Examining the polling data by drilling it down to constituency and booth levels reveals troubling trends in the recently concluded Assembly polls in West Bengal. The outcome is highly questionable for many visible reasons: large-scale deletions through the Special Intensive Revision (SIR) exercise; biased implementation of the Model Code of Conduct; the unusually large and active presence of paramilitary forces, the unprecedented takeover of the State administration by Election Commission of India (ECI)-appointed officers; and disturbing incidents reported during polling as well as counting—all of which are in the open for everyone to see.
But there is one factor that is not obvious, which I mentioned in Part I of this investigation, which is the unreal pace of voting in a large number of booths. The high rate of polling reported and the unusually rapid pace that would have been required to make such polling possible raise troubling questions about the poll process in West Bengal.
Let us take two constituencies, one from each phase: Nandigram from Phase 1 and Bhabanipur from Phase 2.
The data used to build the following scenarios is taken from the ECI and West Bengal Chief Electoral Officer’s (WB CEO) portals and/or from media sources that cite ECI or WB CEO sources. The booth-level vote totals used for the latter part of the analysis are taken from Form 20 data as available on the WB CEO website.
This analysis is a feasibility stress test. It uses publicly reported State-level turnout percentages and applies them to booth-level Form 20 totals to test whether the implied pace of polling stands within the limits of physical feasibility.
If and when ECI releases booth-wise time-stamped polling data and it turns out to be at variance with the data used here, it must be emphasised that it is unlikely to alter the present inferences. If anything, it could only turn out to be a much worse scenario than what is mapped here. If the ECI data were to reveal more than the extrapolated percentage, then the pace of polling would have to be even faster than the already unreal pace. And if it turns out to be less than the extrapolated percentage, then it would be that much more difficult to account for the declared quantum of voting within the actual hours of polling.
If the Election Commission has booth-wise time-stamped data, it must place it in the public domain for public scrutiny and subject it to feasibility stress test.
The first polling slot is from 7 am to 9 am. The reported State-level turnout by 9 am of 18.76 per cent would mean that around 155 voters had voted by 9 am in an average booth in Nandigram. That is in 120 minutes, or 7,200 seconds. If each voter took only 45 seconds to cast her vote and provided downtime was a highly unlikely zero, then 160 votes could have been polled in that time slot.
This is the first warning sign. Nandigram’s reporter: 155 votes cast by 9 am sits close to the absolute ceiling only if we assume each voter took 45 seconds with zero downtime. But if voters took 60 seconds, only 120 votes could be possible. If she took 75 seconds, only 96 votes were possible.
As the time taken rises, polling reduces progressively, even with zero downtime. But if the time taken per vote rose and/or downtime increases, the discrepancy between the feasible and implied number of votes would progressively increase.
In the real world, a polling booth queue does not move like a stopwatch. There are elderly voters, ID checks, pauses, VVPAT wait time, malfunctions or hitches, officer and people movement, confusion, human fatigue, and other ordinary delays.
Let us see the scenario as of 1 pm. By that time, the State-wise aggregate polling was 62.18 per cent. That brings cumulative voting to around 514 votes in an average booth in Nandigram.
However, from 7 am to 1 pm, even given the tight 45 seconds per voter and zero downtime, only 480 votes are feasible, which is below the implied number of 514. At 60 seconds per voter, the feasibility becomes much lower, with the gap between officially implied polling and physically feasible polling growing as time per voter increases or even a minimal downtime is factored in. With 20 per cent downtime and a more realistic rate of 75 seconds per vote, only 77 votes would have been possible.
A man leaves after casting his vote during Phase 1 voting in Nandigram, West Bengal, on April 23, 2026.
| Photo Credit:
SHASHI SHEKHAR KASHYAP
It is this gap between implied voting rate and feasible voting rate that makes the announced polling numbers highly questionable and makes one wonder where all these votes came from.
Let us look at Bhabanipur.
The Bhabanipur numbers hold to the State-wise average as long as we give the same unlikely 45 seconds per vote and add no downtime. But it too begins to coast towards improbability when the implied turnout is examined against booth-level Form 20 data.
That is where the mystery begins to deepen.
Based on Form 20 data, in Bhabanipur, there are 37 High Load booths and 17 Very High Load booths. In the High Load category, average polling was 778 votes, with the lowest at 701 and the highest at 880. In the Very High Load category, the average was 955 votes, with the lowest at 901 and the highest at 1009. Bhabanipur had no Extremely High Load booths.
In Nandigram, there are 84 High Load booths and 68 Very High Load booths. There are also 13 Extreme High Load booths, where the minimum number of votes polled is 1100. This is important because the higher the final booth load, the sharper the anomaly shows up when early turnout percentages are applied to those booths.
Take, for example, booth No. 21 in Bhabanipur constituency, a High Load booth. The total polling according to Form 20 was 999. Between 7 and 9 am, in the highly unlikely scenario of a voter casting her vote in just 45 seconds and with no downtime, the feasible votes that can be cast is 160 votes. Applying the reported 9 am turnout rate of 18.39 per cent to this booth’s Form 20 polling numbers, the implied voting works out to around 184 votes. That means the implied polling breaches the feasibility threshold by about 24 votes even under the fastest and most generous assumption.
If a voter took 60 seconds instead of 45 seconds, and with no downtime, only 120 votes would have been feasible. The implied 184 votes would then breach the feasibility threshold by about 64 votes. If even a small downtime is added, the gap increases further. A mere 12-minute interruption or slowdown in a two-hour segment reduces effective voting time from 120 minutes to 108 minutes. At 45 seconds per voter, feasible votes fall to 144. At 60 seconds per voter, they fall to 108. Even minor real-world frictions thus change the picture dramatically.
Now let us look at booth No.7 of Nandigram, an Extreme Load booth. The total number of polls according to Form 20 was 1158. Between 7 and 9 am, with no downtime, if a voter took 45 seconds to cast her vote, the feasible number was 160. Applying the reported 9 am turnout rate of 18.76 per cent, the implied polling for this parcel works out to around 217 votes.
The implied polling number breaches the feasible threshold by about 57 votes. To process 217 voters in two hours, the booth would have had to move at nearly 33.14 seconds per voter, a little over half a minute per voter!
This is where the issue stops being mere arithmetic and becomes a test of physical reality.
Imagine one voter entering and leaving every 33 seconds for two straight hours, without a single pause, without one slowdown of any sort, no elderly voter walking slow, no wheelchair voter having to be accommodated, no ID issues, not one EVM delay, not one officer taking a pause, not one voter trying to find her name on the list. That is not an election queue. It is a conveyor belt.
If the voter took 60 seconds to cast the vote instead of 45 seconds, the feasibility threshold comes down to 120 votes. The implied 217 votes would then exceed the feasible number by 97 votes. If those votes were to be processed, the booth could not have moved at 60 seconds per voter. It would still have needed the same extraordinary pace of around 33 seconds per voter, with no interruption.
The same booth gives another warning during the 1 pm to 3 pm parcel. If the implied number of votes in that parcel was 162, then at 45 seconds per voter with zero downtime, the feasible number would have been 160. Even under that unreal assumption, it breaches the threshold by 2 votes. At 60 seconds per voter, feasible votes fall to 120, increasing the breach to 42 votes. To reach 162 votes in that two-hour parcel, the pace would have to be about 44.44 seconds per voter, again with zero downtime.
In both constituencies, we can find many such booths. Between 7 am and 9 am, at the pace of 45 seconds per vote with zero downtime, the implied number of votes breaches the feasible number in 23 booths in Bhabanipur. In Nandigram, the count is 103.
The first row in the above table is already the most generous scenario one can offer to the polling system. It assumes one voter every 45 seconds. It assumes zero downtime. Even then, many booths breach the feasibility threshold.
The second row adds only 12 minutes of pause, slowdown, or interruption in a two-hour slot. That small change increases the number of stress booths sharply. In Bhabanipur, it rises from 23 to 38. In Nandigram, it rises from 103 to 136. Remember, this is the first two hours of polling.
Many constituencies in West Bengal have similar suspect trends.
All these suspect trends become visible when we closely examine the data recorded in Form 20 and compare it with publicly reported time-slot turnout percentages. They are all hiding in plain sight.
Until these discrepancies and irregularities are credibly explained by the ECI, the declared results are bound to remain seriously suspect.
The 2026 Bengal mandate will, therefore, continue to carry a blemish of doubt. This is highly undesirable for the country’s electoral democracy.
The only course of action that erases this blemish of doubt is this: a disclosure of booth-wise time-stamped turnout data, reconciliation of Form 17A with Form 17C, and both of them with Form 20, and putting the presiding officers’ diaries, CCTV records of queues, EVM event logs, and related polling records in the public domain for scrutiny.
Parakala Prabhakar is a political economist and author of The Crooked Timber of New India.
Also Read | Bengal’s violent transition into a new political era
Also Read | Why the BJP’s big win in West Bengal is not a surprise
