Safety, Institutions and the Environment

The complete report has seven parts and one 204-page master PDF. The research cutoff for this part is 3 October 2026. Every figure retains its own observation year. Page references in the original reading guide below refer to this volume’s printed pages within the master PDF; use the contents here to move through the web edition.

The numbers touch real lives.

Safety, justice, clean air and water belong in the same report card.

A road can get wider while the number of people dying on it rises. A court can put its records online while cases keep waiting. A city can have cleaner air than before and still have air that falls short of the standard. I wanted this part to explain those situations without making you fight through a pile of official language.

I write on MisterLove because I like taking a difficult subject apart until its meaning becomes clear. My work moves between research, business and technology. In this volume, the most useful question is often a small one: does this number count people, cases, places, capacity or something else?

The Matrix India posts started the series. The tribal-crime chart is one I return to here. I checked its entries against the original NCRB table, included the places it left out and explained why an empty rate cannot be turned into zero.

This is a before-and-after report, with 2014 as the organising point. It is not a claim that every change came from one government. Reporting rules, population, state administration, courts, communities and long-running programmes all belong in the explanation.

I also do not want an old observation to pass as a new one. A report released in 2026 can describe crimes in 2024 or emissions in 2022. Where I could not find a fair pre-2014 comparison, I say so and use the available dated comparison.

Start with the subject you care about. The table gives the numbers. The paragraphs explain each row, then show what remains unanswered. You should not need a degree, a political loyalty or a love of long reports to follow it.

Lovepreet Singh
MisterLove

Sources: 1, 2.

A number needs a meaning.

The date, the unit and the coverage are part of the number.

One lakh means 100,000. One crore means 10 million. A rate per lakh compares a count with a population of 100,000. If a place has 20 recorded cases among two lakh people, its rate is 10 per lakh. A case is not necessarily one victim.

A percentage point is the gap between two percentages. Going from 60% to 70% adds 10 points. It is about 16.7% growth compared with the earlier 60. A rate and a growth percentage answer different questions.

FY means financial year, April to March. Calendar-year crime totals cover January to December. Court and prison figures often count the stock on one date. A stock of pending cases cannot be added to annual new cases as if they were separate events.

The research cut-off is 3 October 2026. This part covers the main national measures in these subjects and the supplied tribal-crime chart. Parts 1-4 cover the economy, infrastructure, living conditions and production. Part 6 will examine international rankings, state differences and the wider chart audit.

Latest means the latest observation verified for that measure, not a completed statistical account of every day in 2026.

The total changed. So did the rules.

Police records count registered cases, not every offence that happened.

Cognisable means the kind of offence for which police can normally investigate without first obtaining a magistrate's order. IPC is the old Indian Penal Code. BNS, the Bharatiya Nyaya Sanhita, replaced it from 1 July 2024. Special and Local Laws include offences under separate central or state laws.

The first row is the main criminal code. The second is the other laws collected in this report. The third adds those two groups. Do not add the total again. The later calendar year combines IPC cases from the earlier part of 2024 with BNS cases from the later part.

NCRB also changed its collection. It excluded the heads for simple hurt under IPC Section 323, food and drug adulteration under Sections 272-276, and criminal intimidation under Section 506. Other forms of hurt remain. A lower total therefore cannot be read as a clean measurement of less actual crime.

One FIR can contain several offences. NCRB's principal-offence rule usually counts it under the most serious offence. An FIR is the police's First Information Report; it is not a court finding that somebody is guilty.

For IPC/BNS cases, the 2024 conviction rate is 53.3%. Its denominator is cases in which trial was completed, not all complaints or all registered cases. Cases still waiting for trial are outside that calculation. Reporting, investigation and judgment are different stages.

Definitions and collection changes are in the report itself. More registration can reflect better reporting as well as more offences.

Sources: 3, 4, 6, 7.

Fewer registered murder cases.

A case count and a population rate tell two parts of the story.

The 2024 count is 6,152 lower than in 2013, a fall of about 18.5%. In plain terms, an earlier count of 100 became about 81. The count also fell by 672 between 2023 and 2024, about 2.4%.

The rate adjusts for the size of the population. The reported rate fell from 2.7 to 1.9 cases per 100,000 people. Those published rates are rounded. They help distinguish a change in the number of cases from a change relative to the number of people.

These are murder cases registered by police. A single case can involve more than one person killed. In 2013, NCRB separately reported 33,901 murder victims; replacing the earlier case count with that victim count would mix two units.

NCRB records disputes as the most common stated motive in 2024, in 9,607 cases. Personal vendetta or enmity accounted for 3,638; gain for 1,460. These are police classifications of cases, not a complete psychological explanation of every killing.

This is a clear decline in the registered murder series. It does not mean violence of every kind fell by 18.5%, or that every district became safer. Assault, domestic abuse, sexual violence and offences never registered need their own evidence.

A national rate also spreads very different local experiences into one average. Someone living in a district facing violence can face a worse situation even when the country's total falls. The table describes the countrywide record, not a promise about an individual's safety.

Cases, victims and accused persons are separate units. Population rates here use NCRB population estimates for the stated years.

Sources: 3, 4.

Different offences moved differently.

Read the whole category and the rape row separately.

The first row covers several offences against women. It rose by about 42.6%. The second compares those cases with the female population: about 52 cases per lakh women became about 65. This rate is not calculated using all men and women together.

The rape row fell by about 12.4% in the registered-case series. The cruelty row remained around 1.2 lakh cases. A fall in rape cases and a rise in the wider category can both occur, because the wider category includes many other offences.

The latest all-crime count is below 2023's 4,48,211. That one-year fall does not erase the longer-period rise. Likewise, the long-period rise does not tell us that registration increased in every intervening year.

The law changed around this baseline. The 2013 criminal-law amendment broadened sexual-offence provisions, and POCSO introduced a separate framework for children. Legal definitions, willingness to complain and police recording affect comparisons across time.

The figures leave out violence that never becomes a registered case. They cannot establish that an unreported assault did not happen. They also cannot tell us how much of a rise came from more violence and how much from more reporting.

A police complaint is not proof of guilt. An acquittal is not, by itself, proof that the complaint was invented. Investigation and the evidence required in court determine the outcome. The table is useful precisely when we keep those different questions apart.

Women and child categories overlap; adding them would double-count some cases. These are registered cases, not a victimisation survey.

Sources: 3, 4.

Much more crime is being recorded.

The category grew, with kidnapping and POCSO cases making up large shares.

Children means people below 18 in this report. The all-crime count is about 3.22 times the 2013 count. The rate is also higher, so population growth alone does not explain the larger recorded category.

The last two rows break down 2024. Kidnapping and abduction made up about 40% of recorded child-crime cases. POCSO made up about 36.9%. POCSO means Protection of Children from Sexual Offences. These rows are parts of the total, not extra cases to add to it.

The POCSO law began in 2012, close to the earlier comparison point. The legal framework and the detail collected expanded over the longer period. The rise cannot honestly be turned into a claim that the same narrowly defined risk became exactly 3.22 times greater.

A registered kidnapping category also needs the case's facts. It cannot be read as a count of children kidnapped by strangers alone. The statistical label does not describe who the accused was or what happened in every case.

The total rose from 1,77,335 in 2023 to 1,87,702 in 2024, about 5.8%. This is a continuing rise in the latest registered total. It deserves attention even while definitions and reporting are kept in view.

Girls can appear in both the women and children chapters. One event may therefore be described in more than one victim-group table. Those chapters help us understand different groups; they are not separate piles that can all be added to calculate a new national crime total.

A dash in the 2013 breakdown means no matched figure is supplied here. It does not mean zero offences occurred.

Sources: 3, 4.

The national totals need context.

The tribal-crime chart is about a specific recorded category.

Scheduled Castes, or SCs, and Scheduled Tribes, or STs, are legally recognised groups. The later SC count is about 41.3% above 2013. The later ST count is about 46.7% above 2013. Both are lower than their 2023 counts.

The recent tables concern crimes and atrocities against these groups within NCRB's stated scope. The ST table says atrocities concern offences committed against STs by non-SC/ST persons. It excludes IPC-only cases without the SC/ST Act where the offender is from SC/ST groups. It is not a count of every crime involving a tribal person.

The legal framework and reporting coverage changed over the long comparison, including amendments to the atrocities law. These before-and-after totals should not be treated as an unchanged survey of all violence, discrimination or actual risk.

The published 2024 rates are 27.7 per lakh SC people and 9.5 per lakh ST people. Their denominators still come from the 2011 Census. They are not rates calculated from a fresh 2024 or 2026 count of those populations.

The ST national total fell by 2,994 from 2023 to 2024. Manipur alone fell from 3,399 to 289, a reduction of 3,110. Across the remaining places together, the net change was an increase of 116. This arithmetic shows how much one state's swing shaped the national change.

It does not tell us why that swing occurred. The next two pages reproduce all 36 places in the official table, sorted by published 2024 rate, and explain the errors in the supplied chart.

The rate is specific to the SC or ST population. The 2011 denominator and legal scope are essential parts of the comparison.

Sources: 2, 3, 5.

Tribal crime, place by place.

Registered cases by year. Rate means cases per lakh ST people in 2024.

Kerala has the highest published rate, 29.9, with 145 cases. Madhya Pradesh has the largest count, 3,165, with a rate of 20.7. A smaller population can produce a higher rate from fewer cases. These are recorded rates in this specific category, using ST population from the 2011 Census. They are not a ranking of every kind of danger faced by tribal people.

Sorted by the published 2024 rate, highest first. Rounded rates should not be recalculated from the table's rounded population figures.

Sources: 2, 5.

The rest of the same table.

An unpublished rate stays unpublished. It is different from a recorded zero.

The supplied chart gives Punjab a rate of 0.2. NCRB prints no rate. It also prints no rate for Chandigarh and Puducherry, where the chart shows zero. Haryana and Delhi are missing from that chart; both appear here. The five 'No rate' places have no notified ST population denominator in the cited Census table. Their recorded cases remain visible.

A recorded zero does not establish absence of unreported violence. This category and its legal scope are explained on page 8. All 36 places are included.

Sources: 2, 5.

The recorded count rose sharply.

A growing digital economy also gives crime more digital routes.

The two 2013 rows add to 5,693. The Information Technology Act covers specified digital offences; some cyber-enabled offences were counted under IPC. The 2024 total is about 18 times that combined earlier count. It rose about 17.9% from 2023.

That large increase is in registered cases. The newer framework collects a broader range of offences, while digital use and opportunities to report have also expanded. Dividing the totals alone cannot show that the risk for each internet user rose 18 times.

Fraud was the recorded motive in 73,987 cases in 2024, or 72.6% of the cybercrime total. A motive category is not the amount of money lost. One case can involve several victims, and a complaint to an online portal is not automatically a police FIR.

The government says its financial-fraud response system saved more than Rs 11,158 crore across over 32.80 lakh complaints by 30 June 2026. These are cumulative figures for the system launched in 2021. They are not the number of 2024 FIRs, an annual loss estimate or proof that all the money reached victims again.

For someone facing an online financial fraud, 1930 and cybercrime.gov.in are the official reporting routes. Prompt reporting helps the response system try to stop funds moving further. The larger report-card question is whether prevention, investigation and recovery keep pace with the growth in recorded fraud.

NCRB case counts and the cybercrime portal's cumulative complaints use different units and cannot be added.

Sources: 3, 4, 5, 8.

Similar crashes. More deaths.

The long-term rise in deaths is much larger than the rise in reported crashes.

Reported crashes rose by about 0.3% between these years. Deaths rose by 39,603, or about 28.8%. The injury count was lower. These rows count different things: an event, people killed and people injured.

The last row divides deaths by reported crashes. It went from about 28 deaths per 100 crashes to about 36. This does not mean each crash has a 36% chance of killing someone. One crash can kill several people. It is a measure of the seriousness of the reported crash record.

Deaths also rose from 1,72,890 in 2023 to 1,77,175 in 2024, about 2.5%. The latest year therefore continued the recent rise. A longer or wider road network has not automatically delivered fewer road deaths.

More vehicles and more travel change exposure. A per-kilometre risk calculation would need the distance travelled, not only these totals. The annual deaths remain important even without that calculation: they measure lives lost.

Road design, speeds, vehicle protection, enforcement and emergency care can all influence the outcome. This table cannot allocate the increase between those causes. Nor can it show that every highway or state moved in the same direction.

This page keeps the ministry's historical series together. NCRB has a separate road-death total of 1,75,142 for 2024. Different reporting systems can produce different totals. Adding those two counts would largely count the same deaths twice.

MoRTH = Ministry of Road Transport and Highways. All main rows use its series, rather than switching sources between endpoints.

Sources: 9, 10.

The long-term count is higher.

The latest small fall follows several years of rising recorded deaths.

The 2024 count is 35,947 above 2013, about 26.7% higher. The reported population rate is also higher, from 11.0 to 12.2 per lakh. More people in the country do not explain the whole increase in the recorded rate.

The 2019 row shows why two endpoints do not tell the entire path. Its count was slightly above 2013, but its rate was lower. Recorded deaths then rose substantially before the latest small fall.

Between 2023 and 2024, the count fell by 672, about 0.4%. That is a change in the latest year, not a reversal of the full rise since 2013. Both facts belong on the page.

NCRB compiles suicides reported to police. These are not counts from a complete clinical survey of mental health. Differences in recognition and reporting affect the record; the figures do not cover every unrecorded death.

The report lists stated causes and occupations. A police-recorded cause is not a medical diagnosis or a complete account of a person's life. Several pressures can exist together. An occupation category says what work the person did, not that the occupation itself caused the death.

These numbers describe a serious loss of life. They support looking closely at prevention, accessible care and social pressures. They do not support blaming every death on a single policy, exam, occupation or personal failing.

No methods of suicide are described. The earlier count and rate are confirmed in the parliamentary annexure; later values use ADSI 2024.

Sources: 10, 11.

More personnel. Many vacancies.

A sanctioned post is permission to employ someone; an actual post is filled.

The number of personnel actually serving rose by 4,39,650, about 25.5%. The approved establishment also grew. The first row is the intended staffing ceiling; it does not mean all those officers were already working.

Subtracting actual staff from sanctioned staff gives the vacancy count. The later shortfall is 31,978 larger in absolute numbers, but smaller as a share of the expanded establishment: about 21.5% instead of 24.6%. Both statements can be true.

Using the report's projected population, the 2024 actual strength is about 155 personnel per lakh people. This includes civil police, District Armed Reserve, Special Armed Police and India Reserve Battalions in the state and UT establishment. Central forces are reported separately.

Women numbered 2,72,535 in that establishment, about 12.6% of its actual strength. A national share does not show whether every station has enough women, whether they hold leadership positions or which duties they are assigned.

This publication describes 1 January 2024, although it was published in 2025. It is not a live count of staff in October 2026. Police and public order are primarily state responsibilities, so staffing differs widely between states.

Staff numbers are one part of capacity. Training, time spent on other duties, investigation skills, equipment and the ability to register a complaint also matter. A larger force alone cannot tell us whether an investigation was fair or finished promptly.

Percentages are calculated from the cited sanctioned and actual totals. The comparison excludes central-force totals from both sides.

Sources: 12, 13.

Many more cases are waiting.

Digital access expanded while the stock of pending cases grew.

High Court pendency rose by about 44.9%. District and subordinate court pendency rose by about 86%. These are stocks of cases awaiting disposal on the two dates. They are not the number newly filed in each year.

The Supreme Court had another 95,954 pending cases in the July 2026 snapshot. Adding the three levels gives 5,64,74,844, about 5.65 crore. No matched 2013 Supreme Court figure is inserted into the table, so it does not pretend to offer an all-level 2013 comparison.

The queue can grow even when courts dispose of many cases, if incoming cases outnumber completed ones. Disposal is a flow over time; pendency is the stock left at a date. A pending case is also not necessarily a criminal case: these totals include different kinds of disputes.

The August 2026 reply reports more than 4.18 crore video hearings, about 1.25 crore electronic filings and 2,584 eSewa centres. These are cumulative service figures. A hearing is not a separate completed case, and several hearings can concern the same case.

At 30 June 2026, the reply records 21,411 working district/subordinate judicial officers against 30,825 sanctioned posts. That gap is another part of the capacity picture. Courtrooms, staffing and case management need to be considered together.

Online filing and access can save journeys and make a case easier to track. They do not prove that waiting times disappeared. To measure speed directly, we need the age and duration of cases as well as the number pending.

Latest pendency is the dated parliamentary NJDG snapshot. It is not a claim to be the live dashboard reading on 3 October 2026.

Sources: 14, 15.

Prisons remain over capacity.

Most people in prison are undertrials, awaiting the outcome of proceedings.

Capacity is the number of people the prison system is designed to hold. It grew by about 30.4%. The number held grew by about 23.5%. Capacity grew faster, so national occupancy fell, but it remained above 100%.

An occupancy rate of 112.2% means roughly 112 people for every 100 places. It does not mean every prison has exactly that crowding. Some are below capacity; others face much greater pressure.

The 2024 values come from NCRB's revision dated 15 September 2026. They count 5,09,012 people. Earlier published 2024 figures quoted elsewhere can differ. This page uses the revised counts and their matching occupancy rate together.

Undertrial prisoners are held while proceedings are ongoing; they have not been convicted in the case for which they are held as undertrials. Their share rose from 67.6% to 72.5%, or 4.9 percentage points. Prison population must not be described as a count of convicted offenders.

The other categories include convicts and people detained under other legal provisions. The undertrial row is part of the total, so adding it to the inmate count would double-count people.

These are year-end snapshots. They do not count everyone admitted over the entire year, and they cannot alone measure sentence length, bail access or conditions inside a particular jail.

The 2024 report was revised in 2026; its observation remains 31 December 2024. Undertrial status is not a finding of guilt.

Sources: 16, 17.

Two ways citizens take part.

Requesting public information and casting a vote need different measures.

RTI means Right to Information. The Act, passed in 2005, lets people request information from public authorities. The cited rejection rate concerns the Union authorities reporting through the central system. It is not the combined rate for every state authority.

The rejection rate fell by 3.95 percentage points. But an application not rejected is not necessarily a complete, useful answer. The February 2026 reply says separate data on partially answered or unresolved applications are not collected.

The Central Information Commission had 29,034 appeals and 3,577 complaints pending on 1 February 2026. Those are different types of pending matters, not the annual number of RTI applications. A low initial rejection rate can coexist with a queue of disputes.

Turnout is votes cast as a share of registered electors. The Election Commission's atlas shows 58.21% in 2009, 66.44% in 2014, 67.40% in 2019 and 66.10% in 2024. The latest share is above 2009 but below the previous two general elections.

The 2014 election belongs to the transition year, so the table uses 2009 as its pre-2014 observation. The atlas's total includes the final voting record; an earlier polling-day percentage should not replace it without saying the coverage changed.

Participation, access to information, case delays and police capacity each describe one part of institutions. None of these figures is a complete democracy or corruption score. International indices and their methods are examined separately in Part 6.

RTI uses financial years; voting uses the named general election. Turnout is not a share of all residents, because only registered electors form its denominator.

Sources: 18, 19.

Improvement is not the finish line.

A city can reduce pollution and still fail an air-quality standard.

NCAP is the National Clean Air Programme, launched in 2019. Its comparison uses 2017-18, not 2014. These results describe its 130 cities, not a national average weighted by how many people breathe each city's air.

PM10 is airborne particulate matter up to 10 micrometres across; PM2.5 is the smaller fraction up to 2.5. Their annual averages are measured in micrograms per cubic metre of air. PM2.5 is part of PM10, so the two concentrations should not be added.

The 72 cities with reductions above 20% and the 26 above 40% sit within the 108 improved cities. These groups overlap. Adding 108, 72 and 26 would count many cities repeatedly.

Only 14 met India's annual PM10 limit of 60, compared with six in the baseline year. The annual PM2.5 limit is 40. WHO's 2021 health-based annual guidelines are lower: 15 for PM10 and 5 for PM2.5, in the same units.

For example, the reply gives Delhi an annual PM2.5 value of 102 in 2025-26. That is above both benchmarks. Mumbai's reported value is 33: below India's annual limit but above WHO's guideline. Meeting the Indian standard and meeting the WHO guideline are different results.

An annual average can also hide very polluted days. The count of cities that improved answers whether measured conditions changed against the programme baseline. The standards answer a separate question about the level that remains.

No matched pre-2014 all-India PM average is invented. Programme results and standards are kept separate; lower pollution is not the same as no pollution.

Sources: 20, 21.

More cover. Uneven changes.

The matched national comparison includes gains outside recorded forest areas.

FSI, the Forest Survey of India, recomputed the earlier figures for a matched comparison in the 2023 report. Its original 2013 headline was different. Using the revised baseline with the revised endpoint avoids treating a mapping revision as new tree growth.

Forest cover rose by 16,630.25 square kilometres, about 2.38%. Separately measured tree cover rose by 20,747.34, about 22.73%. The combined increase was 37,377.59, or about 4.73%. The combined row already contains the first two.

Forest cover means patches larger than one hectare with at least 10% tree canopy, regardless of ownership, legal status or land use. It includes some orchards and plantations. Tree cover separately measures other tree patches and dispersed trees. More mapped cover is not automatically more old natural forest.

Inside recorded forest areas, the matched forest-cover increase was only 440.47 square kilometres. Outside them it was 16,189.78. Most of the net mapped increase was therefore outside recorded areas. That location alone does not establish the kind or ecological quality of every new patch.

The North-East row is part of the national forest total, not extra cover. Across its eight states, cover fell by 3,132.27 square kilometres. A national increase therefore coexists with a substantial regional loss.

The latest assessment puts combined forest and tree cover at 25.17% of India's land area. This is the 2023 assessment, using imagery from earlier observation periods. It is not a measurement of all forests in October 2026.

Use Chapter 9 for the matched historical figures. Cover, legal forest area, natural-forest quality and biodiversity are distinct measures.

Sources: 22.

More emissions. Lower intensity.

The total and the amount per unit of economic output can move differently.

GHG means greenhouse gases, including carbon dioxide, methane and nitrous oxide. CO2-equivalent puts their different warming effects into a common unit. These gases are different from the local particle-pollution measure on page 18.

India's first Biennial Transparency Report, submitted in April 2026, puts emissions excluding land use and forestry about 69.7% above 2005 by 2022. The inventory's latest year is 2022. A 2026 submission is not a 2026 emissions observation.

Intensity means emissions per unit of economic output. It fell by 37.38%, so an earlier index of 100 became 62.62. If output doubles while emissions rise by half, the emissions per unit fall by 25%. A growing total and a falling intensity can therefore both be true.

The official climate-pledge baseline is 2005, not 2014. The matched inventory does not provide a 2013 entry. Forests and other land uses remove part of the emissions; including those removals, net national emissions still rose by 57.33% from 2005 to 2022.

The July 2026 reply reports non-fossil electricity capacity at 54.18% on 30 June. Capacity is equipment installed, not its share of electricity actually generated. The intensity-reduction target is 45% by 2030 and 47% by 2035; those targets are future commitments.

The same reply reports an additional forest-and-tree carbon sink of 2.44 billion tonnes of CO2-equivalent from 2005 to 2022. That is additional stored carbon, not an annual quantity that can simply be subtracted from one year's emissions.

The first row uses report-rounded values; the stated changes use the inventory calculations. Targets, installed capacity, carbon stocks and annual emissions are separate.

Sources: 23, 24.

One cooler year does not end warming.

Compare temperatures against the same long-term reference.

IMD, the India Meteorological Department, compares these years with India's average annual temperature over 1991-2020. The value 0.65 C is a gap above that reference. It is not an outdoor temperature of 65 degrees, or an increase measured only since 2014.

The 2009 observation gives earlier context using the same reference in the latest statement. A matched 2013 value was not verified here, so none is invented. One earlier year's temperature is also not a substitute for a long-term average.

2024 was warmer than 2025 in this measure. Yet 2025 was still the eighth-warmest year in India's record beginning in 1901. Annual temperatures fluctuate; one lower reading does not erase a longer trend.

The 2016-2025 decade was the warmest recorded decade, 0.32 C above the same reference. Across 1901-2025, the reported mean-temperature trend is 0.68 C per 100 years. That is the trend's rate, not a claim that warming began in 2014.

IMD reports 1,274 millimetres of rain in 2025, 110% of the 1971-2020 annual normal. A millimetre measures the depth of rain; 110% means about 10% more than that reference amount. Rainfall uses its own stated baseline, not the temperature baseline.

An annual national average does not tell us where rain fell or whether it arrived in damaging bursts. Heat, floods and drought are local experiences shaped by weather, buildings, land use and preparedness. A warming trend and better disaster protection are different questions, so neither should stand in for the other.

All temperature entries share the 1991-2020 reference. Rainfall uses the report's 1971-2020 normal. Observations stop at 2025.

Sources: 25.

Fewer listed polluted stretches.

The latest river list uses earlier monitoring and a specific pollution test.

A stretch is a section of a river. One river can have more than one polluted stretch. That is why 296 stretches and 271 affected rivers are different counts, and why neither means all 623 rivers were polluted.

The listed stretch count fell by 55, about 15.7%, against the 2018 report. The 2025 assessment uses water-quality monitoring from 2022-23. Its name is the report year, not a claim that every sample was taken in 2025.

CPCB, the Central Pollution Control Board, identifies these stretches mainly through BOD, biochemical oxygen demand. It measures the oxygen used as microbes break down organic matter. A value above 3 milligrams per litre is the assessment criterion used here.

High BOD can leave less oxygen for aquatic life. Low BOD alone does not prove that water is safe to drink: microbes, metals and other contaminants require their own tests. This list is therefore not a complete map of every kind of water pollution.

Monitoring coverage and stretch boundaries change. The 15.7% fall is a change in the published list, not proof that pollution in every Indian river fell by that percentage. A fair matched pre-2014 nationwide stretch comparison is not supplied here.

Namami Gange created 4,263 million litres per day of sewage-treatment capacity by June 2026. That is potential daily treatment, not the verified volume treated every day. Whether sewage reaches the plant, the plant runs properly and its output meets standards determines the actual benefit.

Latest polluted-stretch counts follow the main assessment tables and the March 2026 parliamentary confirmation. Treatment capacity is kept separate from river-water quality.

Sources: 26, 27, 28.

The average hides local pressure.

The amount taken out must be compared with the resource assessed as available.

Recharge is water replenishing the underground resource. The second row is what the assessment counts as available for use after its deductions. Withdrawal is water taken out. A billion cubic metres is one billion units of 1,000 litres each.

The old headline stage was about 62%; the 2025 stage is 60.63%. It does not mean 60.63% of all underground water has been used up. It compares estimated annual withdrawal with the assessed annual extractable resource.

The 2025 share is lower than the 2013 headline, but slightly above 2024's 60.47%. The most recent change was an increase of 0.16 percentage points. Keeping both comparisons avoids presenting every year as an improvement.

Over-exploited means extraction exceeds the annual resource under the assessment. The share of units so classified fell from about 15.70% to 10.80%. But boundaries, the number of units and assessment methods changed, so this is not an unchanged panel of identical local areas.

In 2025, over-exploited and critical units made up more than a quarter of units in Punjab, Haryana, Rajasthan, Delhi and Tamil Nadu. A manageable-looking national average cannot make a stressed local aquifer safe.

These are estimates, not readings from one giant national underground tank. Rain, recharge, irrigation demand and measurement changes affect them. Water level and water quality also need local evidence.

The 2013 assessment was published in 2017; the 2025 assessment was released in November 2025. Historical methods and availability definitions differ.

Sources: 29, 30.

Much more waste is reported processed.

Collection, processing and recycling describe different steps.

The June 2026 government backgrounder reports processing rising from 16% to 82%. That is a gain of 66 percentage points. In the programme's reported waste measure, roughly 16 out of 100 became 82 out of 100.

Collection coverage rose from 43% to 98%, a gain of 55 points. This is the reported urban service indicator. The backgrounder does not spell out its full counting base, so I do not turn it into a claim about 98% of all Indian households.

Collection is getting waste from the point where it is discarded. Processing is handling it through a treatment route. Recycling is recovering material for use again. Waste counted as processed can go through composting, biological treatment, fuel production or other routes; 82% processed does not mean 82% recycled.

The January 2026 ministry reply gave 81.30% processing in its dated administrative record. The June backgrounder reports 82%. Different dates and rounding matter. Neither is presented here as a new October survey.

These are official administrative results, based on programme reporting. They are useful for tracking the reported expansion of services. They do not establish that every collection round happened, that every plant met its standards or that every city performed equally.

New daily waste is a flow. Old waste accumulated in dumps is a stock. A high share of fresh waste processed does not mean every old landfill has been cleared. Cleaner streets, properly treated waste and restored dumpsites require separate evidence.

Urban programme coverage is not all-India rural-and-urban household coverage. Processing and recycling should not be used as interchangeable words.

Sources: 31, 32.

Recovery and recognition grew.

Animal estimates and protected-site designations are different kinds of progress.

The estimated tiger population rose by about 65.4% from 2014 to 2022. The earlier census belongs to the transition year, not a fully pre-2014 year. The latest estimation verified in the July 2026 update still describes 2022.

The final 2022 estimate is 3,682. The earlier minimum estimate of 3,167 covered camera-trapped areas; further analysis included other tiger-presence areas. Those are two stages of the same estimation, not two different annual populations.

This is a scientific estimate using field work, camera traps and modelling. It is not an exact headcount of every tiger alive on the day of publication. The final report also gives an upper estimate of 3,925.

The reserve network expanded from 46 to 58. More reserves mean more sites designated for tiger conservation; the count alone does not describe their size, condition or links between habitats. Project Tiger began in 1973, so the conservation story long predates 2014.

Ramsar sites are wetlands recognised as internationally important under the Convention on Wetlands. The count reached 101 with Glaw Lake in Arunachal Pradesh in August 2026. It was 26 in 2014. The increase is in recognised sites, not proof that India gained 75 entirely new wetlands.

These are substantial gains in the measured indicators. They do not establish that every species or ecosystem recovered. The tiger report describes local declines too. Water quality, habitat connections and pressure on people living around protected areas remain part of conservation.

All three measures keep their own dates. The tiger estimate is 2022; reserve and Ramsar counts are dated 2026 updates. Part 6 follows with rankings and wider chart checks.

Sources: 33, 34, 35.

Sources and reading notes