The Mahatma Gandhi National Rural Employment Guarantee Act is the rare statute whose success can be scored without interpreting anything, because the Act itself supplies the scorecard. It promises a stated number of days of work, within a stated number of days of asking, paid within a stated number of days of working, with a stated allowance if the work does not come and a stated compensation if the wage does not. The MGNREGA implementation record is therefore not, in the first instance, an argument about whether the programme is a good idea. It is a set of measurements against obligations that Parliament wrote in numbers, and the measurements exist in unusual abundance, because the programme runs on a transaction-based management information system that records every muster roll, every wage list and every fund transfer order. Few statutes anywhere generate their own evidence at this scale. The puzzle this article has to solve is that the evidence supports two flatly incompatible public conclusions. The Ministry of Rural Development reports that the overwhelming majority of wage payments are made in time. Independent researchers, working from the programme’s own transaction data, report that the overwhelming majority are not. Both conclusions are drawn from real numbers. They differ because they measure different halves of the same obligation, and the instrument that produces the official figure was built to stop measuring at the exact point where the state-level half of the work ends.

The statute itself, its structure, its drafting and its section-by-section content, belongs to the series’ account of the right to work under the employment guarantee statute, and this article does not retell it. What follows takes the Act as given and asks the implementation questions: what was delivered against each numbered obligation, how the delivery machinery actually moves a wage from a worksite to a bank account, why the recorded figures diverge so sharply from the researched ones, who bears the cost when the machinery fails, and what the courts have been able to do about a compensation entitlement that the measuring system does not compute. The discipline applied throughout is the one this series applies to every enforcement record: every figure carries its source, its period and its base, and no figure from one base is ever set beside a figure from another as though the two were comparable. Most of the confusion in public argument about this programme, and some of the confusion in official argument, is a failure of exactly that discipline.

The yardstick: five obligations written in numbers

An implementation study has to begin by fixing the standard, because a programme judged against expectations will always look different from a programme judged against its statute. The National Rural Employment Guarantee Act, 2005 (Act 42 of 2005), which received assent on 5 September 2005, came into force on 2 February 2006 in two hundred districts, was extended to all rural districts from 1 April 2008 and was renamed for Mahatma Gandhi in 2009, imposes five obligations that can each be tested against a number.

The first is the quantum of work. Section 3 obliges the state government to provide not less than one hundred days of guaranteed wage employment in a financial year to every rural household whose adult members volunteer to do unskilled manual work, and permits up to fifty additional days in areas notified as affected by drought or natural calamity. The hundred days are a household entitlement, not a per-person one, and they are a floor expressed as a guarantee: the statute does not say that the state should aim at one hundred days or provide them on average, but that it shall provide them to a household that demands them.

The second obligation is speed of provision. An application for work, made to the Gram Panchayat or the Programme Officer, carries a dated receipt, and employment must be provided within fifteen days of the application. The dated receipt is not clerical decoration. It is the instrument that starts the clock, proves when the clock started, and founds the third obligation when the clock runs out.

The third obligation is the unemployment allowance under Section 7. Where work is not provided within the fifteen days, the applicant is entitled to a daily unemployment allowance at a rate the Act fixes by reference to the wage rate: not less than one-fourth of the wage rate for the first thirty days of the financial year and not less than one-half of the wage rate for the remaining period. The allowance is the state’s liability, a point Section 22’s funding pattern confirms, and its design is deliberate. A guarantee with no price for non-delivery is a request; the allowance converts non-delivery into a charge on the government that failed to deliver, so that the cheapest course for a state is to open work.

The fourth obligation is speed of payment. Wages are to be disbursed weekly and in any case within fifteen days of the closure of the muster roll on which the work is recorded. The fifth is the price of breaking the fourth: under Schedule II to the Act, a worker whose wages are delayed beyond the sixteenth day after closure of the muster roll is entitled to compensation at the rate of 0.05 per cent of the unpaid wages for each day of delay. The compensation is designed to be automatic rather than discretionary. It does not depend on proving hardship, on the worker’s complaint, or on an officer’s sympathy; it accrues by arithmetic from dates the system itself records, and the state is to pay it upfront after verification and recover it from the functionaries or agencies responsible for the delay, as the Minister of State for Rural Development explained to the Rajya Sabha in December 2022.

The yardstick is wider than the five numbered obligations, and the wider conditions belong in the record because they are the terms on which the work itself is done. Work is ordinarily to be provided within five kilometres of the applicant’s village, and where it is provided beyond that distance, additional wages of ten per cent are payable toward the extra cost of travel and subsistence. Schedule II prescribes the conditions at the worksite: drinking water, rest periods and shade, first aid, and a crèche arrangement where young children are present at the site in sufficient number, with the woman caring for them paid at the programme wage. It prescribes equal wages for men and women, a provision whose delivery is visible in the participation figures discussed below. It also carries the programme’s accident discipline, under which a worker injured in the course of programme work is entitled to medical treatment and to compensation in the event of disablement or death, obligations that fall on the implementing machinery. The implementation literature’s consistent finding on this wider yardstick is that the worksite conditions are the least measured part of the programme: the transaction system counts days, wages and works, but no national series records how many worksites had water, shade or a functioning crèche, so that an entire schedule of entitlements exists in a state of statistical invisibility, honoured or not at the level of the individual worksite and invisible to every dashboard. The pattern recurs throughout this article and is worth naming once here: the programme measures superbly what its systems were built to measure, and the entitlements outside the systems’ design, worksite conditions, unrecorded demand, second-stage delay, share that invisibility in common.

Around these five obligations the Act builds its funding architecture, and the architecture matters to the record because it assigns each failure a debtor. Under Section 22, the Union bears the entire cost of unskilled wages and three-fourths of the material cost, while the state bears the remaining material share and the whole of the unemployment allowance. Delay compensation follows responsibility for the delay. Refusal to accept an application for work or to issue the dated receipt is treated as a contravention under Section 25, punishable with fine. The Act also builds its own verification machinery: Section 17 requires the Gram Sabha to conduct social audits of works, and the programme is planned through an annual labour budget in which districts project demand and the Union approves the projection as the basis of funding. Each of these elements reappears in the record below, because each of them is a place where the record is made, distorted or lost.

How is a day’s pay calculated under the schedule of rates?

A day’s pay is piece-rated, not time-rated. Each state maintains a schedule of rates that prices defined units of work output, digging a stated volume of earth, carrying material a stated distance, so that a worker who completes the standard day’s measured output earns the notified daily wage rate exactly, and a worker who produces less earns proportionally less.

The consequence is that the wage a muster roll records is a measurement before it is a payment. A mate or technical assistant measures the group’s output, the measurement is entered against the muster, and the wage list is generated from the measurement rather than from attendance alone. This is why disputes about wages so often begin as disputes about measurement: a short measurement produces a short wage that no payment system, however fast, can correct, and the worker’s quarrel is then with the measuring chain at the worksite rather than with the paying chain above it. The schedule-of-rates design was inherited from the relief-works tradition and defended on incentive grounds, since it pays for work done rather than presence recorded. Its cost is borne by the slowest workers, the elderly, and those working in hard soil or difficult terrain, for whom the standard output norm may be unattainable in a day, and the programme’s own studies and the audit literature have repeatedly recorded measurements that workers regarded as short. The wage that the rest of this article tracks through the payment system is therefore already, at the moment the muster closes, a contested quantity in a share of cases, and the payment delays analysed below fall on amounts that some workers believed were too small before they were late.

The wage rate itself is fixed outside the statute’s text. Section 6(1) empowers the Union, notwithstanding anything in the Minimum Wages Act, 1948, to notify a wage rate for the programme, subject to a floor of sixty rupees a day that the original text set and that inflation long ago left behind. Until 2008, workers under the programme were paid the state minimum wage for agricultural labour, because the Union had not displaced it. In January 2009 the Union exercised the Section 6 power, froze the rates then in force and delinked the programme wage from state minimum wages; since that delinking, the Union has notified a separate rate for each state for each financial year, revised annually by reference to the Consumer Price Index for Agricultural Labourers. The legal and political consequences of that choice run through the record: the programme’s wage became a Union decision, its level a standing subject of federal negotiation, and its relation to the wages a worker could earn or claim elsewhere a question the implementation sections below take up in the comparison the seed questions demand.

The architecture the statute does not contain

Nothing in the Act describes how a wage actually travels. The Act speaks of disbursement; the machinery of disbursement is administrative, built after enactment and rebuilt several times since, and it is the machinery, not the statute, that determines when money reaches an account. Understanding the record requires understanding that machinery in some detail, because the article’s central finding sits inside it.

In the programme’s first decade, wages moved through state-level arrangements in which funds released by the Union were held by states and districts and paid out through banks, post offices and, in the earliest years, sometimes in cash at the worksite. From April 2016 the programme migrated to the National Electronic Fund Management System, under which the payment chain was centralised and made fully electronic. Under this system, the life of a wage payment runs as follows. Work is done and recorded on a muster roll at the worksite. The muster is closed, the work is measured, and the attendance and measurement data are entered into the programme’s software. A wage list is generated from the muster data, naming each worker, the days worked and the amount due. A fund transfer order, the FTO, is then generated from the wage list: it is the electronic payment instruction that carries the workers’ names, account details and amounts to the paying system. The FTO requires digital signatures at the block level, conventionally described as a first and a second signature, after which it is pushed to the Union’s payment platform. There the order is processed through the Public Financial Management System, sanctioned against the funds available for the state, routed through the payment bridge and the sponsoring banks, and credited to the worker’s bank or post office account.

The programme’s own operating discipline divides this chain into two stages with two owners. The first stage runs from closure of the muster roll through data entry, generation of the wage list, generation of the fund transfer order and its digital signature by panchayat and block officials. It is the responsibility of the state government, and it is mandated to be completed within eight days of the closure of the muster. The second stage runs from the signed fund transfer order through processing by the Union ministry and its payment systems to the credit of wages in the worker’s account. It is the responsibility of the Union government, and it is mandated to be completed within a further seven days. Eight and seven make fifteen: the two administrative stages together exactly fill the statutory period for payment, so that the chain has no slack. A day lost anywhere is a day of statutory breach, and the design question of who lost it has a designed answer, because each stage has a different government as its owner.

Who is responsible for each stage of a wage payment?

The state government owns the first stage: closure of the electronic muster roll, data entry, generation of the wage list and the fund transfer order, and its digital signature by panchayat and block officials, all within eight days. The Union government owns the second stage: processing the signed order and crediting the worker’s account within a further seven days.

The division was an accountability reform, and on its own terms a sensible one. Under the earlier arrangements, a delayed wage could be blamed in a circle, the block on the district, the district on the state, the state on the Union’s release of funds, with the worker unable to fix responsibility anywhere. The staged electronic chain timestamps every hand-off. Each fund transfer order carries the dates of its own generation, signature, processing and credit, so that in principle the system can say, for every late payment, whose seven or eight days were exceeded. In principle is the operative phrase, because the system’s delay computation, the calculation that decides whether compensation is owed and how much, uses only some of those timestamps. Which ones it uses is the subject of the next section, and it is the most consequential design decision in the programme’s history.

What happens to a fund transfer order after it is signed?

Once the second digital signature is applied at the block, the fund transfer order leaves the state’s hands and enters the Union’s payment platform, where it is checked, sanctioned against the state’s available balance through the Public Financial Management System, routed through the payment bridge to the sponsoring bank, and finally credited to the worker’s account.

What can go wrong in that passage is different in kind from what goes wrong before it. The first stage fails through human process: staff shortages, connectivity, measurement disputes, a signatory absent from the block office. The second stage fails principally through money and data. If the state’s programme account has exhausted the funds the Union has released, signed orders queue unsanctioned, however complete their paperwork, until further funds arrive. If a worker’s account details or Aadhaar mapping fail validation at the bridge, the transaction is rejected and returns down the chain for correction and regeneration, beginning a second journey whose delay belongs to no one’s stage in particular. The second stage is also largely automated, a fact the government’s defence relies on heavily, and automation cuts both ways: an automated stage has no discretion to exercise, but it also has no officer to whom a worker can appeal, and when it stops, it stops for a funding reason or a data reason that the worker cannot see and the block cannot fix.

Four payment regimes in two decades

The staged chain described above is the fourth payment regime the programme has run, and the record of each regime explains the next one’s design. In the first regime, from 2006, wages were disbursed through the administrative hierarchy itself, in many places in cash at or near the worksite, against the muster roll. The defects were those of every cash chain: leakage between the sanctioned amount and the received amount, payment delayed by the physical movement of money, and a record that existed on paper, in the custody of the people it might incriminate. The second regime, adopted from 2008, required wages to be paid through bank and post office accounts in the worker’s own name. This was a structural reform of the first importance: it created, for crores of rural households, a first account, and it made the wage in principle traceable to a named person. Its costs were the costs of a thin banking network, distance to branches, dormant accounts, and a new intermediary class of banking correspondents, but it abolished the cash chain’s easiest thefts.

The third regime electronicised the instruction layer: state-level electronic fund management systems, then the National Electronic Fund Management System from April 2016, moved the authorisation of payment onto the staged, timestamped chain this article has described, and centralised the actual payment in the Union’s systems rather than in state treasuries. The fourth regime, layered onto the third from 2023, changed the payment’s address from an account to an Aadhaar number and its attendance proof from a paper muster to a photographed one; it is treated in its own section below. The direction of travel across the four regimes is consistent: each reform moved discretion away from the local chain and information toward the centre, and each traded one class of failure for another. Cash failed by theft and delay at the point of payment. Accounts failed by exclusion at the point of access. Electronic authorisation failed by queue at the point of funding. Aadhaar addressing fails by mismatch at the point of identity. An implementation record has to score each regime against its predecessor rather than against perfection, and on that scoring the sequence is one of genuine, cumulative improvement in traceability accompanied by a steady migration of the failure point away from the village, where a worker could at least find the responsible person, toward systems in Delhi and in payment infrastructure where no responsible person is findable at all.

The migration of the failure point has a distributional consequence that the delay statistics do not capture. A worker cheated in the cash regime knew a theft had occurred and often knew by whom; a worker whose transaction is rejected by a payment bridge, or whose attendance photograph failed to upload, confronts an event with no author and no scene. The programme’s grievance data reflects the difference: complaints about non-payment remain the largest category of grievance in the programme’s own recording, but the remedy for a systems failure lies in correction of data held in several institutions at once, the job card database, the bank’s records, the payments corporation’s mapper, and the block’s software, and the worker’s part in that correction is to shuttle between the institutions carrying proof of identity to each. The modernisation of payment is therefore best described as a transfer: of the wage’s security, upward, and of the error’s cost, downward. The transfer was worth making on the evidence of the first stage’s transformation, but an honest record prices both sides of it.

Where the clock stops

The management information system at the heart of the programme computes delay, and displays it, for the first stage only. The recorded delay of a wage payment is the number of days between closure of the muster roll and the generation and signature of the fund transfer order, measured against the eight days the first stage is allowed. Delay in the second stage, the days a signed order spends in the Union’s processing before the money reaches the worker’s account, is not computed as delay in the system’s compensation logic, with the consequence the brief for this article states exactly: the extent of second-stage delay, and the compensation payable in respect of it, remain uncomputed. The timestamps exist. The fund transfer order’s journey is recorded. What does not happen is the arithmetic that would convert the second stage’s timestamps into a delay figure and a compensation liability.

This is the point at which the article’s namable claim can be stated, because everything else in the record hangs on it. The clock stops when the state’s part ends. The statute imposes one obligation running from completion of work to payment in the worker’s hands, and attaches compensation to the whole of it. The administrative system divides the obligation into a state stage and a Union stage. The measurement system records delay in the first stage and not in the second. The official figure for timely payment therefore describes half of a two-half process, the half the states perform, and the compensation the statute makes automatic goes uncomputed, and so unpaid, for the half the Union performs. A worker cannot claim compensation that has not been calculated, and a compensation that is never calculated need never be budgeted. The provision has not been repealed; it has been left outside the boundary of the instrument that would give it effect.

The findable artifact below traces a single payment along the whole chain and marks, step by step, where the recorded clock is running and where it has stopped.

Step in the life of one wage payment	Stage	Responsible level	Mandated period	Delay recorded by the system?
Work completed; muster roll closed	Stage one begins	Gram Panchayat and block officials (state)	Day zero of the payment count	The recorded clock starts
Work measured; muster data entered	Stage one	Block and Programme Officer staff (state)	Within the stage’s eight days	Yes
Wage list generated from the muster data	Stage one	Block (state)	Within the stage’s eight days	Yes
Fund transfer order generated from the wage list	Stage one	Block (state)	Within the stage’s eight days	Yes
Fund transfer order digitally signed, first and second signatures	Stage one ends	Block and Programme Officer (state)	By day eight	Yes; the recorded clock stops here
Order processed and sanctioned through the Public Financial Management System	Stage two	Union Ministry of Rural Development	Within the stage’s seven days	No
Payment routed through the payment bridge and sponsoring bank	Stage two	Union systems, the National Payments Corporation and banks	Within the stage’s seven days	No
Wages credited to the worker’s bank or post office account	Stage two ends	Bank or post office	By day fifteen from muster closure	No; the credit exists, the delay is not computed

The rows marked as not recorded are the article’s argument, and they fall entirely in the second stage. Every step the Union performs sits outside the delay computation, and every step the state performs sits inside it. The asymmetry cannot be explained by technical difficulty, because the data for the unrecorded steps is generated by the same systems that record the rest; the fund transfer order’s processing and credit dates were available to researchers who obtained the transaction data, which is how the independent studies discussed below were able to compute the very delays the official computation omits. Nor can it be explained by the compensation rule, which is written against the whole fifteen days and not against the first eight. It is a boundary drawn around the measuring instrument, and like the sibling finding in the sanitation cluster, where the count of toilets constructed and the count of households using them were answers to different questions presented as one, the divergence here is architectural rather than arithmetical: the two sides of the public argument are not miscalculating a shared quantity but reporting different quantities under one name. The manual scavenging record presents the same structure in another field, where the statutory count of a prohibited practice and the counted record of its persistence diverge because the definitions doing the counting were never the same. In this programme the divergence is temporal rather than definitional, but the lesson for a reader of official figures is identical: ask where the clock stops before asking what it says.

The Ministry of Rural Development has not been able to dismiss the finding, because it rests on the programme’s own data. The researchers who first published it record that the Ministry acknowledged the correctness of their findings, and the Supreme Court’s orders in the proceedings discussed below proceeded on the same basis. The programme’s standard operating procedures for timely payment have since spoken of compensation across the whole period, and a format was proposed under which compensation for second-stage delay would be recovered from the stage-two stakeholders responsible. The computation in the live system, however, continued to be made on the first stage, and the gap between the acknowledged position and the operating position is itself part of the implementation record: a defect can be conceded at the level of correspondence and judgment while persisting at the level of code.

Two sets of numbers, both accurate on their own bases

The official figures first, stated as the Ministry states them and on the base the Ministry uses. In May 2017 the Ministry of Rural Development announced that 89.08 per cent of wage payments in the financial year then current had been made within fifteen days of the work being done. In December 2022 the Minister of State for Rural Development told the Rajya Sabha that the percentage of payments generated within fifteen days in that financial year stood at 95.55 per cent. In its later statements the Ministry has said that over the last five years more than 96 per cent of fund transfer orders have consistently been generated within fifteen days of closure of the muster roll, and that 97 per cent of fund transfer orders are generated on time. Read the verbs. The operative measure in the official series is generation: the fund transfer order generated, the payment generated, within fifteen days of muster closure. Generation is the end of the first stage. A fund transfer order generated on day eight and credited to the worker’s account on day forty is, on the official base, a timely payment, because the clock that produced the percentage stopped at generation.

The independent figures are computed on the whole journey. Rajendran Narayanan of Azim Premji University, Sakina Dhorajiwala and Rajesh Golani of LibTech India analysed over ninety lakh wage transactions across ten states for the financial year 2016-17, using the programme’s own transaction records, in a study published as “Analysis of Payment Delays and Delay Compensation in MGNREGA”. They found that only 21 per cent of payments were completed within the statutory fifteen days when both stages were counted. The first stage was not the principal problem in their sample: it was the second stage that consumed the time, with the Union government alone taking an average of over fifty days to transfer wages electronically after the fund transfer order reached it. Their follow-up analysis of the first two quarters of 2017-18 found 32 per cent of payments completed on time, an improvement that still left two-thirds late. A later LibTech India analysis of more than three crore wage transactions for 2021-22 found that only 29 per cent of payments were credited within the seven days allotted to the second stage alone. The same study programme computed what the measurement boundary costs in compensation: in the 2016-17 sample, the delay compensation actually payable on the whole journey was about Rs 36 crore, while the compensation the system actually calculated, on the first stage only, was about Rs 15.6 crore. More than half of the entitlement the statute created in that sample was never computed, and what is never computed is never paid.

Neither side of this argument is fabricating. The Ministry’s percentages are percentages of fund transfer orders generated within the first stage’s discipline, and by that measure the system’s performance genuinely improved across the decade: the staged electronic chain, the standard operating procedures and the pressure of the Supreme Court’s supervision compressed the first stage from the chronic delays of the cash and state-treasury era into a process that in most states completes within its eight days in the great majority of cases. The researchers’ percentages are percentages of workers paid within the statute’s fifteen days, and by that measure the obligation continues to be breached in the majority of transactions, because the second stage, the Union’s own, is where the queue forms. The two series cannot be averaged, trended together or quoted interchangeably, and a report that sets the Ministry’s 96 per cent beside the researchers’ 21 per cent as though one refuted the other has misunderstood both. They answer different questions. Only one of them answers the statute’s question.

For the reader who has to quote these figures, the working rule is short. A timely-payment percentage is meaningless until three things are attached to it: the base (orders generated, or wages credited), the period (which financial year, and whether the whole year), and the source (the Ministry’s statement of its own measure, or a named analysis of the transaction data). Journalists who attach all three will find that the story is not that one side is lying; the story is the boundary itself, and it is a better story, because it explains how a government can report success in good faith while workers wait two months for wages in the same good faith’s jurisdiction.

The days delivered against the guarantee

The second measurement is simpler and in some ways more damning, because it requires no reconstruction of anybody’s clock. The guarantee is one hundred days per household per year. The programme’s average delivery, on the Ministry’s own figures laid before Parliament, has run at about half of that. The Minister of State for Rural Development informed the Rajya Sabha that the average days of employment provided per household were 51.54 in 2020-21, 50.07 in 2021-22, 47.84 in 2022-23, 52.07 in 2023-24 and 50.24 in 2024-25, a five-year run in which the average never rose above fifty-two days or fell below forty-seven. An average is not the guarantee’s test, since households differ in how much work they demand, and a household that demands forty days and receives forty has been treated exactly as the Act requires. But the distribution behind the average shows how the guarantee operates at its ceiling. On LibTech India’s computation from the programme’s own management information system data, published in its national review of implementation for 2024-25, the households that completed the full one hundred days were 7.6 per cent of the households that obtained work in 2023-24 and 7.0 per cent in 2024-25. Roughly ninety-three of every hundred working households stopped short of the guaranteed maximum, most of them far short, at an average near fifty days.

The same review records the divergence between the programme’s register and its reach. Households registered under the programme rose from 13.80 crore in 2023-24 to 14.98 crore in 2024-25, an increase of 8.6 per cent. Households that actually obtained work fell in the same period from 5.51 crore to 5.35 crore, and person-days generated fell from 288.83 crore to 268.44 crore, a decline of 7.1 per cent. The register grows because registration is the gateway to a right and households rationally hold the gateway open; delivery shrinks or grows with the funds released, the works opened and the administrative capacity of the year. A registration figure is therefore a measure of potential demand, and a worked figure is a measure of funded supply, and the widening distance between the two series is the shape of a guarantee operating inside an appropriation. The review also notes a geographic hole in the national totals: no employment was generated in West Bengal in either year, because Union funds for the state stood stopped, a matter taken up in the funding section below, so the national delivery figures for those years describe a programme absent from one of its largest states.

The pandemic year stands as the control experiment in this series. In 2020-21, when reverse migration carried workers back to their villages and the Union raised the programme’s funding substantially in mid-year, the system delivered at a scale it has not matched before or since: on the Ministry’s figures given to Parliament, 11.17 crore individual workers obtained employment in that year, against 8.75 crore in 2022-23 and 8.34 crore in 2023-24, and the average days per household reached 51.54 even as the denominator swelled with new demand. The lesson the year teaches is double-edged. The machinery can expand enormously when money and administrative priority arrive together, which refutes the claim that the programme’s ordinary shortfall is a law of nature. But even in that year of maximum effort, the average household received barely half the guaranteed days, which shows how far the system’s ordinary capacity sits below the entitlement’s formal promise, and how much of the shortfall is therefore structural rather than episodic.

The lean season: the guarantee as seasonal insurance

The demand the programme serves is not spread evenly across the year, and its implementation has to be judged partly as a piece of seasonal machinery. Agricultural employment in most of rural India concentrates in the sowing and harvest peaks and collapses in the long lean season between them; the programme’s demand curve runs inversely, rising when farm work disappears. The design anticipates this: the entitlement is available on demand at any time, but its insurance value is realised in the months when the alternative is distress migration or consumption credit at moneylender rates. The delivery record’s seasonal failures are therefore weighted failures. A work opened in the harvest month serves a worker who had alternatives; a work not opened in the lean month fails a worker who had none. The drought mechanism belongs to the same design: the additional fifty days the Act permits in notified calamity areas exist to thicken the guarantee exactly where and when the agricultural economy has failed, and in the drought years of the middle of the last decade the provision was notified for affected districts across several states and programme delivery expanded sharply there, the expansion the Swaraj Abhiyan proceedings then supervised. The converse also holds in the record: in years and states where the programme’s funds ran short in the final quarter, the shortfall landed in the lean season’s tail, when demand is most desperate and the wage delay least absorbable.

One further feature of the delivered days deserves record because it is a design success measurable in the same data. The Act requires that at least one-third of the workers be women. On the Ministry’s figures laid before Parliament, women were 4.60 crore of the 8.34 crore workers who obtained employment in 2023-24, about 55 per cent, a share that has held above half for years. The self-selection design, a wage set at a level that does not displace male migration wages in many regions, equal wages for men and women by the Act’s express command, and work located within the village combined to make the programme, in delivery, majority-female, a record no comparable public works programme in the country matches.

Planning from below, in text and in practice

The Act’s planning design runs upward: the Gram Sabha recommends the works to be taken up and their order of priority, the Gram Panchayat prepares a development plan and a shelf of projects, the shelf aggregates through the block and district into the district plan, and the labour budget is built from the aggregation, so that in the statute’s imagination the funded programme is the sum of village decisions. The implementation record of planning is the distance between that imagination and the annual ritual the audit found. Plans were prepared, but as compilations of works proposed by the implementing tier rather than as expressions of the Gram Sabha’s priorities, with quorums thin and recommendations ratified rather than debated; shelves of projects ballooned beyond any plausible execution capacity, which is how the early audit could find only about thirty per cent of approved works implemented, since an approval list built to satisfy every proposer’s expectation is not a plan but a register of hopes. The labour budget, nominally the sum of these plans, was in practice negotiated downward against the expected release, importing the appropriation’s ceiling into the plan’s base and completing the inversion of the statutory design: the plan was supposed to generate the budget, and instead the budget, anticipated, generates the plan.

The planning failure matters to the wage and days record because it is upstream of both. Works that exist only on a shelf cannot be opened when demand arrives in the lean season, and the lag between an unpredicted demand and a newly sanctioned work is measured in exactly the weeks the guarantee’s fifteen-day promise does not tolerate. Districts and states that maintained a ready shelf of sanctioned, technically estimated works could respond to demand spikes, drought notifications and the pandemic’s return migration at speed; those that planned on paper discovered, at the moment of need, that the sanctioning and estimation pipeline was the binding constraint. The planning record thus closes the circle of this article’s argument: the programme’s failures present at the counter as delay, rationing and missing days, but they originate in a chain of upstream decisions, about plans, budgets, staffing and measurement boundaries, each defensible in isolation and jointly fatal to the guarantee’s text.

Counting the unit: person-days, households and the moving database

The figures in the delivery sections reward a methodological pause, because the programme’s units are treacherous and most public error about it is unit error. The person-day is the basic coin: one person working one day. It aggregates unlike things, a day worked by a household’s only earner and a day worked by its fourth member counting identically, and it says nothing by itself about how the days were shared. Every average in the debate is a ratio of person-days to some household base, and the base is a choice: households registered, households that demanded work, or households that obtained work. The Ministry’s five-year series of averages near fifty days is computed over households that obtained employment; the same person-days divided by registered households would yield a figure near twenty, and divided by households that demanded work a figure between the two. None of these is the true average; each is the average over a differently defined population, and a speaker who wishes to make the programme look generous or miserly can do so, without inventing a single number, by selecting the denominator.

The database underneath the ratios also moves. The management information system is a living register in which transactions are posted, corrected and regenerated after the fact, so that two honest extractions of the same financial year taken at different times differ. The reader can watch the effect in miniature in this article’s own sources: for 2023-24, the average days of employment per household appear as 52.07 in the Minister of State’s answer to the Rajya Sabha and as 52.42 in LibTech India’s computation from the programme’s data, a small divergence with no villain in it, produced by extraction timing and base conventions in a database that does not freeze. The lesson generalises beyond this programme and is one of the series’ recurring findings: in any large administrative dataset, the figure is a function of the extraction, and citing a number without its extraction is citing half a fact. Registration figures carry the same warning in the opposite direction. A job card is a household document; a worker is a person; an active worker, in the Ministry’s usage reported in December 2023, is a person who has worked at least one day in the last three financial years; and the three populations, 14.98 crore registered households in LibTech’s 2024-25 count, 25.25 crore registered workers in the Ministry’s December 2023 count, 14.35 crore active workers in the same count, are all real and all different. Public argument moves among them without announcement, and a reader who holds the units apart will find that many celebrated contradictions in the programme’s statistics dissolve into denominator changes, while the genuine findings, the delay boundary chief among them, survive every denominator because they are built into the instrument.

The works and the assets: the other half of the record

A wage is paid against a work, and the implementation record of the works themselves is the half of the programme’s history that the payment debate tends to crowd out. Schedule I to the Act lists the permissible categories of work, concentrated on water conservation and harvesting, drought proofing and afforestation, irrigation, renovation of traditional water bodies, land development, flood control and rural connectivity, with individual-beneficiary works on the lands of specified categories of households added over time. The Act’s disciplines are structural: a ratio of wage cost to material cost of sixty to forty is to be maintained at the district level, contractors are prohibited, and at least half the works by cost are to be executed through Gram Panchayats. The design intends the wage bill to leave assets behind it, and the asset question, whether a hundred days of digging produced a pond that holds water, is the question on which the programme’s economic defence ultimately rests.

The audit record on assets is mixed in a specific pattern. The Comptroller and Auditor General’s 2013 performance audit, examining the programme’s first five years, found that of the works approved in annual plans in the states it covered, only about 30 per cent were implemented in the audit period, and that 7,69,575 works, on which Rs 4,070.76 crore had been spent, stood incomplete after periods of one to five years, with a further large body of expenditure recorded as not resulting in durable assets. The pattern the audit identified was administrative rather than mysterious: works sanctioned in numbers far beyond the implementing capacity of panchayats and blocks, technical staff too thin to measure and supervise, and completion certificates chasing sanctions rather than structures. Later programme policy pushed the measurable margins: water conservation was made a mission priority, assets were geo-tagged on a public map, 1.45 crore of them by the Ministry’s count in 2017, and individual assets on household land, wells, ponds, livestock shelters and plantations, grew as a share of the portfolio, shifting the asset question from the quality of public works to the distribution of private ones.

The honest summary of the asset record is that it varies by exactly the variables that govern the wage record: the technical and administrative capacity of the state, and the presence of an audit institution that inspects outcomes rather than expenditures. Where social audit units and engineering staff exist in strength, works complete and assets persist; where the programme is run as a wage-distribution exercise with works attached, the incomplete-work percentages of the early audit repeat themselves. For this article’s purpose the point is narrower. The asset debate and the wage debate are usually conducted as though one could be won while the other is lost, the programme defended as employment or attacked as digging and refilling. The record shows them rising and falling together on the same administrative variable, which is what one would expect of two outputs of a single delivery system.

The wage in the wider labour market

The level of the wage, not only its punctuality, is part of the implementation record, because Section 6’s delinking made the level a standing decision rather than a settled fact. The chronology is short. From 2006 to 2008 the programme paid the state’s minimum wage for agricultural labour, the Act’s default before the Union notifies its own rate. In January 2009 the Union notified separate rates, freezing the amounts then in force, and thereafter revised them annually by the change in the Consumer Price Index for Agricultural Labourers. Two consequences followed. The programme wage ceased to track the legal minimum wage of the state, and in a large number of states it settled below that minimum, a position the Act permits because Section 6 operates notwithstanding the Minimum Wages Act, but one in visible tension with the principle the Supreme Court stated in Sanjit Roy v State of Rajasthan in 1983, that paying a worker on state works less than the minimum wage amounts to forced labour within Article 23. The tension is managed legally by the non-obstante clause and politically by the argument that the programme is relief employment rather than an employment relationship, an argument whose strength the reader can judge against the Act’s own description of its wage as a wage.

Expert and parliamentary bodies have repeatedly proposed closing the gap. A committee chaired by Mahendra Dev, appointed to review the wage indexation method, recommended that the programme wage be linked to the state minimum wage or indexed to a rural consumer price index that better reflects workers’ costs; the recommendation was not adopted. The Parliamentary Standing Committee on Rural Development, in February 2024, recommended that programme wages be raised to at least Rs 375 a day, adopting the figure a committee chaired by Anoop Satpathy had proposed in 2019 as a national minimum wage, and the committee noted that the programme’s daily wages for 2023-24 were lower than the daily earnings of agricultural labour in the preceding year in all major states except Gujarat. The notified rates for 2024-25, on the analysis of the Ministry’s notification published in the press, topped out at Rs 374 a day in Haryana, with the average notified rate across states a little above Rs 280. Real wages under the programme, on the twenty-year review already cited, stagnated from 2009-10 onward once inflation is accounted, a stagnation that itself explains part of the delivery record: a guarantee whose wage drifts below the market wage will be demanded fiercely in the lean season and in distress, and lightly in the peak season and in prosperity, which is a defensible shape for a safety net and an indefensible one for a statutory right to work at a legal wage. States are free to top up the notified rate from their own funds, and a few have done so at times, converting a Union-set wage into a state one only where the state’s politics and finances permitted.

The demand that never enters the record

Every figure in the preceding sections is computed over recorded events: recorded musters, recorded applications, recorded households. The deepest distortion in the record is the demand that was never recorded, because an unrecorded application starts no clock, founds no unemployment allowance, and appears in no denominator. The programme’s demand statistics are therefore, by construction, statistics of demand the administration chose or managed to register, and the gap between those statistics and the demand that existed is unmeasurable from inside the system. It can only be approached from outside, through audit.

The Comptroller and Auditor General’s performance audit of the programme, Report No. 6 of 2013, covering the period from April 2007 to March 2012, examined the base of the record and found it porous. In 1,402 of the Gram Panchayats it test-checked, 36.43 per cent, dated receipts for applications for work were not given at all. In 2,068 test-checked Gram Panchayats, 53.74 per cent, employment registers were not maintained or not properly maintained. The audit drew the entitlement conclusion expressly: in the absence of proper records, the entitlement to unemployment allowance could not be worked out, depriving beneficiaries of rights the Act confers, and the lapse called for fixing accountability. Audit chapters on individual states recorded the mechanism in its purest form, noting that on the management information system applications for work were shown as exactly equal to work allotted, a bookkeeping identity that abolishes unmet demand by definition: if the system records as demand only what it supplies, the supply can never fall short. State audit also recorded applications made orally and never documented, allotments made without letters, and beneficiaries surveyed in large numbers reporting that no dated receipt had ever been issued to them.

The unemployment allowance is where this recording failure does its legal damage, and the allowance’s record is the clearest case in the programme of an entitlement defeated upstream of any refusal. The allowance accrues only where a dated application exists and fifteen days pass without work. Where the application was oral, unreceipted or unregistered, the legal precondition is unprovable, and the allowance is not so much denied as unborn. The 2013 performance audit found unemployment allowance not paid in 47,687 cases in the states it examined, and noticed non-payment of the allowance across the states generally. The incentive structure completes the explanation. The allowance is the state’s own liability, charged outside the Union’s wage funding, so every recorded application that goes unmet for fifteen days creates a debt for the government responsible for recording it. A rational administration under fiscal pressure under-records demand, not necessarily by conspiracy but by the ordinary friction of understaffed panchayats, unavailable forms and receipts that are nobody’s priority. The Act anticipated the failure mode and made refusal to issue a receipt a contravention under Section 25, but a contravention punishable with a small fine, prosecuted by the same machinery that benefits from the omission, has not proved a working deterrent.

The audit record preserves one episode that captures the allowance’s strange position from the other side. In two blocks of Sitapur district in Uttar Pradesh, 860 workers demanded unemployment allowance for a period in 2007 in which they said work had been requested in writing and not provided. The District Programme Coordinator rejected the claim on the ground that work in another scheme had been offered; the workers’ union appealed; the Coordinator could produce no documentary evidence that work had been offered; and the Commissioner directed payment of the allowance, some Rs 14.99 lakh. The episode then entered the audit report as a case study in avoidable payment, because the allowance, a state liability, had been paid out of programme funds. The allowance is thus doubly exposed: unpaid where the records fail the worker, and irregular where it is paid from the wrong pocket. What the episode proves, beneath the accounting objection, is that the entitlement was real, was claimed through the machinery, and was established precisely because the workers in that instance held written applications the administration could not answer. The exception demonstrates the rule. Where the paper exists, the right can be enforced; the system’s ordinary practice is that the paper does not exist.

Nothing in this account is advice on an individual claim. A worker pursuing a particular payment or allowance should take the dated documents to the Programme Officer, the district Ombudsman appointed under the programme’s instructions, and, where those fail, the district legal services authority, because individual claims turn on individual records and limitation discipline is unforgiving. What general analysis can do is identify where the record is made and lost, and for this programme it is lost earliest, at the application counter, before any question of funds or payment stages arises.

Rationing: who receives the days that exist

Between the recorded demand and the delivered days sits the allocation process, and the research literature’s central finding about it is that the programme rations. Rationing in this context has a precise meaning: households that want work at the programme wage, and would take it, do not receive it, not because a rule excludes them but because the supply of opened works, sanctioned musters and funded days is smaller than the demand at the counter. The most careful measurement of the phenomenon remains the study of Bihar by Puja Dutta, Rinku Murgai, Martin Ravallion and Dominique van de Walle, published as an assessment of the employment guarantee in that state, which found substantial unmet demand for work among poor households and an allocation of the available days that did not track need closely: households with better connections and resources were not systematically excluded, but the days available were too few to reach all who sought them, and the poorest state in the country was operating one of the thinnest programmes relative to its poverty. Their finding generalises into the pattern comparative work calls the programme’s inverse geography: for much of its history, the states with the deepest rural poverty have generated fewer person-days per poor household than states with stronger administrations and shallower poverty, because delivery capacity, not need, is the binding constraint.

Rationing operates through mechanisms that leave no refusal to record. A work is not opened in the week demand peaks, because the labour budget’s envelope for the quarter is spent or the technical staff cannot measure another site. A muster is issued for ten days where twenty were sought. A Gram Panchayat, anticipating that wages will be delayed for want of funds, discourages applications it cannot honour, and the discouraged application joins the unrecorded demand of the previous section. The queue is real but invisible, which is why programme data can show demand and supply in near balance, the Ministry was able to tell Parliament that in 2025-26 nearly all rural households demanding work had been offered employment, while village-level research continues to find workers who wanted more days than they received. Both statements describe the same system from different ends: the offer is measured against the demand that survived the funnel, and the funnel is the finding.

The distributional consequence matters more than the aggregate. When days are rationed, the households best placed to obtain them are those with the information, the proximity to the panchayat and the ability to wait for delayed wages, and the households rationed out are disproportionately those for whom the insurance value of the work was highest, casual labour households without assets, single-woman households, and households whose members cannot absorb a two-month wage delay and so withdraw from the queue into costlier private credit or migration. Wage delay and rationing therefore compound: the same funding shortage that delays the wage also shrinks the days, and it shrinks them selectively against the workers the Act most intended to reach. The fifty-day average is not fifty days for everyone; it is the mean of a distribution in which a minority of households approach the guarantee and a long tail receives very little, and the implementation question is as much about the tail as about the mean.

The MGNREGA implementation record in one household’s season

The aggregate record can be reassembled into a single season, as an arithmetic illustration built only from values documented in the sections above, to show how the system’s separate failures accumulate in one household’s year. Take a casual-labour household in a rain-fed district, two working adults, registered and holding a job card. In the lean season the adults apply for work; the application is taken orally at the panchayat, no dated receipt is issued, and work opens three weeks later when a sanctioned work with a ready estimate begins, a sequence whose lateness founds an unemployment allowance claim that the missing receipt makes unprovable. Across the season the household obtains fifty-two days of work, the national average for a household that worked in 2023-24 on the Minister’s figures, shared between the two adults against the one hundred the card entitles them to jointly demand. The days are worked in spells; each spell’s muster closes, and the wages for the spells reach the household’s account in intervals that the end-to-end studies describe, some within the fortnight, others after six or eight weeks, the delay concentrated after the fund transfer orders are signed, in the stage no dashboard displays. The interest cost of those intervals is real and unrecorded: consumption in the waiting weeks is financed from the local shop’s credit at the customary premium, and the season’s borrowing repays itself out of the wages when they land. For the longest delay, the Schedule II formula yields a compensation of some tens of rupees, which is not computed, because the delay fell in the unmeasured stage, and is therefore not paid; the Ministry’s national figure for the compensation that was paid, Rs 5.27 lakh for a year of hundreds of crores of person-days, is what this household’s share of the entitlement looks like when aggregated honestly. The household ends the season having received, in cash terms, roughly half the wage value the statute held out to it, at times of the system’s choosing, with no single official having acted unlawfully in any way the household could prove. That is the implementation record at ground level: not a story of refusal, but of a guarantee filtered through recording, rationing, funding and measurement until what emerges is a benefit, a real one, at about half the statutory size.

The delivery tier: the people inside the first stage

The first stage’s eight days are not performed by software. They are performed by the thinnest administrative tier in Indian government: a Gram Rozgar Sahayak or Sevak for the panchayat, often covering more than one; a mate at each worksite, drawn from the workers and paid at the programme’s skilled rate; technical assistants responsible for measurement across many works; data entry operators at the block who convert paper and field data into the system’s records; and a Programme Officer who is, in most states, the Block Development Officer carrying the programme as an additional charge among many. The Act’s administrative expenses are funded within a norm of six per cent of programme expenditure, and the tier’s staffing has never matched the transaction load the design assumes: a single block may process thousands of musters in a peak month through a handful of functionaries, two of whom must digitally sign every fund transfer order. Stage one fails, when it fails, at predictable points in this tier: the signatory is on leave or transferred and the signature queue waits; the technical assistant’s measurements backlog and musters cannot close; connectivity or a software change stalls data entry for days. These are capacity failures, not corruption, and the first stage’s great improvement over the decade is substantially a story of this tier being progressively equipped, with dedicated data systems, standard operating procedures and monitoring that named the delay at the level of the individual order.

The tier’s thinness also explains features of the record that pure funding analysis misses. Measurement, the act on which the wage itself depends, is the scarcest skill in the chain, and the programme’s works have always queued for the technical assistant’s visit; a delayed measurement delays the muster closure that starts the payment clock, so that part of what appears as payment delay is works-supervision delay upstream of the clock entirely, invisible to the stage analysis because the clock has not started. The mate system concentrates a second risk at the worksite: the mate records attendance and leads the work group, and social audits’ findings of proxy attendance and inflated musters are findings about this office and its supervision. The design choice underneath the tier is fiscal and deliberate. A programme that capped its administrative cost at six per cent chose breadth of wage coverage over depth of administration, and many of its signature failures, the unrecorded application, the short measurement, the unsigned order, are that choice arriving at the counter. Judged as a whole, the delivery tier performs a volume of transactions that has few parallels in world administration, and it performs them with staffing that no private payment system would attempt; the record’s criticism is properly addressed not to the tier but to the establishment tables that starve it.

The grievance machinery and its record

A worker failed by the system has, in design, a ladder of remedies inside the programme before any court. Complaints lie to the Programme Officer at the block, with appeal upward to the District Programme Coordinator; the Act makes refusal to accept an application or issue a receipt a contravention; and over the structure sits a district Ombudsman, an independent authority for the programme created not by the Act but by Union instructions issued in 2009, empowered to receive complaints, inquire, and direct corrective action and compensation recommendations. The Ombudsman is the kind of institution this series has learned to watch: a statutory programme’s forum resting on executive instructions, its authority real but borrowed, its coverage dependent on whether states appoint and fund it. Appointments across districts have been incomplete and intermittent in the programme’s history, and an office that exists in the instructions and not in the district is, for the worker, indistinguishable from no office.

The programme’s own grievance recording shows what workers actually complain of: non-payment and delayed payment of wages dominate the complaint volume, followed by complaints about work not provided and about job cards and registration. The distribution is itself evidence, since a grievance system records the failures the worker can see. A worker can see an unpaid wage; she cannot see an uncomputed compensation or an unrecorded application in her own name, so the system’s two most structural failures generate almost no grievances, and the grievance statistics understate them by construction. Disposal figures, complaints closed within stated periods, measure the office’s throughput rather than the worker’s remedy, because a complaint about a wage delayed by a funding queue is typically closed by the wage’s eventual arrival, which the system counts as resolution and the worker experienced as two months of borrowing. None of this makes the machinery useless: individual officers do recover individual wages through it, and the Ombudsman’s inquiry power has in places disciplined local practice. It makes the machinery an instrument calibrated to the visible failures, running in parallel to the invisible ones, and a complete implementation record has to keep both calibrations in view.

The courts and the compensation that cannot be claimed

The programme’s payment failures reached the Supreme Court in Swaraj Abhiyan v Union of India, public interest proceedings arising out of the drought that affected several states in 2015 and 2016, in which the Court treated the employment guarantee and food security statutes as the measure of the governments’ obligations in the drought districts. In its judgments of 2016, reported at (2016) 7 SCC 498 and in the connected parts of the same proceedings, a bench of Justice Madan B. Lokur and Justice N.V. Ramana held that there is no informal capping of funds under the programme, recorded that there had been unconscionable delay on the part of the Government of India in releasing funds under both the wage and the material components, and directed the Government of India to ensure that compensation for delayed payment beyond fifteen days is made to workers. The Court observed that when the rights of lakhs of people are affected by delayed payment of legitimate dues, a constitutional breach is committed by the state, whether the Government of India or a state government is responsible for the particular delay. It also directed the constitution of the Central and State Employment Guarantee Councils the Act requires, bodies whose uncertain existence the audit had also remarked.

The wage question returned to the same bench, and in May 2018, in the judgment reported as 2018 INSC 547, the Court converted the general directions into a specific holding on the payment chain. A worker, the Court held, is entitled to payment of wages within a fortnight of the work being done, and on failure is entitled to the compensation Schedule II prescribes. Critically for the staged architecture, the Court refused to let the two governments divide the obligation into innocence: the burden of compliance, it held, is on the state governments and Union Territory administrations as well as the Centre, and one entity cannot pass the burden to another. Bureaucratic delay and red tape, the Court said, cannot be pedalled as an excuse to deny workers their wages, and delays are simply not acceptable. The Centre was directed, through the Ministry of Rural Development and in consultation with the state governments, to prepare an urgent, time-bound and mandatory programme to secure payment of wages and compensation, in the interest of the workers and in furtherance of the rule of law.

What did the Supreme Court direct on delayed MGNREGA wages?

In Swaraj Abhiyan the Supreme Court held in 2018 that a worker is entitled to wages within a fortnight of the work and to compensation on failure, that compliance burdens the Centre and the states jointly and cannot be passed between them, and that the Centre must frame a time-bound programme securing payment of wages and compensation.

Measured against the measurement problem at this article’s centre, the judgments both achieved a great deal and struck a structural limit. They established, at the highest level, that the fifteen-day obligation is single and indivisible, that the Centre’s stage is inside it, and that compensation is owed for breach of the whole. They moved money: the proceedings repeatedly pressed the Union on fund releases, and the researchers’ analysis of the first stage records improvement under the Court’s supervision. What a judgment cannot do is rewrite the programme’s software. The compensation entitlement the Court affirmed still has to be computed transaction by transaction by the management information system, and the system computes it on the first stage. In the continuing proceedings, the Court was told that only a small fraction of the compensation assessed as due had actually been paid. The worker’s practical position therefore remains what the architecture makes it: the entitlement is real, the forum that vindicated it is the Supreme Court acting for the class, and the individual claim, which depends on a computation no instrument performs, has no working path. Litigation under this statute has consequently taken a systemic form, public interest proceedings about releases, computations and programmes, rather than the claim-by-claim form the Act’s design supposed, and that substitution of the systemic for the individual is itself a finding about implementation.

The arithmetic of a small entitlement: delay compensation in practice

Delay compensation deserves a section of its own arithmetic, because the rate that sounds derisory in the abstract and the aggregate that sounds substantial in the studies are the same entitlement viewed from opposite ends. The rate is 0.05 per cent of the unpaid wages for each day of delay beyond the sixteenth day from closure of the muster roll. Take a worker owed ten days of wages at a notified rate of Rs 220 a day, a sum of Rs 2,200, whose payment arrives sixty days after the muster closed, forty-four days of compensable delay. Her compensation is a twentieth of one per cent of Rs 2,200 multiplied by forty-four: a little under fifty rupees. The sum is not nothing to her, but it is not a sum for which an individual claim, a journey to the block office, a contested verification, is a rational expenditure of a day’s labour, and everyone in the system knows it. Compensation of this design can only ever function as an automatic computation credited with the wage, a discipline on the payer rather than a remedy the payee pursues. That is plainly what the drafters of Schedule II intended: the formula needs no discretion, no assessment of loss, no hearing, only two dates and a multiplication.

The aggregate tells the same story at national scale. The LibTech sample for 2016-17, ten states and over ninety lakh transactions, put the compensation actually payable at about Rs 36 crore against about Rs 15.6 crore the system computed, in a sample covering a fraction of one year’s payments; extrapolated in the mind across the whole programme and the whole period, the uncomputed entitlement is a large sum of money that has silently not been transferred to the poorest wage-earners in the country. What has actually been paid is documented by the Ministry itself: in a statement on wage payment, the Ministry recorded that delay compensation rules had been notified in 27 states and Union Territories and that the compensation paid for delays in financial year 2024-25 stood at Rs 5.27 lakh. Five lakh rupees, across a programme that in that year paid wages on hundreds of crores of person-days with, on the independent analyses, a majority of end-to-end payments late, is not a rounding error on the entitlement; it is the measurement of the entitlement’s practical extinction. The mechanics of payment add a final turn: the state is to pay compensation upfront and recover it from the functionaries or agencies responsible for the delay, so that even where compensation is computed, paying it sets off an internal recovery process against identifiable officials, a design that guarantees institutional resistance at exactly the point of payment. A provision can be killed by many instruments; this one was killed by a computation boundary, an upfront-payment rule and a recovery clause, while remaining, throughout, in force.

The funding question: a guarantee inside an appropriation

The programme’s deepest structural tension is not administrative but constitutional in the practical sense: the entitlement is uncapped and the money is not. Section 3 obliges the state to provide one hundred days to every household that demands them; nothing in the Act makes the obligation conditional on the size of the year’s budget. The money, however, arrives through the ordinary appropriation process, in which the Union budgets a sum for the programme, districts submit labour budgets projecting the demand they expect to fund, the Union approves a labour budget for each state, and releases flow against that approval. The Swaraj Abhiyan Court held that the agreed labour budget is not an informal cap on the statutory obligation, and the Ministry’s own statements describe the programme as demand-driven, with additional funds sought from the Ministry of Finance as and when required and pending liabilities reimbursed at the start of the following year. The formal position is therefore that the entitlement governs and the budget follows. The record of releases and arrears describes a more complicated practice.

How does the approved labour budget shape the year’s work?

The labour budget is the district’s funded projection of demand, approved by the Union before the year begins, against which releases are made. Because works are opened, musters issued and wages sanctioned within the envelope the budget and the releases create, the projection functions in practice as the year’s working ceiling on employment, whatever its formal status as an estimate.

The mechanics of the shortfall are visible in the rollover of liabilities. People’s Action for Employment Guarantee, the network of organisations and researchers that tracks the programme, has estimated that roughly a fifth of the funds available in a year are absorbed by clearing the previous year’s unpaid wage and material liabilities, so that a nominal allocation overstates the year’s working resources by that share before a single new muster is opened. Allocations have repeatedly been exhausted well before the year’s end, with supplementary demands then raised; the pattern of a budget estimate, an in-year crisis of pending wages, and a supplementary correction is a recurring feature of the programme’s financial history rather than an aberration of any single government. The allocation retained for 2025-26 was Rs 86,000 crore, the same figure as the year before, in a programme whose wage rates are revised upward annually and whose register of households continues to grow. As a share of the economy, central expenditure on the programme has run between 0.3 and 0.4 per cent of gross domestic product in the years since 2015-16, apart from the pandemic surge, down from a peak of 0.6 per cent in 2009-10, on the figures assembled in Jean Dreze’s twenty-year review of the programme published by Ideas for India. The People’s Action for Employment Guarantee has argued that an allocation several times larger would be needed to fund the entitlement at full demand; the Ministry’s answer is that allocations follow demonstrated demand and that every admissible liability is eventually paid, an answer whose truth depends on how much of the demand was ever allowed to demonstrate itself.

The arrears themselves have an internal order that the funding debate usually flattens. A year’s unpaid liabilities are not only wages: they include material payments to suppliers and skilled workers, administrative costs, and the compensation and allowance dues this article has tracked. When releases tighten, the queue is not neutral. Wage payments carry the statute’s compensation and the courts’ attention, so they are prioritised in practice; material dues wait longest, and the waiting propagates into the works programme itself, because suppliers who carry a season of unpaid bills raise their prices, withdraw from programme works, or must themselves be carried by the panchayat’s credit, and engineers facing unpaid material accounts sanction fewer new works. State audit reports across the years record material liabilities ageing far beyond the wage delays that draw the headlines. The funding shortfall therefore degrades the programme on two clocks at once: the worker’s wage clock, which the courts supervise, and the supplier’s payment clock, which nobody supervises, and the second clock’s failure reappears the following season as fewer works opened and thinner technical supervision, feeding the rationing described above. A funding settlement that cleared wages while leaving material dues to age would not have settled the programme’s finances; it would have moved the queue to where the cameras are not.

The funding lever can also be pulled in the opposite direction, against a state rather than for one. Section 27 of the Act empowers the Union to issue directions and to withhold releases where a state fails to comply with the Act’s requirements. From 2022, the Union stopped fund releases to West Bengal under this power, citing irregularities in implementation; the state government disputed the stoppage and the matter was litigated. The LibTech India national review records the consequence in the employment data: no employment was generated in West Bengal in 2023-24 or 2024-25. Whatever the merits of the compliance dispute, the episode demonstrates what the funding architecture makes possible: a statutory entitlement held by households in one state can be switched off, for years, by a funding decision taken in a dispute to which the households are not parties, and the Act’s own machinery provides the switch. A guarantee that can be suspended by appropriation in one state, and rationed by appropriation everywhere, has a legal character different from its text, and any honest implementation record has to hold both facts at once: the text is unconditional, and the money is not.

The fair statement of the policy question, which the verdict section below takes up, is whether that gap is a failure of the programme or of the appropriation. The programme’s machinery, the labour budget, the staged payment chain, the release system, is built to transmit funding decisions faithfully; it transmits scarcity as faithfully as abundance. A legislature that enacts an uncapped entitlement and then funds a fraction of its full-demand cost has made a decision that the machinery merely executes, and the queue that forms in the second payment stage, the works not opened in the lean months, and the demand never recorded at the panchayat counter are all, at one remove, the same decision arriving at different counters.

One muster, three payment streams: the caste-segregated fund transfer orders

In March 2021, by a circular dated 2 March, the Ministry of Rural Development changed the payment architecture in a way that turned the programme into an unintended experiment on its own second stage. Until then, the workers on a single muster roll, who had worked together at one site for the same days, were paid through a single fund transfer order. The new instruction required payment to be separated by category: three fund transfer orders for one muster, one for Scheduled Caste workers, one for Scheduled Tribe workers and one for all others. The Ministry’s stated purpose was accounting: to reflect accurately the flow of funds to the different population groups, in line with budgetary practice that tracks expenditure under the Scheduled Caste and Scheduled Tribe component heads. Parliament’s Standing Committee on Rural Development questioned the arrangement and asked for its rollback, and workers’ organisations objected that the same work, on the same muster, was being paid at different times to different castes.

The objection was borne out by the transaction data. LibTech India’s analysis of 18 lakh wage transactions across ten states for the period from April to September 2021 found systematic differences in processing time by category: in the first stage, fund transfer orders were completed on time for 46 per cent of Scheduled Caste payments, 37 per cent of Scheduled Tribe payments and 26 per cent of payments to the residual category. Reporting in The Hindu, tracking the system’s operation in October 2021, recorded co-workers on the same muster roll in Rajasthan paid weeks apart by category, and in Madhya Pradesh the second stage completed within seven days for about half of Scheduled Caste and Scheduled Tribe transactions but for only 7 per cent of transactions in the residual category. LibTech India’s larger analysis of more than three crore wage transactions for the full year 2021-22 found the same ordering in the second stage: 47 per cent of Scheduled Caste payments, 42 per cent of Scheduled Tribe payments and 33 per cent of other payments credited within the stage’s seven days. The sequencing had a mechanical cause: when payment streams are separated and funds are released in tranches, the streams are not processed simultaneously, and the category whose order is processed later waits, not because any official decided that a particular worker should wait, but because the architecture converted a single queue into three queues of unequal speed. In November 2021, by an office memorandum dated 1 November, the Ministry partially rolled the arrangement back, restoring the single muster and single fund transfer order with category-wise amounts indicated within it for accounting purposes, and the segregated payment arrangement was thereafter discontinued.

The episode matters to this article beyond its own facts, in three ways. It is the clearest proof available that second-stage timing is a function of administrative and funding queues rather than of anything workers do, since the same workers, on the same muster, were paid at measurably different speeds for no reason but the stream their order travelled in. It demonstrates the fragility of a payment architecture that can be re-engineered by circular, without legislative debate, in ways that distribute a statutory benefit unequally by caste, the ground on which the Constitution most forbids distribution. And it shows the correction mechanism working at the speed of publicity rather than of audit: the arrangement was identified, measured and reversed within a year because independent researchers could read the transaction data, an accountability loop that exists only because the programme’s records are public, and one that operated faster than any of the Act’s own institutions.

The payment system’s new gates: Aadhaar-based payment, attendance by application, and the deleted

The second rebuilding of the payment architecture concerns not the order in which wages travel but the address to which they travel and the proof that work was done. Two changes, both taking full effect in the period from January 2023 to January 2024, altered both, and their interaction with the register of workers is the most contested part of the record since 2023. Because Aadhaar sits inside both changes, the legal framework of the identifier itself, its statutory basis and the Supreme Court’s treatment of it, is addressed in the series’ account of the Aadhaar Act and the architecture of identity-linked delivery; what follows is confined to what the changes did inside this programme.

The Aadhaar Based Payment System routes a wage not to a bank account number but to the worker’s Aadhaar number as a financial address: the payment bridge credits whichever bank account the National Payments Corporation’s mapper associates with that Aadhaar. The Ministry’s stated reasons address real defects of the account-based system: workers change accounts, account details recorded at blocks go stale, and payments fail or wander. The system was directed by an order dated 30 January 2023, its deadline extended five times through that year, and made mandatory for all wage payments with effect from 1 January 2024. At the point of mandate, on the Ministry’s own data reported in December 2023, 34.8 per cent of the programme’s 25.25 crore registered workers were still not eligible for Aadhaar-based payment, and 12.7 per cent of the 14.35 crore workers classified as active were not eligible either. Eligibility is a chain of three links, the Aadhaar seeded in the job card, the Aadhaar seeded in the bank account, and the account mapped at the payments corporation, and a break at any link, a name spelled differently in two databases, a closed account never replaced in the mapper, a seeding done at a bank the worker has left, stops the wage entirely rather than delaying it.

The register itself contracted sharply in the same period. Ministry figures reported in January 2024 recorded the deletion of about 7.6 crore workers from job cards in the twenty-one months from April 2022, a deletion rate far above the programme’s earlier years, coinciding with the drive to make every card payment-ready. The Ministry’s position, stated to Parliament, is that deletions follow a standard operating procedure on defined grounds, duplicate cards, cards issued on forged documents, permanent migration, and workers recorded as not willing to work, with verification including the Gram Sabha, and that job cards are not deleted for failure to link Aadhaar with a bank account. Analyses of the deletion data by researchers tracking the programme found that the large majority of deletions in the drive were recorded under the head of workers not willing to work, a ground that sits uneasily beside a register of households actively seeking employment, and field organisations reported workers discovering their deletion only when they next demanded work or when a wage failed to arrive. Documentary effects recorded by LibTech India and other research groups include wages misdirected where an Aadhaar was mapped to another person’s account, payments rejected for mapping failures outside the worker’s knowledge, and genuine workers required to re-register, re-seed and re-verify to restore a status they had never surrendered. The stated purpose of the register cleanup, removing duplicate and ineligible records so that payments reach the right person, is a legitimate administrative object, and a guarantee programme paying ghost cards would fail its own record. The documented cost is that the burden of the state’s data errors was placed on the worker least equipped to correct them, and that the deletion drive’s timing made the register smaller at exactly the moment the payment system’s eligibility demands on it became strictest.

The attendance gate changed in parallel. From 1 January 2023, the Ministry required attendance at worksites to be captured through the National Mobile Monitoring System, a mobile application recording two time-stamped, geo-tagged photographs of the work group each day. The stated purpose is the suppression of bogus attendance and proxy musters, a genuine and long-documented fraud channel. The documented difficulties are of a different texture: worksites without connectivity, a single smartphone, usually the mate’s, carrying the whole group’s attendance, photograph uploads failing and attendance for a day’s real work going unrecorded, and workers, disproportionately older workers and women without phones of their own, made dependent on an intermediary for the proof of their labour. An unrecorded day is an unpaid day, and unlike a delayed wage it generates no compensation claim, because the system’s position is that the work was never shown to happen. The pattern across both changes is consistent, and it is the pattern the evidence discipline of this article requires to be stated in both directions: each change answered a real defect of the older system, and each transferred the cost of the new system’s failures from the administration to the worker, in the currency of wages delayed, rejected or never generated.

The rejected transaction deserves its own line in the record, because it is the failure mode the new architecture produces in bulk and the one least visible in the timeliness statistics: a rejected payment is not late, it has exited the queue, and it re-enters only if someone regenerates it. Rejections arise from the validation chain at the payment bridge and the banks: account closed or dormant, name mismatch between the job card and the account, Aadhaar mapped to a different account, account frozen under know-your-customer rules. The Ministry’s own steps for timely payment, as stated to Parliament in December 2022, include directing states to ensure timely regeneration of rejected transactions and tagging fund releases with priority for them, an official acknowledgement that rejection and regeneration had become a standing workload rather than an exception. The scale in audited pockets is not trivial: a Comptroller and Auditor General audit of the programme in Jammu and Kashmir for the period from 2019-20 to 2023-24 recorded 3,16,178 rejected transactions involving Rs 77.24 crore, of which about ninety per cent, 2,83,196 transactions involving Rs 48.91 crore, were wages of unskilled workers. One Union Territory’s five-year rejection ledger thus holds more failed wage payments than many districts record successful complaints, and each rejection returns the wage to the block for a correction process the worker experiences as a fresh application for money already earned. Rejection rates belong in any honest timeliness account beside delay rates: a system can improve its measured delay while worsening its rejections, and the worker is indifferent between the two failures, since both end in an unpaid month. The record’s verdict on the new gates must therefore be stated whole: they tightened the identity of the payee and loosened nothing else, they reduced certain frauds, and they added a mass failure channel whose remedy lies in data correction work that the system’s design assigns to the person with the least capacity to perform it.

Social audits: the record from below

Against the electronic record stands the Act’s own human audit. Section 17 requires the Gram Sabha to conduct regular social audits of all works in the Gram Panchayat, and the Mahatma Gandhi National Rural Employment Guarantee Audit of Schemes Rules, 2011, framed under Section 24, oblige states to maintain independent social audit units and prescribe the audit’s process. The model implementation, built in Andhra Pradesh through its Society for Social Audit, Accountability and Transparency, professionalised the instrument: trained village auditors, drawn substantially from worker households, verify musters against workers’ testimony in public hearings where officials must answer. Meghalaya in 2017 became the first state to give social audit a statutory basis of its own across public services. The audits’ evidentiary value for this article is that they test the record against the village rather than against itself.

What they find, across states and years, is a consistent distribution. Muster-level irregularities, names of persons who did not work, inflated measurement, work recorded and not executed, recur in audit findings at rates no honest account can wave away, and the 2013 performance audit separately noticed tampering with muster rolls in ten states. The audits’ weakness lies after detection: of the amounts found misappropriated or irregular, the share actually recovered and the share of responsible officials actually proceeded against have persistently lagged the findings, and audit coverage itself, the proportion of Gram Panchayats audited in a year as the rules contemplate, has frequently fallen well short of the mandate in the states without an institutionalised unit. A social audit also has a defined evidentiary limit for this article’s central question: it can verify whether the muster is true, but the wage’s journey after the muster, through the stages where the unrecorded delays accumulate, is invisible to a village hearing. The audit and the transaction data are complementary instruments pointed at different halves of the record, and the programme is unusual in possessing both; it is a commentary on the programme’s governance that the two halves are so rarely read together.

What the record credits

An implementation record that lists only failures would fail its own discipline, because the same data that exposes the second stage also documents real achievements, and the counterfactual, what preceded this programme, is part of the evidence. Three credits stand out.

The first stage of wage payment was transformed. In the programme’s first years, wages travelled through layers of intermediaries and cash handling that generated both delay and leakage as routine features; the staged electronic chain, whatever its measurement boundary, made every payment traceable, dated and attributable, and the Ministry’s generation figures, 89.08 per cent within fifteen days in its May 2017 statement, rising to the mid-nineties in its later statements, record a first stage that improved under measurement, standard operating procedures and judicial pressure. The lesson cuts against a lazy reading of this article: measurement works, which is why the decision about what to measure matters so much.

Direct credit to workers’ own accounts, for all its new gates, ended the era in which a wage could be skimmed by every hand it passed through, and the programme’s wage bill became, in principle, fully reconcilable against named individuals, a property the researchers’ studies themselves exploit. The programme also performed the insurance function its design promised, at the moments of greatest need: in drought years it expanded in the affected districts under the additional fifty days the Act permits, and in the pandemic year it absorbed returning migrant labour at a scale no other instrument attempted, delivering work to 11.17 crore individual workers. Its labour-market effects extend beyond its own musters: the best-known econometric study of the programme’s early years, by Clement Imbert and John Papp, estimated that the guarantee raised earnings in the private casual labour market in the districts where it operated, because an employer of last resort with a posted wage improves every worker’s bargaining position, including workers who never enrol. The insurance credit deserves separate statement because it operates even where the days fall short. A household that knows work can be demanded in the village, at a known wage, within a stated period, holds an option whose value does not depend on exercising it in full: the option disciplines the private labour market’s lean-season wage offers, underwrites consumption credit from lenders who know the entitlement stands behind the borrower, and substitutes for the distress responses, the sold asset, the withdrawn schoolchild, the migration undertaken in desperation rather than by choice, that lean seasons historically extracted. The twenty-year review of the programme’s evidence records effects in this direction across the study literature: consumption smoothing in programme households, reduced reliance on the most exploitative credit, and migration decisions made less desperate where the local programme functioned. Insurance of this kind is systematically undervalued by the delivery statistics, which count the days worked and not the catastrophes that did not happen, and a fair implementation record has to enter the uncounted credit alongside the counted debits, while noting the asymmetry that defines the programme’s tragedy: the insurance value depends on the entitlement being credible, and every unrecorded application, rationed season and two-month wage delay spends the credibility the insurance runs on. The programme’s greatest success and its characteristic failure are thus the same fact at different moments, a guarantee believed in, and a guarantee being taught, season by season, to be believed in less.

A record that omitted these credits would be advocacy, not audit.

The government’s case, the criticism, and a verdict

The government’s position deserves its strongest form, because parts of it are right. The first stage, it says, is where the historical delay lived: in paper musters, absent signatories, treasury queues and cash. Measuring it, timestamping it and publishing it produced the improvement the figures show, and the electronic system is a genuine advance on everything it replaced. The second stage is largely automated; where delay occurs there, the cause is the availability of funds against a demand-driven programme whose wage bill cannot be known in advance, not administrative idleness, and a government that must reconcile an open-ended entitlement with a finite appropriation will sometimes pay late for the same reason any payer with a hard budget constraint pays late. On the register, the government says that a programme paying a quarter of a billion registered names must periodically purge duplicates, the deceased and the fictitious, and that Aadhaar-based payment, by following the person rather than a changeable account, protects workers from the failed and misdirected payments of the older system. On funding, it says allocations are raised through supplementary demands when demand requires, that admissible liabilities are reimbursed, and that no year’s workers have ultimately gone unpaid by the programme as a whole.

The criticism answers each point without denying its kernel. Fund availability is not weather; it is the product of an appropriation the same state proposes and a legislature passes, and to classify second-stage delay as a funding event rather than an administrative failure is to rename the breach, not to answer it. The statute, in any event, does not ask why a wage is late: the compensation entitlement in Schedule II attaches to delay, full stop, and a system that declines to compute compensation for the stage its own government controls has not interpreted the provision but suspended it. The measurement defence proves too much: if measuring the first stage improved it, the refusal to measure the second is a choice about where improvement shall not be demanded. On the register, the criticism accepts the purge and disputes its incidence: the deletions fell heaviest in the compliance drive’s own window, the ground most used was unwillingness to work, and the workers deleted included, on the documented record, people who had worked in the very period of their deletion.

The verdict this record supports is therefore specific rather than general. The programme’s failures of delivery are not principally failures of local honesty or of worker behaviour, and they are not principally failures of technology, which has on balance improved every process it has been honestly applied to. They are failures of two decisions taken at the centre of the system: the decision to fund an uncapped entitlement as though it were a capped scheme, and the decision to bound the measurement of delay at the point where the centre’s own responsibility begins. Both decisions are reversible without amending the Act, the first by appropriation practice and the second by extending a computation the data already supports. Until they are reversed, the accurate one-sentence account of the programme is the one its own figures give when each is returned to its base: a guarantee that delivers, to the households it reaches, about half its promised days; pays their wages through a chain whose first half is measured and prompt and whose second half is neither; and owes, for the unmeasured half, a compensation it does not calculate. This is the series’ counting discipline applied for the sixth time, and the second in which the divergence between official and independent accounts is architectural rather than arithmetical: as with the enforcement record of the anti-terror statutes, where the count depends on the stage of the process at which counting stops, and the food statute’s distribution record, where the same grain is counted at different points of the chain, the argument ends not when one side’s arithmetic defeats the other’s but when each figure is restored to the boundary that produced it.

The design implications follow from the diagnosis, and they are unusually concrete for a programme of this scale, because the defects are specific rather than atmospheric. The measurement repair is a software decision: compute delay and compensation across the whole fifteen days, on data the system already holds, and credit compensation automatically with the delayed wage, as the formula’s design intends; the objection that this would expose the Union to a compensation liability it has not budgeted is the confession that the liability exists. The funding repair is a budgeting decision: place the programme’s appropriation on the footing the entitlement requires, by estimating full-demand cost honestly, releasing against it automatically rather than in negotiated tranches, and treating the rollover of liabilities as the debt it is; the People’s Action for Employment Guarantee’s costing exercises exist precisely to make that estimate, and a legislature that prefers a lower number can vote it, but the vote should be visible as a decision to ration a right rather than absorbed silently into queues. The demand repair is an institutional decision: take the recording of applications out of the hands of the office that pays for unmet demand, through independent registration channels and automatic receipts, so that the unemployment allowance’s trigger is no longer guarded by its debtor. None of these repairs requires amending the Act. All three have been recommended, in substance, by the programme’s own auditors, researchers or courts. That they remain unmade is the final datum of the implementation record, and it converts the article’s verdict from a finding about administration into a finding about choice.

For the researcher or journalist working this record, the practical instruments are the transaction data itself, which the programme publishes, and a private discipline of source-keeping: extracts retained with their dates of extraction, the base of every percentage noted beside the figure, and the correspondence with ministries preserved, a working file best kept in an encrypted workspace designed for journalists to protect sources and research material, because the figures in this field are contested precisely in proportion to their consequences. For the worker, the payload is simpler and has run through every section: the entitlement is dated, receipted and compensable on paper, and the paper, the dated application, the job card, the muster entries, is the whole of the worker’s leverage, to be created, kept and copied at every step. The statute kept its side of the drafting bargain; it wrote the promises in numbers. The record shows a system that has learned to keep some of those numbers, and chosen, so far, which ones.

Frequently Asked Questions

Q: How many days of MGNREGA work does a household actually get?

The statute guarantees up to one hundred days per household per financial year, but delivered days run at about half that. On the Ministry of Rural Development’s figures laid before the Rajya Sabha, the average days of employment per household that obtained work were 51.54 in 2020-21, 50.07 in 2021-22, 47.84 in 2022-23, 52.07 in 2023-24 and 50.24 in 2024-25. The distribution matters more than the average: on LibTech India’s computation from the programme’s own data, only 7.6 per cent of households that worked in 2023-24 completed the full one hundred days, and 7.0 per cent in 2024-25. Many households demand fewer than one hundred days and receive what they demand; others are rationed by opened works and funds. The honest summary is that the guarantee functions as a ceiling few reach rather than a norm most receive.

Q: Why do official and independent MGNREGA wage delay figures differ?

Because they measure different halves of the same journey. A wage payment runs in two stages: the state must close the muster, prepare the wage list and generate and sign the fund transfer order within eight days, and the Union must then process the order and credit the worker’s account within a further seven days, fifteen in all. The programme’s information system computes delay for the first stage only, so official percentages, such as the Ministry’s statement that over 96 per cent of fund transfer orders are generated within fifteen days, describe generation, not credit. Independent analyses of the transaction data count the whole journey: LibTech India found only 21 per cent of payments completed within fifteen days in 2016-17. Both figures are accurate on their own bases; only the second answers the statute’s question.

Q: Is delay compensation actually paid?

Barely. Schedule II entitles a worker to compensation at 0.05 per cent of the unpaid wages for each day of delay beyond the sixteenth day after the muster roll closes, and the Supreme Court in Swaraj Abhiyan directed that it be paid. Three defects defeat it in practice. The information system computes compensation on the first payment stage only, so compensation for the Union’s stage is never calculated; what is not computed cannot be claimed. The amount per worker is small, typically tens of rupees, so individual claiming is irrational. And the state must pay upfront and recover from the officials responsible, which builds resistance into the payment point. The Ministry itself recorded total delay compensation paid across the country in 2024-25 at Rs 5.27 lakh, against a programme paying wages on hundreds of crores of person-days.

Q: Why is unemployment allowance almost never paid?

The allowance under Section 7, at least a quarter of the wage rate for the first thirty days and half thereafter, accrues only where a worker applied for work and fifteen days passed without work being provided. Its trigger is a dated, recorded application, and that is where it fails. The Comptroller and Auditor General’s 2013 performance audit found dated receipts for applications not given in over a third of test-checked Gram Panchayats and employment registers unmaintained in over half, so unmet demand left no paper from which an allowance could be computed; the audit found the allowance unpaid in 47,687 cases. The allowance is also the state government’s own liability, outside the Union’s wage funding, so the office that must record the application is the office that pays when the application goes unmet, an incentive no amount of instruction has overcome.

Q: How do MGNREGA wages compare with state minimum wages?

They began equal and were then separated. From 2006 to 2008 the programme paid the state minimum wage for agricultural labour. In January 2009 the Union used Section 6 of the Act, which operates notwithstanding the Minimum Wages Act, 1948, to notify its own rates, frozen at the levels then in force and revised annually thereafter by the Consumer Price Index for Agricultural Labourers. The result is that the notified programme wage sits below the state minimum wage in a large number of states, and below prevailing agricultural earnings in most. A committee chaired by Mahendra Dev recommended relinking the programme wage to the state minimum wage; the recommendation was not adopted. The Parliamentary Standing Committee on Rural Development recommended in February 2024 that programme wages be at least Rs 375 a day; the highest notified rate for 2024-25 was Haryana’s Rs 374. States may top up the notified rate from their own funds.

Q: Has Aadhaar based payment excluded workers?

The evidence supports a careful yes, alongside the government’s denial that exclusion is the system’s purpose. Aadhaar Based Payment, mandatory for all wage payments from 1 January 2024 after five deadline extensions, routes wages to whichever account the payments corporation’s mapper associates with the worker’s Aadhaar, so a break in the chain of seeding and mapping, a differently spelled name, a closed account, stops the wage entirely. On the Ministry’s own data reported in December 2023, 34.8 per cent of registered workers and 12.7 per cent of active workers were not yet eligible for the system when it became compulsory. Ministry figures reported in January 2024 recorded about 7.6 crore workers deleted from job cards in the twenty-one months from April 2022, during the compliance drive. The Ministry states that deletions follow verification on defined grounds and are not made for failure to link a bank account; researchers documented genuine workers deleted, and wages misdirected or rejected through mapping errors.

Q: Do social audits change anything?

They change the evidentiary position, which is a precondition of everything else. Section 17 requires the Gram Sabha to audit all works, and the Audit of Schemes Rules of 2011 oblige states to maintain independent social audit units; Andhra Pradesh’s dedicated society became the model, training auditors drawn from worker households to verify musters against workers’ testimony in public hearings. Audits consistently surface muster-level irregularities, ghost entries, inflated measurement and unexecuted work, and the Comptroller and Auditor General’s 2013 audit separately noticed muster tampering in ten states. The limits are after detection: recovery of amounts found irregular and action against responsible officials lag the findings persistently, coverage falls short of the annual mandate in states without a strong unit, and an audit can verify a muster but cannot see the wage’s later journey through the payment stages. Audits discipline the worksite half of the record; the payment half needs the transaction data.

Q: Is MGNREGA funding adequate to meet demand?

Not at full demand, on any published costing. The entitlement is uncapped: every rural household may demand one hundred days, and more in notified calamity areas. The money is an annual appropriation filtered through approved district labour budgets. The People’s Action for Employment Guarantee has estimated that meeting full demand would require an allocation several times the recent level, and that roughly a fifth of the funds available in a year are absorbed by the previous year’s unpaid liabilities before new work begins. Central expenditure on the programme has run between 0.3 and 0.4 per cent of gross domestic product since 2015-16, apart from the pandemic year, on Jean Dreze’s twenty-year review. The allocation for 2025-26 was held at Rs 86,000 crore while wage rates rose and registrations grew. The Ministry’s position is that the programme is demand-driven and that supplementary funds are sought when needed; the record of exhausted allocations, rolled-over arrears and mid-year crises describes a budget that follows demand late rather than funding it in advance.

Q: What is a fund transfer order in MGNREGA wage payment?

A fund transfer order, or FTO, is the electronic payment instruction that carries a wage from the programme’s records to the payment system. After a muster roll is closed and the work measured, the block generates a wage list naming each worker, the days worked and the amount due, and the fund transfer order is generated from that list, carrying the workers’ names, account or Aadhaar details and amounts. It requires two digital signatures at the block level before it is pushed to the Union’s platform, where it is processed through the Public Financial Management System and the payment bridge to the worker’s account. The FTO is the hinge of the payment architecture: its generation ends the state’s stage of the process, its credit ends the Union’s, and the programme’s delay statistics turn on which of those two events is being counted.

Q: What is the National Electronic Fund Management System and how did it change wage payment?

The National Electronic Fund Management System, rolled out from April 2016, replaced state-level electronic payment arrangements with a single centralised chain. Before it, funds released by the Union were held by states and districts and paid out through their treasuries and banking channels, so a delayed wage could be attributed in a circle among block, district, state and Centre. Under the centralised system, the fund transfer order travels electronically from the block to the Union’s payment platform, every hand-off is timestamped, and wages are credited directly to workers’ accounts by the Centre. The reform made each payment traceable and fixed responsibility for delay by stage, which is a genuine administrative advance. It also removed the states’ ability to pay from contingency funds while awaiting Union releases, centralising the funding queue along with the paperwork, so that when Union funds run short the stoppage is national and immediate rather than local and cushioned.

Q: Why are wages still delayed after the fund transfer order is signed?

Because signature completes only the paperwork; the money must still move. After the second digital signature, the order is processed and sanctioned on the Union’s platform against the funds released for that state. If the state’s programme balance is exhausted, signed orders queue unsanctioned until further releases arrive, however complete their documentation: this funding queue is the largest single cause of second-stage delay in the independent analyses, with the Union’s processing alone averaging over fifty days in the 2016-17 transaction study. The second cause is data failure: orders whose account or Aadhaar mapping fails validation are rejected at the payment bridge and return for correction and regeneration. Neither cause is visible in the official delay measure, which stops at the order’s generation, and neither attracts delay compensation, because compensation is computed on the same truncated measure.

Q: Who pays delay compensation and unemployment allowance under MGNREGA?

Different governments, by design. Delay compensation under Schedule II follows responsibility for the delay in principle, but its payment mechanics rest with the state: the Minister of State for Rural Development told the Rajya Sabha in December 2022 that the state government shall pay the compensation upfront after verification and recover the amount from the functionaries or agencies responsible for the delay. Unemployment allowance under Section 7 is wholly the state government’s liability, which Section 22’s funding pattern confirms: the Union bears the entire unskilled wage cost and three-fourths of material cost, while the state bears the remaining material share and the allowance. This division is a large part of both entitlements’ history. The allowance is owed by the government that records the application, and compensation must be advanced by a government that then has to pursue its own officials or the Union’s agencies for recovery.

Q: Why were MGNREGA wage payments separated by caste category in 2021?

By a Ministry circular dated 2 March 2021, which required that workers on a single muster roll be paid through three separate fund transfer orders, for Scheduled Caste, Scheduled Tribe and other workers, instead of one. The Ministry’s stated purpose was accounting: to reflect accurately the flow of funds to different population groups, in line with budget heads that track expenditure under the Scheduled Caste and Scheduled Tribe components. The effect, measured by LibTech India in the transaction data, was that the categories were paid at systematically different speeds: in the full year 2021-22, the second stage was completed within its seven days for 47 per cent of Scheduled Caste payments, 42 per cent of Scheduled Tribe payments and 33 per cent of other payments. Parliament’s Standing Committee on Rural Development asked for a rollback, and by an office memorandum of 1 November 2021 the Ministry restored the single fund transfer order, with category amounts shown within it; the segregated arrangement was discontinued.

Q: What happens when a worker’s wage transaction is rejected?

The wage leaves the payment queue and waits for human correction. A transaction is rejected when validation fails at the payment bridge or the bank: the account is closed, dormant or frozen, the name on the job card does not match the account, or the Aadhaar is mapped to a different account than the worker expects. The rejection returns down the chain to the block, where the record must be corrected and the transaction regenerated before the wage travels again. The Ministry has directed states to ensure timely regeneration of rejected transactions and to tag fund releases with priority for them. The volumes are large: a Comptroller and Auditor General audit in Jammu and Kashmir recorded 3,16,178 rejected transactions between 2019-20 and 2023-24, about ninety per cent of them unskilled workers’ wages. A rejection is not counted as a delay in the official statistics, and no compensation attaches to it, although the worker’s experience, an earned wage unpaid for weeks, is identical.

Q: Do women receive their full share of MGNREGA work?

They receive more than the share the Act sets as a floor. The Act requires that at least one-third of the workers be women, and the delivery record has run well above that for years: on the Ministry’s figures laid before Parliament, women were 4.60 crore of the 8.34 crore workers who obtained employment in 2023-24, about 55 per cent, and the share has held above half across recent years. Equal wages for men and women are prescribed by the Act itself, and the notified rate does not differ by sex, which distinguishes the programme from the private casual labour market, where women’s daily earnings run well below men’s. The design features that produce the share are self-selection at a wage that does not displace male migration earnings in many regions, work within the village, and payment into the woman’s own account where the account is hers. The record on this measure is the programme’s clearest delivery success.

Q: How does MGNREGA affect agricultural wages outside the programme?

It raises them, on the best-known evidence. A statutory employer of last resort offering a posted wage within the village improves the bargaining position of every casual worker, including those who never enrol, because the outside option against which private employers bid has improved. The most cited econometric study of the programme’s early years, by Clement Imbert and John Papp, estimated that the guarantee raised private-sector earnings in the districts where it operated, with gains concentrated in the agricultural lean season when the programme’s alternative is most valuable. The effect is one reason landowner opposition to the programme has been a recurring feature of its politics, and one reason economists treat the programme’s cost differently from a pure transfer: part of the wage bill returns as higher earnings for workers who never appear on a muster. The size of the effect varies with how credibly the local programme delivers; a guarantee that pays two months late disciplines private wages far less than one that pays on time.

Q: What is the National Mobile Monitoring System in MGNREGA?

It is the mobile application through which attendance at programme worksites must be captured. From 1 January 2023 the Ministry required attendance to be recorded through the application with two time-stamped, geo-tagged photographs of the work group each day, replacing the paper muster as the system’s proof that work was done. The stated purpose is the suppression of bogus attendance and proxy musters, a documented fraud channel under the paper system. The recorded difficulties are practical: worksites without reliable connectivity, dependence on a single smartphone, usually the mate’s, for the whole group’s attendance, failed uploads that leave a day’s real work unrecorded, and workers without phones of their own, disproportionately older workers and women, made dependent on an intermediary for proof of their labour. An unrecorded day generates no wage and no compensation claim, because the system’s position is that the work was never shown to happen.

Q: Why did MGNREGA work stop in West Bengal?

Because the Union stopped the money. Section 27 of the Act empowers the Central Government to issue directions to a state and to withhold fund releases where the state fails to comply with the Act’s requirements. From 2022 the Union stopped releases to West Bengal under this power, citing irregularities in the state’s implementation; the state government disputed the stoppage and the matter was litigated. The consequence in the programme’s data, recorded in LibTech India’s national review, is that no employment was generated in West Bengal in 2023-24 or 2024-25: a programme that is a statutory entitlement for households was absent from one of its largest states for two full financial years. The episode is the funding architecture’s limiting case. It shows that the guarantee’s practical existence depends on a flow of releases between two governments, and that a compliance dispute between those governments can suspend the entitlement of households who are parties to neither side of the dispute.

Q: What records should a worker preserve to claim unpaid wages or compensation?

The records that start and prove the clocks. First, the job card itself, with its entries of work demanded, work given and wages due, which the household holds and which should be updated at every transaction. Second, proof of the application for work with its date: the dated receipt the panchayat or Programme Officer is required to issue, since the fifteen-day period for providing work, and with it the unemployment allowance, runs from that date, and an oral application leaves the allowance unprovable. Third, the muster and payment particulars for each spell of work: the dates worked, the closure of the muster, and the date the wage was actually credited, from the passbook or account statement, since delay compensation runs from the sixteenth day after muster closure. A worker pursuing a specific claim should take these documents to the Programme Officer, the district Ombudsman and, if those fail, the district legal services authority; general accounts of the programme cannot substitute for advice on an individual case.

Q: How did MGNREGA perform in the pandemic year of 2020-21?

At a scale the programme has not matched in any other year, because demand and funding rose together. As reverse migration carried workers back to their villages during the lockdowns, the Union raised the programme’s funding substantially in mid-year, and the system absorbed the returning labour: on the Ministry’s figures given to Parliament, 11.17 crore individual workers obtained employment in 2020-21, against 8.75 crore in 2022-23 and 8.34 crore in 2023-24. The average days per household reached 51.54 even as the number of households swelled with new demand, near the top of the programme’s historical range. The year functions as the record’s control experiment. It demonstrates that the delivery machinery can expand enormously when appropriations and administrative priority arrive together, which refutes the claim that the ordinary shortfall, averages near fifty days against a hundred-day guarantee, is a fixed property of the programme rather than of its funding.