An analysis of Delhi’s draft electoral rolls released as part of the special intensive revision (SIR) process showed that the software-driven check for “logical discrepancies” in their records is flagging a disproportionately higher number of Muslims.

Muslims, who form 13.6% of electors on Delhi’s draft roll, made up 20.3% of those flagged for “logical discrepancies”, and 30% of those flagged under five reasons indicative of family size. Who got flagged was decided not by the Electoral Registration Officers (EROs), the statutory custodians of the rolls, but by the Election Commission of India’s (ECI’s) centralised ERONET software.

Delhi’s draft roll numbers

The Hindu digitised Delhi’s draft SIR rolls, which had 97.54 lakh voters, of whom 33.12 lakh were flagged for notices by the ECI, to study patterns. The latter set included two distinct sets of electors: 13.79 lakh were flagged for being “Unmapped with Last SIR” and another 19.33 lakh for “logical discrepancies” based on eight reasons.

Our analysis shows that Muslims accounted for 13.6%, or roughly 13.3 lakh, of the 97.54 lakh electors (refer to the box for methodology). This is not much higher than the 12.9% representation of Muslims in the 2011 Census.

Among the 13.79 lakh who were “Unmapped with Last SIR,” members of the community accounted for only 7.6%, which meant that higher proportion of Muslims than non-Muslims were able to map themselves with the last SIR.

But among the umbrella category of “logical discrepancies” featuring 19.33 lakh electors, Muslims accounted for 20.3%, which is 12.8 percentage points higher than their share among unmapped electors and 6.7 percentage points higher than their share among total electors in the draft roll.

To put it another way, for every 1,000 Muslim electors in the draft roll, 318 were flagged for “logical discrepancies”, while the corresponding figure for the rest of the non-Muslim electors was 197 per 1,000.

This means that Muslims were 60% more likely to be marked for “logical discrepancies” than non-Muslims.

Interestingly, a similar data analysis for Bihar’s draft rolls showed that Muslims accounted for 18.4% of deleted electors, not much higher than their 17.7% share of the adult population according to the 2011 Census. The concept of logical discrepancies was not used during the Bihar SIR.

‘Logical discrepancies'

The eight reasons for flagging under “logical discrepancies” could be broadly classified into four categories. One, name mismatches with self or with parent. Two, five reasons based on family size or children. Third, “parents marked ineligible in the last SIR,” which only had a miniscule number of 183 names. And a strange fourth, “no reasons mentioned”, in which nearly 60,000 electors were flagged. In this category, too, the share of Muslims was disproportionately higher at nearly 20%.

In the first category, though the share of Muslims was higher when compared with their share among electors, the difference was of a lesser magnitude. About 15.5% of all electors flagged for mismatches in names were Muslims, which is 1.9 percentage points higher than their 13.6% share in the draft rolls.

It is in the second category, namely, “progeny >= 6”, “parent age difference >50”, “grandparent age difference <40”, “progeny age gap <9 months”, and “parent age difference <15”, where the proportion of Muslims among those flagged is much higher. These reasons are also indicators of family size.

Of the 6.75 lakh electors who were flagged for one or more of these reasons, 30% were Muslims, which is more than double their 13.6% share in the draft electoral rolls. Of the 1.4 lakh electors flagged with “Progeny >=6” as reason, the share of Muslims went to 58% — over four times higher than their share among electors.

If one were to look at the rate among Muslims and non-Muslims flagged for one or more of these five reasons, it turns out that Muslims were nearly three times more likely to get flagged. The numbers showed that while 152 out of every 1,000 Muslim electors got flagged for these reasons, the same was only 56 for every 1,000 non-Muslim electors.

These patterns indicate that the reasons could have been designed in a manner that predominantly overlapped with the fact that the fertility rate is higher among Muslims. Yet, these higher rates are observed to be consistently falling, as per recent National Family Health Survey (NFHS) reports.

As per NFHS reports, the total fertility rate for Muslim women came down from 4.41 to 2.36 between 1992-93 and 2019-22. The same for Hindu women came down in the same period from 3.3 to 1.94, with the gap between Muslim and Hindu women narrowing.

It is also unclear how the software does these age-specific calculations collated for the discrepancies, considering that a large number of older people in India did not precisely remember their birth date owing to home deliveries being more common earlier.

Even assuming that all electors had accurately mentioned their ages and dates of birth for these calculations to be right, it is unclear whether these reasons per se disqualified a person from being an elector or if it was the ECI’s responsibility to scrutinise these.

Not in the orders

Neither the ECI’s order for SIR in Bihar (dated June 24, 2025) nor its subsequent direction to conduct SIR in 12 more States and Union Territories (U.T.s) as part of Phase 2 (dated October 27, 2025) had any mention of issuing notices for “logical discrepancies” to electors who managed to find a place on the draft SIR rolls by submitting duly filled enumeration forms.

ECI’s directions for Phase 2 had only mentioned that EROs would issue notices to those added to the draft rolls, but could not be linked with the previous SIR rolls — which many States now report as the “unmapped” list of electors.

The ECI introduced the flagging of “logical discrepancies” first in West Bengal, midway during Phase 2 of the SIR, without any formal announcement.

The issue of notices to electors for “discrepancies” found a mention for the first time only in ECI’s direction issued on May 14, 2026 for the ongoing Phase 3 of SIR — in 19 States and U.T.s.

Adding an additional clause, it said, “ERO to issue notice to those persons who could not be linked or there is discrepancy in their linking with the electoral roll of the last revision.”

Till date, the ECI has not explained the reasons for which an elector on the draft SIR rolls could be flagged for “logical discrepancies”.

Moreover, though it directed EROs to issue notices, the decision on who would be flagged was not theirs, but decided by the centralised ERONET software that has now come under scrutiny. This was confirmed by a senior election official who did not want to be identified.

A detailed questionnaire was sent to the ECI, seeking responses on what each of these reasons meant, how they were decided, whether any orders were issued in this regard, were any changes made to the reasons between Phase 2 and Phase 3 of SIR and whether the software used any artificial intelligence or deep learning methods. The ECI did not respond till Tuesday night.

(With inputs from Sreeparna Chakraborty)

Methodology: The Hindu downloaded Delhi’s draft SIR rolls for all ACs, used optical character recognition (OCR) to extract text, and made a database. We downloaded the lists of electors who received notices for different reasons and mapped them with the roll database using EPIC IDs. To estimate religion, we used an open-source classifier built by Rochana Chaturvedi and Sugat Chaturvedi (Political Analysis, 2023). The model breaks names into short sequences of characters and learns which patterns go with each religion, so that it can classify names. We ran it on each elector’s name combined with their relative’s name for better accuracy. The authors trained it on about 1.15 lakh households across 17 States, and in their tests, this combined-name version separated Muslims from non-Muslims with an accuracy of about 97%. Our manual check of a random sample of electors also found this to be accurate. 

Explaining reasons: Interpretation of the “logical discrepancy” reasons, based on conversations with field-level officials. “Progeny >= 6” — six or more electors named the same person in the 2002 rolls as their parent; “Parent age difference >50” and “Parent age difference <15” — the parent was over 50 or under 15 when the child was born. “Grandparent age difference <40” — elector and their grandparent are less than 40 years apart. “Progeny age gap <9 month” — siblings were born less than nine months apart.

Note: A minor share of Muslim and non-Muslim electors flagged for logical discrepancies were marked with more than one reason. The rate and shares are calculated by counting every reason for every flagged elector. Similar trend held even if only the reason mentioned first for every elector is counted.