plainly, because both framings exploit it: losing camps questioned the machines, winning camps defended them, and the same party could occupy either position in successive electoral cycles. The trajectory reading draws a different lesson from the 2004 transition. The year the machine became universal was also the year the verification question became structural rather than occasional, because from 2004 onward every vote in every parliamentary election existed only as an entry in a sealed electronic memory, and the paper evidence that had underwritten recounts since 1951 existed nowhere at all.

One consequence of the 2004 transition deserves separate record, because it altered the experience of elections for every participant and is easy to overlook in a story told through litigation. Under paper, an Indian general election ended in instalments: counting spread across days, leads changed as distant centres reported, and the interval between the close of poll and the making of a government was filled with a rolling negotiation conducted in public against incomplete numbers. The machine compressed the count into a single day of a few hours, and with the compression came a new political rhythm in which verdict, concession and government formation could all occur within twenty-four hours of the last count. Election officials valued the compression as a security good in itself, since a result known quickly is a result harder to unsettle by pressure on individual counting centres, and the parties reorganised their own operations around it, building the war rooms and data units that a one-day count rewards. The compression also removed the interval in which a doubtful count under paper could be interrogated before it hardened into a government. A disputed paper count could be reopened box by box while the outcome still hung in the balance of days; a disputed machine count is disputed after the outcome has been announced, celebrated and acted upon, which raises the practical threshold for disturbing it far above the legal threshold an election petition sets. Critics of the machine era have rarely isolated this temporal effect, and defenders have rarely acknowledged it, but it belongs in any complete account of why the verification question acquired its urgency after 2004. The faster the count became, the more the system’s legitimacy came to depend on checks that operate before or during the single day of counting, and the less comfort any participant could take from remedies scheduled for the weeks after it.

The Decade of Trust: Challenges, Demonstrations and the 2010 Study

For five years the universal machine operated with its legitimacy largely unexamined in public, and the Commission treated the question as closed. The general election of 2009 reopened it. After the Bharatiya Janata Party’s second successive defeat, its senior leader L. K. Advani stated publicly that the machines’ reliability was open to question and that several democracies had retreated from electronic voting, a position the party’s G. V. L. Narasimha Rao developed into a book-length indictment of the machine system. The Commission responded with the most theatrical instrument in its possession. From 3 to 8 August 2009, at Vigyan Bhawan in New Delhi, it placed one hundred machines, drawn at random from strong rooms in ten states, before invited sceptics, party nominees and technical experts, and invited them to demonstrate that the machines could be tampered with. The Commission’s recorded outcome was that no participant demonstrated tampering, and it published the exercise as a vindication.

What did the Commission do with one hundred machines at Vigyan Bhawan in August 2009?

Between 3 and 8 August 2009 the Election Commission placed one hundred electronic voting machines, drawn at random from strong rooms in ten states, at Vigyan Bhawan in New Delhi, and invited parties, petitioners and technical experts to demonstrate tampering under observation. The Commission recorded that no participant succeeded in demonstrating tampering on any machine.

The Commission treated the week-long exercise as a public vindication of the fleet. Its critics disputed that reading, on the ground that access to the machines was supervised, time-bound and hedged by conditions, so that the failure to demonstrate tampering in a hall in New Delhi proved, at most, that tampering could not be improvised on demand by visitors working under the institution’s own rules.

The critics’ substantive answer came not at Vigyan Bhawan but in a laboratory. In February 2010 an anonymous source supplied a Hyderabad technologist, Hari K. Prasad of Netindia, with a genuine control unit obtained from official storage, and Prasad worked on it with J. Alex Halderman of the University of Michigan and the Dutch researcher Rop Gonggrijp, who had led the campaign that ended machine voting in the Netherlands. Their peer-reviewed study, authored by Scott Wolchok, Eric Wustrow, Halderman, Prasad and colleagues and presented at the ACM Conference on Computer and Communications Security in October 2010 under the title Security Analysis of India’s Electronic Voting Machines, demonstrated two attacks on the real device. In the first, the team replaced the control unit’s display board with a look-alike board fitted with a Bluetooth radio, which could receive instructions from a mobile phone and alter the totals the machine displayed and stored. In the second, a pocket-sized clip-on device attached to the memory chip could rewrite stored votes between polling and counting. The study’s authors were careful about what their work proved: the attacks required physical access to a machine, and their claim was not that any Indian election had been stolen but that the system’s security rested on custody procedures and on the secrecy of the design, and that a determined insider with access could defeat both. The state’s response hardened the dispute. On 21 August 2010 Mumbai police arrested Prasad in Hyderabad in connection with the theft of the machine from a Mumbai collector’s office, and held him for refusing to name his source; he was released on bail, and later in 2010 received the Electronic Frontier Foundation’s Pioneer Award. The exchange left both positions at their strongest and irreconcilable on procedure: the researchers held that a voting system that cannot survive independent examination cannot demand trust, and the Commission held that unauthorised possession of a machine was a crime against election security, that its supervised challenges were the lawful forum for examination, and that no attack had ever been shown on a machine under election conditions, with the full chain of seals, randomisation and agent oversight in force.

The decade of trust ended in a reversal of political positions that the record should state without varnish, because it disciplines every partisan reading of the story. In 2009 the doubts had come from the Bharatiya Janata Party after defeat; after the state elections of February and March 2017, in which that party won Uttar Pradesh and Uttarakhand by large margins, the doubts came from the other side. The Bahujan Samaj Party alleged on the day of the Uttar Pradesh result that the machines had been tampered with, the Aam Aadmi Party took up the allegation in Delhi, and on 9 May 2017 the AAP legislator Saurabh Bharadwaj demonstrated in the Delhi Legislative Assembly a look-alike device which he operated to show how a machine might be made to miscount. The Commission’s reply was categorical: the device demonstrated was a prototype look-alike and not an Election Commission machine, and a demonstration performed on a replica built by the demonstrators could prove nothing about the fleet in the strong rooms. On 12 May 2017 the Commission met the national and state parties, forty-two parties attending, at which most parties supported machine voting with a paper trail while a minority pressed for a return to ballot papers, and on 20 May 2017 Chief Election Commissioner Nasim Zaidi announced the EVM challenge, a public invitation to the parties to attempt tampering with real machines taken from the strong rooms of the states that had voted that spring. The challenge of 3 June 2017, whose course the FAQ section of this article records in detail, ended with the two participating parties, the Nationalist Congress Party and the Communist Party of India (Marxist), declining to attempt a tampering demonstration at all, and with Zaidi declaring the tamperability question closed. The CPI(M) answered that the framework of the challenge had been too restrictive to settle anything. The sequence from March to June 2017 is the strongest single exhibit for each framing and therefore settles nothing between them: the progress reading sees accusers retreating when handed real machines, the controversy reading sees an institution writing the rules of its own examination and then grading the paper. The trajectory reading notes what both miss. By 2017 the parties were no longer arguing about whether the machine recorded votes correctly in the polling station. They were arguing about who controls the conditions under which correctness may be tested, which is the verification question wearing its political clothes, and it is the same argument the Supreme Court would be asked to settle, in statistical dress, two years later.

The Paper Trail Verdict: Swamy, the Delhi High Court and the Judgment of 8 October 2013

The laboratory dispute entered the courts through the one participant who had argued from the start that the answer was architectural rather than forensic. Subramanian Swamy, then president of the Janata Party, filed a writ petition in the Delhi High Court in 2009 contending that a machine with no independent record of each vote could not satisfy the requirements of a free and fair election, and seeking a direction that the Commission add a paper trail or a paper receipt to the act of voting. On 17 January 2012 the Delhi High Court declined to issue the direction Swamy sought, holding that it could not direct a return to paper ballots or dictate the Commission’s technology choices, but the judgment contained the observation around which the appeal was built: the Court recorded that in their existing form the machines could not be described as tamper-proof in an absolute sense, and it expected the Commission to engage with the paper trail proposal in consultation with political parties. Meanwhile the Commission’s own position was moving on a parallel track. At an all-party meeting on 4 October 2010, parties had pressed the Commission to explore a voter-verified paper audit trail, and the Commission assigned the design of a printer unit to its two manufacturers under the Technical Expert Committee’s guidance. Field trials of the prototype ran in July 2011 at Thiruvananthapuram, Delhi, Cherrapunji, Jaisalmer and Leh, a second trial followed in July and August 2012 after redesign, and on 19 February 2013 the Technical Expert Committee approved the final VVPAT model, which the Commission demonstrated to political parties on 10 May 2013. On 14 August 2013 the Conduct of Elections Rules 1961 were amended to permit the printer unit’s use, and on 4 September 2013 VVPAT units were used in a real election for the first time, in all twenty-one polling stations of the Noksen assembly constituency by-election in Nagaland.

The Supreme Court heard Swamy’s appeal against the Delhi High Court with this record before it, and the judgment of 8 October 2013 in Subramanian Swamy v Election Commission of India, delivered by a bench of Chief Justice P. Sathasivam and Justice Ranjan Gogoi, must be stated by its holding, because the holding is narrower and stronger than most summaries of it. The Court held, first, that a paper trail is an indispensable requirement of free and fair elections, and that voter confidence in the machines could be achieved only with its introduction. It held, second, that EVMs fitted with the VVPAT system ensure the accuracy of the voting system, treating the printer not as a rival to the machine but as its completion. It directed, third, that the Commission introduce VVPAT in gradual stages or area by area in the ensuing general elections, the choice of states and booths being left to the Commission, and it directed the Government of India to provide the financial assistance required for procuring the units, the Commission having already sought sanction for twenty thousand VVPAT units, ten thousand from each manufacturer, at a cost of about Rs 38.01 crore. The judgment’s reasoning turned on expression as much as on audit: a vote, the Court wrote, is an act of expression of immense importance in a democratic system, and the voter who expresses it is entitled to the fullest transparency the system can provide. In the vocabulary of this article, the 2013 bench rebuilt, in judicial form, the standard that Parur had assumed and that the paper ballot had embodied by accident: the record of the vote must exist in a form that does not depend on the recording device for its own verification. Crucially, the Court achieved this by augmentation rather than replacement. It did not question the legality of machine voting under Section 61A, it did not direct any return to paper ballots, and it praised the Commission’s own development work. The architecture after October 2013 was therefore a machine with a paper shadow, and every subsequent dispute in the story is a dispute about the shadow: how much of it must be counted, by whom, and with what consequence when shadow and memory disagree.

The Three-Phase EVM-VVPAT Trajectory Map

The argument of this article can be carried in a single table, and the table below is the findable artifact of the piece: the Three-Phase EVM-VVPAT Trajectory Map, which divides the story by architecture rather than by controversy, and assigns to each phase its governing verification question and the litigation and protocol development through which that question was worked on. Phase One runs from the Parur trial of 1982 to the universal machine election of 2004 and the paperless decade that followed it, and its question is whether a count stored only in a sealed memory can be checked at all. Phase Two runs from the Swamy judgment of October 2013 to the first fully VVPAT-covered general election of 2019, and its question is what the paper trail is for: a voter’s momentary reassurance, an audit instrument, or the true record in case of dispute. Phase Three opens with the counting protocol battles of 2019 and continues through the judgment of 26 April 2024, and its question is quantitative and forensic at once: how large a sample of paper must be counted, and what inspection of the machine’s burnt memory a losing candidate may demand. The phases overlap at their edges, as the table shows, because the judgment that created Phase Two was delivered while Phase One’s fleet was still the only fleet available for the election of 2014. Training staff and counting officials who circulate the Commission’s published counting-day manuals as presentation files, the kind of deck a PPT viewer opens on any screen, meet the phases in the same compressed form: one set of instructions for machines alone, one for machines with printers, and one for the draw of lots that selects the stations whose slips will be counted by hand.

Phase	Years	Architecture	Governing verification question	Litigation and protocol development
Phase One: the machine alone	1982 to 2014	EVM without paper trail under Section 61A (1989)	Can a count held only in sealed electronic memory be independently checked?	A. C. Jose v Sivan Pillai (1984); full rollout directed May 2003; Vigyan Bhawan challenge 2009; Wolchok and others study 2010
Phase Two: the machine with a paper shadow	2013 to 2019	EVM with VVPAT added as printer unit	What is the paper trail for: voter reassurance, audit instrument or record of last resort?	Subramanian Swamy v Election Commission (8 October 2013); Noksen first use (4 September 2013); eight Lok Sabha pilots 2014; Goa full state 2017
Phase Three: the counted sample	2019 onward	EVM with VVPAT universal, slips counted at sampled stations	How much paper must be counted, and what forensic inspection may a candidate demand?	Naidu v Union of India (8 April 2019, five stations per segment); ADR v Election Commission (26 April 2024, symbol loading unit sealing and burnt-memory check)

The map exposes what the two rival framings conceal. The technological-progress framing reads the rows as ascent: each phase corrects the last, and the story tends toward a finished system. The controversy framing reads the rows as recurrence: each phase generates a scandal, and the story is a wheel. The trajectory reading reads the middle column. In each phase the architecture answers the previous phase’s verification question by relocating it. Phase One answered the paper era’s chaos by abolishing the independent record, and thereby created the question of the sealed memory. Phase Two answered that question by printing a record, and thereby created the question of the sample, because a record that exists everywhere but is counted almost nowhere has an ambiguous status that litigants were bound to test. Phase Three has answered the sample question, twice, with numbers chosen by the Supreme Court, five stations and five percent, and has thereby created the forensic question of the burnt memory, which the 2024 judgment converted into a candidate’s purchasable right of inspection. A story in which each settlement manufactures the next question is not well described as either progress or scandal. It is well described as an architecture arguing with itself in public, through the only forums available to it: committee rooms, demonstration halls and courtrooms.

What the Voter Sees: Seven Seconds Behind Glass

The VVPAT slip is the only object in the entire architecture that the voter personally inspects, and the design of the inspection deserves the scrutiny the brief’s framework demands, because the seven seconds behind glass carry the whole theory of the paper trail. The voter presses the button against a candidate’s name and symbol on the ballot unit. The printer unit, connected to the ballot unit and standing in the same compartment, prints the slip: serial number, name, symbol. A lamp illuminates the slip behind a transparent window for about seven seconds, long enough for the voter to read the three fields and compare them with the choice made, and then the slip is cut and falls into a sealed compartment at the base of the printer. The voter leaves with nothing. Every element of that sequence was chosen against an alternative that somebody advocateded and the record rejected. The slip is shown rather than handed over, because a slip the voter carried could be marked, traded or coerced, and the secrecy of the ballot would leak through the paper that was meant to protect the count; the Association for Democratic Reforms’ allied proposals in the 2023 litigation, which included giving the voter physical access to the slip so that the voter might deposit it in a separate box, were refused by the 2024 bench, which held that physical access to slips would be problematic and impractical and would invite misuse, malpractice and disputes. The slip is printed on thermal paper, the medium of the receipt, and petitioners pressed the Court on the legibility of thermal printing over time; the architecture answers with the retention calendar described earlier in this article, in which the slips’ working life is the petition window rather than the archive’s century. And the window shows the slip for seconds rather than minutes, because the compartment must serve a queue of voters at the pace of an Indian polling day, a constraint of scale that recurs at every design decision in this story and that foreign comparisons, built on smaller electorates, systematically underweight.

The deeper point of the seven seconds is epistemic. Before 2013, the voter’s knowledge of the vote ended at the button and the lamp: the machine acknowledged a press, and everything after the press was institutional assertion. The window inserts the voter’s own eyes into the chain between intention and record at exactly one point, the point of creation, and the audit trail exists so that a sample of what the voters collectively saw can later be compared, by other eyes, against what the machines collectively stored. Where the comparison has been run, across the tens of thousands of stations counted since 2019, the eyes and the memory have agreed in every case the record attributes to the machines themselves. The verification question persists not because the comparison keeps failing, but because the system rations the comparison, and rationing is a policy choice that each generation of litigants is entitled to contest afresh. That is the sense in which the slip, a scrap of thermal paper visible for seven seconds, is the most consequential printed object in Indian public life: it is the physical point at which the republic’s largest act of collective decision becomes, briefly and partially, checkable by the people who performed it.

Building the Paper Trail, from Noksen 2013 to the Full VVPAT Election of 2019

The phased introduction the 2013 judgment authorised proceeded constituency by constituency, and its milestones mark the conversion of a courtroom direction into national infrastructure. In the general election of 2014, VVPAT units accompanied the machines in eight parliamentary constituencies chosen as pilots: Lucknow, Gandhinagar, Bangalore South, Chennai Central, Jadavpur, Raipur, Patna Sahib and the Mizoram parliamentary constituency. In the assembly elections of early 2017 the Commission deployed about 53,500 VVPAT units, across thirty-three assembly constituencies in Punjab, six in Manipur, three in Uttarakhand, thirty in Uttar Pradesh and all forty in Goa, which thereby became the first state in India to vote entirely with a paper trail. The Gujarat and Himachal Pradesh elections of late 2017 extended the coverage, and for the seventeenth Lok Sabha election of April and May 2019 every polling station in the country, more than a million of them, carried a VVPAT unit beside the ballot unit. The slip the voter sees is deliberately ephemeral as a personal possession: it is printed behind a glass window, shows the serial number, name and symbol of the candidate chosen, remains visible for about seven seconds, and then drops into a sealed compartment of the unit, so that the voter verifies the record without acquiring it and the paper archive accumulates anonymously, in polling order, inside a box that only a counting officer may open.

Universal coverage changed the politics of the paper trail overnight, because a trail that exists everywhere raises the question of why it is read almost nowhere. The Commission’s protocol after 2013 provided for the physical counting of slips from one randomly selected polling station in each assembly constituency or segment, chosen by a draw of lots in the presence of candidates or their agents on counting day, in addition to counting triggered by any candidate’s statutory request in a disputed case. Civil society litigation pressed on exactly this joint. The organisation Common Cause moved the Supreme Court in 2017 in proceedings against the Union of India seeking the use of paper-trail machines in all elections and a substantially expanded practice of cross-verification, and its petition formed part of the continuous litigation that kept the counting protocol, rather than the machine itself, at the centre of the argument between 2017 and 2019. Political petitions ran in the same channel: in Kamal Nath v Election Commission of India in 2018, the Court declined a plea for random verification in ten percent of polling stations in the Madhya Pradesh assembly election and rested on the reputation the Commission had built as an impartial constitutional authority, and in November 2018 the Court dismissed the Nyaya Bhoomi petition seeking a wholesale return to ballot papers. By the eve of the 2019 general election the terrain of dispute had narrowed to a single variable, the size of the sample, and every actor in the system understood that the number chosen would be read as the judiciary’s own measure of how much verification a free and fair election requires. That narrowing is the signature event of Phase Two: the question had migrated from the existence of the record to the statistics of its audit, from architecture to arithmetic, without ceasing to be a question about trust.

The state elections of late 2017 and 2018 ran the enlarged system at full stretch before the general election arrived. Gujarat and Himachal Pradesh voted with VVPAT in December 2017, and in November and December 2018 the five state elections in Madhya Pradesh, Rajasthan, Chhattisgarh, Telangana and Mizoram were conducted with printer units at every polling station, so that by the time the seventeenth Lok Sabha was called, universal paper trail coverage had already been rehearsed across most of the country’s electoral geography. The rehearsals generated their own friction, recorded here because the 2019 petitioners built their case partly on it: parties complained of printer units failing in polling-day heat and of replacement delays at individual stations, and the Commission answered with replacement figures and the argument that a fleet operated at the scale of a million stations will exhibit failures at rates that are operationally normal and electorally immaterial where the affected votes are recorded and the trail preserved. The Kamal Nath proceedings of 2018, in which the Court declined to order ten percent verification for Madhya Pradesh, showed the judiciary already settling into the posture it would formalise in April 2019: the protocol belonged to the Commission, the Court would police its outer edges, and demands to enlarge the sample would need to show more than unease.

Five Slips: The Verification Ruling of 8 April 2019

The decisive proceeding of the 2019 season was brought by twenty-one opposition parties led by the Andhra Pradesh Chief Minister N. Chandrababu Naidu, and it asked the Court to direct that VVPAT slips be physically counted for at least fifty percent of the machines in every assembly segment before results were declared. The petitioners’ case was statistical and experiential at once: a sample of one station per segment could not generate rational confidence in a constituency of roughly three hundred stations, malfunctions and delays reported in the 2017 and 2018 state elections had corroded party trust in the fleet, and the delay of some days that a fifty percent count would cause was a price the petitioners stated they were willing to accept for a result all sides could believe. The Commission’s answer was also stated at its strongest. It told the Court that a fifty percent manual count would delay results by five to six days, that manual counting reintroduces the very human error and booth-level dispute that the machine was built to abolish, and that it had consulted the Indian Statistical Institute, which had advised that a sample of 479 machines, properly drawn, could yield 99.99 percent accuracy in detecting a wrong result at the constituency level. The bench of Chief Justice Ranjan Gogoi and Justices Deepak Gupta and Sanjiv Khanna delivered its order on 8 April 2019, three days before polling opened, and the order is best stated by its operative holding: the number of machines subjected to physical verification of the paper trail was raised from one to five per assembly constituency or assembly segment, the five to be selected at random under the existing guidelines, and the Court recorded its purpose as securing the greatest degree of accuracy and satisfaction in the electoral process, noting that the increase required neither substantial additional manpower nor substantial delay. The prayer for fifty percent was refused, and a review petition against the order was dismissed on 7 May 2019, while counting was under way.

The 2019 count therefore proceeded on the five-station protocol, and its conduct supplied the factual record on which the next round of litigation would be fought. Across more than twenty thousand polling stations selected by lot, counting officials opened the VVPAT compartments and tallied the printed slips against the control unit figures by hand, constituency by constituency, under the eyes of candidates’ agents. The tallying across the 2019 general election and the state elections that followed it became the Commission’s central empirical exhibit: when the Association for Democratic Reforms returned to the Supreme Court in 2023, the Commission’s affidavit record, recited in the 2024 judgment, stated that 41,629 EVM and VVPAT units had passed through random verification, that more than four crore printed slips, over forty million, had been tallied against electronic counts, and that not one mismatch attributable to the machines had been found, the single recorded discrepancy arising from a presiding officer’s failure to delete mock poll votes before polling began. The petitioners’ reading of the same record was different, and it is marked here as disputed rather than resolved: a verification exercise designed and administered by the institution whose machines are in question, on a sample that institution’s own guidelines fix, cannot by itself settle the public’s entitlement to check, because the objection was never only that the sample might miss an error. The objection was that the power to define the audit had never left the audited institution. That objection, in its strongest form, is about the location of the verification power, and it survived the arithmetic of 2019 intact.

The Arithmetic of Assurance: Samples, Software Independence and the Audit Idea

Beneath the courtroom exchange of 2019 sat a genuine intellectual disagreement that neither side’s slogans captured, and naming it precisely is the fastest way to see why the sample question cannot be settled by arithmetic alone. In 2006 the computer scientist Ronald Rivest, writing with John Wack for the American National Institute of Standards and Technology, gave the field its organising concept: software independence. A voting system is software-independent, on their definition, if an undetected change or error in its software cannot cause an undetectable change in the election outcome, which is a property not of the device’s quality but of the system’s design, since it specifies what evidence survives a failure of the very component that records the vote. A paperless machine is software-dependent by construction: if its programme miscounted, nothing outside the programme could show it. A machine with a voter-verified paper record can be software-independent in design, but only to the extent that the paper is actually checked, which converts Rivest’s design question into a sampling question. Philip B. Stark of the University of California, Berkeley, supplied the sampling tradition’s strongest form from 2008 onward in his work on risk-limiting audits: choose the sample by a stated statistical rule, count the paper by hand, and keep enlarging the count until either the reported outcome is confirmed to a pre-announced risk limit or a full recount replaces the reported outcome altogether.

Measured against that tradition, both Indian positions of 2019 were partial. The Indian Statistical Institute’s advice, that a sample of 479 machines could yield 99.99 percent accuracy, answers a detection question at national scale: how many units must be checked to be confident that the fleet as a whole is not producing wrong outcomes. It does not answer the constituency question a losing candidate actually asks, which is whether the result in one segment, among roughly three hundred polling stations, was right, because detection power spread across four thousand segments is thin at any single segment where a close result turns on a handful of stations. The petitioners’ fifty percent demand answered the constituency question intuitively but without a stopping rule or a risk limit, at a cost in delay that the Commission priced at five to six days, and with the unexamined cost that a hand count of half the slips reimports human counting error on a massive scale. The Supreme Court’s five stations were therefore neither the statisticians’ optimum nor the petitioners’ assurance. Five is a confidence-generating number, chosen to be visibly more than one and operationally less than a recount, and the order’s own language, the greatest degree of accuracy and satisfaction, pairs the two nouns honestly: accuracy is a statistical property, satisfaction is a political one, and the protocol was designed to produce both in a single exercise. Sandeep Shastri’s essays on the election process treat that pairing as the Commission’s normal working condition, an institution that must be right and must be believed to be right, with the second obligation not reducible to the first. The trajectory reading adopts the pairing as its own conclusion about Phase Three. The sample is where the verification question became a number, and a number, unlike a principle, can always be argued with; which is why the number has been litigated in 2018, 2019 and 2023, and why no judgment fixing it has yet made the argument stop.

The ADR Judgment of 26 April 2024: Deference with Two New Doors

The Association for Democratic Reforms filed its writ petition in 2023 asking the Supreme Court to go where the 2019 bench had refused to go: to direct cross-verification of the electronic count against the paper slips in every polling station, or at least a substantially enlarged count, and allied petitions sought a return to the ballot paper itself. Argument in the spring of 2024 ranged across the whole trajectory this article has traced, from the design of the microcontroller to the German precedent, and the bench of Justice Sanjiv Khanna and Justice Dipankar Datta delivered judgment on 26 April 2024, in the middle of the eighteenth Lok Sabha election, for which polling had begun on 19 April. The judgment must again be stated by its holdings, because its structure is a refusal joined to two grants. The Court refused to direct one hundred percent counting of VVPAT slips, holding that the data and the record of verifications since 2019 disclosed no ground to enlarge the five-station protocol. It refused to order any return to ballot papers, holding that the machine system had eliminated booth capturing and bogus voting and that electoral reform would be undone by reintroducing paper as the instrument of voting. It rejected the suspicion of manipulation of the burnt firmware as unfounded, recording the Commission’s evidence that the programme is burnt once and cannot be modified, that the firmware knows nothing of candidates or parties, and that the expert-vetted M3 fleet operates inside layered technical and administrative safeguards. In a separate concurring opinion, Justice Datta framed the matter as one of trust between the pillars of democracy, warning that blind distrust corrodes the institutions it claims to defend, while insisting on a critical and constructive approach to improvement of the system.

The two grants were precise, and both operate on parts of the machine system that no counting protocol touches. First, the symbol loading unit, the device used before polling to load candidate symbols into the VVPAT units of a constituency, was directed to be sealed on completion of the loading process undertaken on or after 1 May 2024, signed across the seal by candidates or their representatives, and stored in the strong room with the machines for at least forty-five days after the declaration of results, to be opened and examined as the machines themselves are. Second, the burnt memory and microcontroller of five percent of the machines, control unit, ballot unit and VVPAT together, in each assembly constituency or segment, was directed to be open to checking and verification by a team of engineers from the manufacturers, after the announcement of results, on the written request of the candidate placed second or third behind the winner, the request to be made within seven days of the declaration, the candidate to identify the machines by polling station or serial number and to pay the cost of verification, with the expense refunded if a machine is found to have been tampered with, and the District Election Officer to certify the authenticity of the memory after the check. The bench also suggested, without directing, that the Commission examine a machine for counting paper slips and the printing of a bar code on each slip. Read architecturally, the 2024 judgment does something subtle to the verification question. It declines to enlarge the audit of the count, and simultaneously creates a forensic right over the instrument, exercisable by the two candidates with the most direct interest in the result, at their own expense and under a refund rule that prices frivolous suspicion. Verification after April 2024 is therefore no longer only a sample drawn by the institution. It is also, for the first time in the story, a remedy that a named loser can invoke against named machines.

Two features of the second direction repay attention, because together they mark the first occasion on which the architecture has priced verification and sold it to the person with the strongest motive to use it. The remedy belongs to the candidates placed second and third, and to no one else: not to the press, not to a voter, not to a party organisation as such, and not to the winner. Standing to inspect the memory follows the result sheet, which ties the forensic right to a demonstrable stake in the specific count and screens out the general curiosity that the Commission has always treated as a security risk in itself. The cost rule performs the same screening in money. The requesting candidate pays the notified expense of the manufacturers’ engineering team, and recovers it only if a machine is found to have been tampered with, a refund condition that converts the inspection into a wager: a candidate who genuinely believes a named machine in a named station was altered can test the belief at a price that a vindicated suspicion makes free, while a candidate performing suspicion for an audience must pay for the performance. Whether the wager is priced fairly, and whether seven days is time enough for a defeated candidate to identify the stations worth testing, are questions the direction leaves to experience, and they are recorded here as open rather than answered. What the direction settles is the principle, new in this story, that the memory of the machine is not beyond the reach of the count’s losers. For thirty years the control unit’s memory could be read authoritatively only by the institution that owned the machine. After 26 April 2024 it can also be read, on conditions, at the instance of the two people the count most directly disappointed, and that transfer of a fragment of the verification power, from the audited institution to the audited result’s named victims, is the judgment’s most architectural act.

The judgment’s afterlife illustrates the relocation dynamic in miniature, and it is recorded here as protocol development continuing past the verdict. The Commission framed standard operating procedures for the burnt-memory verification, issuing them in June and July 2024, and the Association for Democratic Reforms returned to the Court contending that the procedures, as first framed, contemplated clearing or reloading data on machines under verification, which in its submission would destroy the very evidence the exercise existed to inspect. The Court’s response in the ensuing proceedings restated the judgment’s intention in operational terms: the verification was meant to have engineers from the manufacturing firms examine the burnt memory and certify, in the presence of those entitled to attend, that the memory and software had not been tampered with, and it was not a licence to erase or reload polling data in the course of the check. The Commission accepted the recasting, proposing amended procedures under which the manufacturers’ engineers would issue a certificate of intactness after examination. A single paragraph of litigation, conducted after the supposedly settling judgment, thus had to decide what the settling judgment meant at the level of a technician’s hands. Defenders of the 2024 settlement read the episode as ordinary implementation. The trajectory reading records it as the verification question at its smallest scale yet: not how much paper, not whose memory, but whether the act of checking the memory would itself be allowed to alter the memory. No phase of this story has ever produced a settlement so complete that the next procedural document did not reopen it.

The World Compares: Germany, the Netherlands and the Paper Principle

India’s argument has a foreign mirror, and the comparison disciplines both sides of it. The Netherlands, the earliest European adopter of machine voting, withdrew approval from its Nedap machines in 2007 after the campaign group led by Rop Gonggrijp, Wij vertrouwen stemcomputers niet, or We Do Not Trust Voting Computers, demonstrated in 2006 that a machine could be made to miscount and to divulge how a vote was cast, and the Dutch state returned to paper ballots for the elections that followed from 2008. Germany’s Federal Constitutional Court went further as a matter of doctrine. In its judgment of 3 March 2009, on complaints against the use of voting computers in the 2005 Bundestag election, the Second Senate held that the principle of the public nature of elections, derived from Article 38 with Articles 20(1) and 20(2) of the Basic Law, requires that the essential steps of voting and of the determination of the result be examinable by the citizen reliably and without specialist knowledge; the machines used in 2005 failed that test, the ordinance authorising them was unconstitutional, and although the 2005 Bundestag was not dissolved for want of any indication of actual malfunction, machine voting in Germany ended with the judgment. The United States took a third path after its own crisis of confidence in the punch-card count of 2000: federal law from 2002 onward funded electronic machines, state after state then legislated for voter-verifiable paper records after security studies from universities including Princeton and Johns Hopkins exposed direct-recording machines that kept no independent record, and the American settlement, like the Indian one, converged on electronics for the act of voting and paper for the act of auditing.

Ireland supplies the comparison’s melancholy footnote. The Irish government purchased a fleet of Nedap-based machines and used them in pilot areas in 2002, then suspended national rollout while a Commission on Electronic Voting examined the system and reported that it could not certify the machines’ secrecy and accuracy to the standard Irish elections required. The unused fleet sat in storage for a decade at continuing cost, a standing embarrassment to successive governments, until the state disposed of the machines for scrap. The Irish episode is the purest case in the comparative record of a verification failure that was never about proven error: no Irish result was shown to be wrong, and the machines died anyway, because a democracy that could not explain to its own voters how a result could be checked chose to warehouse the technology rather than defend it. India’s institutions, facing the same structure of doubt, chose the opposite strategy at every fork: defend the machine, stage the demonstration, accept the court’s augmentation, enlarge the sample, open the memory. The two strategies frame the range of democratic answers to a single question, and the fact that the question bankrupted one fleet and built another is the comparison’s real lesson.

The comparison cuts against lazy uses in both directions. Critics of the Indian system cite Germany for the proposition that courts committed to open elections must distrust machines, but the German standard, examinability by the citizen without specialist knowledge, is one that India’s paper trail was expressly designed to approach and that Germany applied to machines with no paper record at all, and the Indian Supreme Court in 2024 had the German reasoning before it and reached a different institutional conclusion on a different technical record. Defenders of the Indian system cite the same comparison in reverse, noting that the Netherlands and Germany could retreat to paper because their booths had never known organised capturing on the Indian scale, and that a return to paper in India would restore the physical vulnerabilities that the paper era’s record, from stuffed boxes to captured booths, documents across the 1970s, 1980s and 1990s. The trajectory reading takes from the comparison the one proposition all three democracies converge on, beneath their opposite institutional choices: the legitimacy of a voting system is decided by the quality of the public check on the count, not by the elegance of the recording device, and every durable settlement, whether paper-only in Germany, machine-only briefly in the Netherlands, or machine-plus-paper in India, has had to purchase that check explicitly and defend its price in public.

Three Framings, One Question: Adjudicating the EVM Story

The named disagreement of the brief can be decided on the record assembled here. The technological-progress framing, held in its strongest form by the Commission’s institutional histories and by the celebratory strand of commentary after 2004, holds that the machine solved the documented pathologies of paper, that each later addition was refinement, and that the verification story is one of steady completion ending in a system that is, in the Commission’s phrase, non-tamperable within its safeguards. That framing fails its own evidence at one point: it cannot explain why the institution that had declared the question closed in 2009 was directed by the Supreme Court in 2013 to build a paper apparatus it had not planned, nor why the Court spent two further judgments, in 2019 and 2024, fixing the size and the forensic reach of the check. A system whose completeness must be judicially decreed three times in eleven years was not complete. The controversy framing, held in its strongest form by the security researchers of 2010 and by the litigation campaigns that followed each adverse result, holds that the machine era is a sequence of exposed weaknesses answered by institutional denial, and that trust in the count is misplaced until the public can check the whole of it. That framing fails its evidence at the opposite point: across more than forty million slips tallied in verification exercises recited by the Court in 2024, no machine-attributable mismatch has been produced, the two staged public challenges of 2009 and 2017 ended without a demonstrated tampering under any conditions the challengers accepted, and the framing cannot explain why the system’s harshest institutional critic inside the process, the Supreme Court of the Swamy judgment, chose augmentation of the machine over abolition of it when abolition was squarely requested again in 2024.

The trajectory reading adjudicates between them by keeping the unit of analysis architectural. Within the institution-under-stress framework, the Commission appears in this story neither as a triumphant moderniser nor as a defensive monopolist, but as an institution whose authority is repeatedly tested at the joint where its technology meets the public’s entitlement to check, and the integrated three-tensions framework locates that test inside the standing tensions of the republic between central direction and dispersed control, and between speed of result and depth of consent. The scholarly record divides along the same seam, and adjudicating it is part of the brief the framings set. The security literature, of which the Wolchok and Halderman study of 2010 is the Indian exemplar, evaluates a voting system by its worst case under adversarial conditions, and on that standard a closed device with a rationed audit can never be pronounced sound, only unexamined at the points where examination was refused. The institutional literature, in which Sandeep Shastri’s essays on the election process and E. Sridharan’s essays on institutions under strain are the Indian exemplars the brief names, evaluates the same system by its performance across the largest elections ever administered, and on that standard a fleet that has counted billions of votes, survived the transfer of power between rival coalitions in 2004 and 2014, and produced no verified case of machine-attributable miscounting has earned a presumption that its critics must displace with evidence rather than with possibility. The trajectory reading holds that both standards are honestly held and that neither is complete, because the worst case standard cannot price the chaos of the paper alternative it implicitly recommends, and the performance standard cannot answer the citizen who asks not whether the system has worked but how she would know if it stopped working. The verification question is the name this article gives to the gap between those two standards, and the forty-two year record is the history of Indian institutions negotiating the gap without ever closing it: Parliament in 1989, the committee rooms of the 1990s, the demonstration hall of 2009, the laboratories of 2010, the Supreme Court in 2013, 2019 and 2024, and the counting halls, five stations at a time, in every election since.

The strongest counter to the trajectory reading must be faced at full strength: the 2024 judgment’s deference to the Commission’s design, its refusal to enlarge the sample and its language of institutional trust can be read as a settlement that ends the verification question rather than relocating it, in which case the question’s persistence is an artefact of this article’s framing and not a property of the system. The record since the judgment answers the counter only in part, and the answer is marked here as argument rather than as fact. The judgment itself created two new instruments, the sealed symbol loading unit and the burnt-memory check, whose operation in disputed elections will generate fresh protocol questions about cost, access and the consequences of an adverse certificate, and the Commission’s 2024 suggestion docket, machine counting of slips and bar-coded slips, remains open business that any future design generation will reactivate. A question that the settling judgment itself equips with new instruments has not been closed by the settlement. It has been handed to the next phase of the architecture, whose outline the trajectory map already predicts: the machine will change again, the check will be renegotiated again, and the argument of 1982, that the form of the record decides who may verify the result, will be heard again in whatever vocabulary the next technology supplies.

Frequently Asked Questions

Q: When did India first use EVMs?

India first used electronic voting machines on 19 May 1982, in the by-election to the Parur assembly constituency in Kerala, where the Election Commission directed machine voting in fifty of the constituency’s eighty-four polling stations while the remaining thirty-four voted on paper ballots. The experiment grew out of a prototype the Commission had demonstrated to political party leaders on 6 August 1980 and development work assigned to the Electronics Corporation of India at Hyderabad. The trial ended in litigation: the losing candidate, A. C. Jose, challenged the result, and on 5 March 1984 the Supreme Court set aside the poll in the fifty machine stations because no statute authorised machine voting, ordering a fresh poll there by ballot paper. Statutory use began only after Parliament inserted Section 61A into the Representation of the People Act 1951 with effect from 15 March 1989, and the machines returned to polling stations in November 1998.

Q: What is VVPAT?

VVPAT, the voter-verified paper audit trail, is a printer unit attached to an electronic voting machine that produces an independent paper record of each vote at the moment it is cast. When a voter presses a button on the ballot unit, the VVPAT prints a slip bearing the serial number, name and election symbol of the candidate chosen, displays it behind a glass window for about seven seconds so the voter can confirm that the record matches the choice, and then drops the slip into a sealed compartment inside the unit. The voter cannot take the slip away, so the paper archive remains anonymous and accumulates in polling order. On counting day, the slips of randomly selected polling stations are counted by hand and tallied against the control unit’s electronic figures, which converts the printed archive into an audit instrument independent of the machine’s memory.

Q: What was the Subramanian Swamy verdict 2013?

In Subramanian Swamy v Election Commission of India, decided on 8 October 2013 by a bench of Chief Justice P. Sathasivam and Justice Ranjan Gogoi, the Supreme Court held that a paper trail is an indispensable requirement of free and fair elections and that voter confidence in electronic voting machines could be achieved only with its introduction. The Court held that machines fitted with the VVPAT system ensure the accuracy of the voting system, directed the Election Commission to introduce VVPAT in gradual stages or area by area in the ensuing general elections, and directed the Government of India to provide the financial assistance needed for procurement, the Commission having sought sanction for twenty thousand units costing about Rs 38.01 crore. The verdict completed rather than displaced machine voting: it ordered augmentation of the EVM with a paper shadow and left the choice of areas and booths to the Commission.

Q: Are EVMs hackable?

The question divides into a laboratory answer and an election answer, and the two are disputed between named holders rather than settled. In the laboratory, a peer-reviewed study by Scott Wolchok, J. Alex Halderman, Hari K. Prasad and colleagues, presented in October 2010, demonstrated two attacks on a genuine control unit obtained from official storage: a replacement display board with a radio that could alter totals on instruction from a mobile phone, and a clip-on device that could rewrite stored votes between polling and counting. Both attacks required physical access to the machine. The Election Commission’s answer is that its machines are standalone devices with no network connection, that their programme is burnt once into a one-time programmable chip that cannot be modified, that two-stage randomisation, sealing, mock polls and agent oversight govern custody, and that in the staged challenges of August 2009 and June 2017 no participant demonstrated tampering. The Supreme Court in April 2024 rejected the suspicion of firmware manipulation as unfounded on the record before it.

Q: How does VVPAT verification work?

VVPAT verification is the physical counting, by hand, of the printed slips stored in the VVPAT units of randomly selected polling stations, tallied against the electronic count of the paired control unit. Under the protocol fixed by the Supreme Court’s order of 8 April 2019, five polling stations in every assembly constituency or assembly segment are selected by a draw of lots in the presence of candidates or their agents on counting day, and counting officials open the sealed slip compartments of those stations and count the slips candidate by candidate at a dedicated table. The selection is random so that no one can predict which stations will be audited, and the exercise covers more than twenty thousand polling stations in a general election. Across the verifications conducted from 2019 onward, the Commission’s record recited by the Supreme Court in 2024 reported 41,629 units verified and more than four crore slips tallied without a machine-attributable mismatch.

Q: What did the 2024 ADR verdict say?

In Association for Democratic Reforms v Election Commission of India, decided on 26 April 2024 by Justices Sanjiv Khanna and Dipankar Datta, the Supreme Court dismissed the petitions seeking one hundred percent cross-verification of electronic counts against VVPAT slips and a return to ballot papers, upholding the existing protocol of physically counting slips from five randomly selected polling stations per assembly segment. The Court issued two directions. Symbol loading units must be sealed after symbol loading undertaken on or after 1 May 2024, signed across the seal by candidates or their representatives, and stored with the machines for at least forty-five days after results. And the burnt memory of five percent of the machines in each constituency must be open to checking by the manufacturers’ engineers on the written request, within seven days of the result, of the candidates placed second or third, at the requesting candidate’s cost, refundable if tampering is found.

Q: When did India go fully EVM?

India went fully electronic in the fourteenth Lok Sabha election of April and May 2004, the first general election in which every parliamentary constituency voted on electronic voting machines. Chief Election Commissioner T. S. Krishnamurthy announced a deployment of about 1.08 million machines for an electorate of roughly 675 million, voting in four phases between 20 April and 10 May 2004. The step followed a graduated expansion: sixteen assembly constituencies in Madhya Pradesh, Rajasthan and Delhi in November 1998, forty-five parliamentary constituencies in the 1999 general election, the full Haryana assembly in February 2000, and complete assembly elections in Tamil Nadu, Kerala, Puducherry and West Bengal in May 2001. A Commission direction of May 2003 settled that subsequent general elections would be held on machines throughout, closing the transition that the Parur trial had opened twenty-two years earlier on an insecure legal footing.

Q: Which constituency first used VVPAT in September 2013?

The Noksen assembly constituency in Nagaland, constituency 51-Noksen (ST), was the first in India to vote with VVPAT, at its by-election on 4 September 2013, when the printer units were used in all twenty-one of the constituency’s polling stations. The deployment came after field trials of the prototype at Thiruvananthapuram, Delhi, Cherrapunji, Jaisalmer and Leh in July 2011, a second trial in July and August 2012, approval of the final model by the Technical Expert Committee on 19 February 2013, and an amendment of the Conduct of Elections Rules 1961 notified on 14 August 2013 permitting the units’ use. The Supreme Court’s Swamy judgment of 8 October 2013 recorded the Noksen deployment, on the Commission’s affidavit, as a successful demonstration of the system, five weeks before directing that the paper trail be introduced across the country in gradual stages.

Q: What is a symbol loading unit, and why did the 2024 judgment order it sealed?

A symbol loading unit, or SLU, is the device used before polling to load the names and symbols of a constituency’s candidates into the VVPAT units, so that the printer can print the correct slip for each button on the ballot unit. Because the loading process is the one moment at which constituency content enters the machine system, petitioners in the 2023 litigation argued that the units were an unguarded joint in the architecture. In its judgment of 26 April 2024 the Supreme Court directed that, on completion of symbol loading undertaken on or after 1 May 2024, the units be sealed in containers signed across the seal by candidates or their representatives, and kept in the strong room along with the voting machines for at least forty-five days after the declaration of results, to be opened, examined and dealt with in the same manner as the machines themselves.

Q: What happens when the slip count and the control unit count disagree at a verified polling station?

A disagreement at a verified polling station sets in motion a recount of the printed slips for that station, conducted again by hand in the presence of candidates or their counting agents, and the returning officer records the outcome in the result documentation for the station before the constituency result is compiled. The single discrepancy recited in the Supreme Court’s 2024 judgment illustrates the failure mode the protocol is built to isolate: it arose because a presiding officer had failed to delete the votes cast in the mock poll before polling began, so the control unit carried test votes into its total, and the paper count exposed the error. What wider consequence a persistent, unexplained mismatch should trigger, whether counting should expand to more stations in that constituency, is a question the counting protocol does not answer in terms, and petitioners pressed it in the litigation of 2023; the 2024 bench left the consequence within the Commission’s protocol.

Q: Why did Germany’s Constitutional Court rule against voting computers in 2009?

Germany’s Federal Constitutional Court ruled on 3 March 2009, in complaints against the use of Nedap voting computers in the 2005 Bundestag election, that the governing ordinance was unconstitutional. The Second Senate derived from the principle of the public nature of elections, in Article 38 read with Articles 20(1) and 20(2) of the Basic Law, a requirement that the essential steps of voting and of the determination of the result must be examinable by the citizen reliably and without any specialist knowledge of the subject. Machines that recorded and totalled votes in a form no voter or observer could inspect failed that requirement, because the public check on the count had been delegated to the device and its manufacturer. The Court did not dissolve the 2005 Bundestag, finding no indication that the machines had malfunctioned or been manipulated, but machine voting in Germany ended with the judgment, and the country returned to paper ballots counted in public view.

Q: What did the 2010 security study of India’s machines demonstrate?

The study, Security Analysis of India’s Electronic Voting Machines, authored by Scott Wolchok, Eric Wustrow, J. Alex Halderman, Hari K. Prasad and four co-authors and presented at the ACM Conference on Computer and Communications Security in October 2010, was the first independent security examination of a genuine Indian control unit, supplied to Prasad by an anonymous source after the Commission declined to release a machine for research. Working on the real device, the team demonstrated two attacks. A replacement display board fitted with a Bluetooth radio could receive instructions from a mobile phone and silently alter the totals displayed and stored. A pocket-sized clip-on device attached to the memory chip could rewrite the stored votes after polling and before counting. The authors’ stated conclusion was conditional: both attacks needed physical access, and the exposure they proved was that the system’s safety rested on custody procedures and design secrecy rather than on properties of the device itself.

Q: What role did the all party meeting of 4 October 2010 play in the paper trail story?

At the all-party meeting of 4 October 2010, political parties pressed the Election Commission to explore a voter-verified paper audit trail for the voting machines, and the Commission acted on the suggestion by assigning the design of a printer unit to its two manufacturers, Bharat Electronics and the Electronics Corporation of India, under the guidance of its Technical Expert Committee. The meeting matters to the sequence of the story because it shows the paper trail entering the Commission’s own engineering programme from the parties’ side of the table, a full year before the Delhi High Court’s observations in the Swamy litigation and three years before the Supreme Court’s direction of October 2013. When the Court later mandated VVPAT, it was accelerating a design process the institution had already begun, field trials in 2011 and 2012 included, rather than imposing a technology the institution had refused to consider.

Q: What is the Technical Expert Committee, and which designs must it approve?

The Technical Expert Committee is the standing body of engineers and academics, drawn principally from the Indian Institutes of Technology, that the Election Commission constituted to evaluate the voting machines’ designs and to approve changes to them before deployment. Its approval operates as a gate in the system’s development: the final model of the VVPAT printer unit received the Committee’s approval on 19 February 2013 before the Commission demonstrated it to political parties on 10 May 2013 and used it at Noksen that September, and the M3 generation of machines described in the Supreme Court’s 2024 judgment was vetted by the Committee before entering the fleet. The Committee’s existence is the institutional core of the Commission’s design defence, since every hardware generation in the story passed through expert scrutiny outside the two manufacturing firms, while critics answer that the scrutiny remains internal to the state and is no substitute for examination by independent researchers.

Q: How many machines did the fully electronic general election of 2004 deploy?

The general election of 2004 deployed about 1.08 million electronic voting machines, a figure announced by Chief Election Commissioner T. S. Krishnamurthy for an electorate of roughly 675 million voting in four phases between 20 April and 10 May 2004 across all 543 parliamentary constituencies. The scale is the fact that distinguishes the Indian machine programme from every European experiment with which it is compared: the Netherlands and Germany each used machines in the low thousands of polling places, while India’s 2004 deployment put a machine in effectively every polling station of a continental electorate, with the stock manufactured by two public sector firms at Bangalore and Hyderabad. That scale is also why the Indian debate about verification took a statistical form, since a physical check of the whole paper record, once a paper record existed, would mean counting slips at more than a million stations.

Q: Which state voted entirely with VVPAT for the first time in 2017?

Goa was the first state in India to vote entirely with VVPAT, at its assembly election of February 2017, when printer units accompanied the machines in all forty of the state’s assembly constituencies. In the same round of state elections the Commission deployed about 53,500 VVPAT units in selected constituencies elsewhere, thirty-three in Punjab, six in Manipur, three in Uttarakhand and thirty in Uttar Pradesh, so Goa’s complete coverage stood out as the proof of concept for universal deployment. The state’s small size made it the natural test bed for the logistics of a full paper trail election: procurement, transport, training of polling staff, and the counting-day handling of sealed slip compartments in every station. Two years later the seventeenth Lok Sabha election of 2019 extended the Goa pattern to the entire country, with a VVPAT unit at every polling station in India.

Q: What happened at the Election Commission’s EVM challenge of June 2017?

The EVM challenge of 3 June 2017 was the Commission’s second staged public test of its machines, announced by Chief Election Commissioner Nasim Zaidi on 20 May 2017 after several parties alleged tampering in the state elections of February and March that year. All recognised national and state parties were invited to attempt tampering with machines drawn from strong rooms in Punjab, Uttarakhand and Uttar Pradesh, under the challenge framework. Only two parties, the Nationalist Congress Party and the Communist Party of India (Marxist), registered to participate. Fourteen machines were produced, and on the day neither party attempted to tamper with a machine: the NCP team stated an academic interest in the process, and the CPI(M) team sought and received a detailed demonstration instead. Zaidi declared the tamperability question closed, while the CPI(M) afterwards described the exercise as too restrictive to settle it.

Q: What did the Delhi High Court say in the Swamy litigation in January 2012?

On 17 January 2012 the Delhi High Court disposed of Subramanian Swamy’s writ petition seeking a paper trail for the voting machines by declining to issue the direction he sought. The Court held that it could not order a return to paper ballots, nor dictate the Election Commission’s choice of voting technology, and it left the development of a paper trail to the Commission in consultation with political parties. Two features of the judgment carried the case forward on appeal. The Court recorded that the machines in their existing form could not be described as tamper-proof in an absolute sense, an observation that deprived the Commission of the strongest version of its design defence. And the Court’s expectation that the Commission engage seriously with the paper trail proposal framed the question that the Supreme Court answered differently in October 2013, when it converted the High Court’s expectation into a constitutional direction and held the paper trail indispensable.

Q: What is a mock poll, and why is it conducted before polling begins?

A mock poll is the test vote conducted on the morning of polling, before the station opens to electors, in which polling agents of the candidates cast a block of test votes on the machine, ordinarily at least fifty, and the presiding officer then displays the recorded result, confirms that the machine has counted the test votes correctly for each candidate, clears the memory to zero in the agents’ presence, and seals the control unit for the real poll, with a mock poll certificate signed by the presiding officer and the agents present. The ceremony is the first link in the chain of custody that the Commission defends as half of the machine’s security, because it proves to the candidates’ own representatives, station by station, that the device begins the day empty and counting accurately. Its failure mode is equally instructive: the one discrepancy recited in the 2024 judgment arose because mock poll votes at a station had not been deleted before the poll began.

Q: Why did the 2019 petitioners demand counting of slips from half the polling stations in each segment?

The twenty-one opposition parties led by N. Chandrababu Naidu asked the Supreme Court in 2019 to direct physical counting of VVPAT slips from at least fifty percent of the machines in every assembly segment, and their stated reason was confidence rather than arithmetic alone: a single randomly selected station among roughly three hundred could not, in their submission, generate rational assurance about the segment’s count, and reported malfunctions in the elections of 2017 and 2018 had left parties and voters doubting results they were asked to accept on the institution’s word. The petitioners told the Court they were willing to accept a delay of days in the declaration of results as the price of a count every side could believe. The Commission answered that a fifty percent count would delay results by five to six days and reintroduce manual error, and cited Indian Statistical Institute advice that a sample of 479 machines could yield 99.99 percent accuracy; the Court’s order of 8 April 2019 fixed the sample at five stations per segment.