Investigations & impact
Journalism that mattered
These are not client case studies. They are the stories where reporting met a record, a record met a reader, and something afterwards was different — a report republished, a route sealed, a bill itemised, a junction rebuilt. Each entry sets out what we found, how we verified it and what followed.
Headline impact metrics
41
Investigations published
18,740
Documents reviewed
23
Policy or process changes
0
Sources exposed
Counts cover January 2023 to June 2026 and exclude routine daily reporting. “Policy or process change” means a published decision by an institution that cites, follows or directly responds to our reporting. We do not claim credit for changes we cannot document.
Flagship investigation
Eight months, 214 wards, one spreadsheet that would not reconcile.
Above: Ward-level residual chlorine readings redrawn from the audit annexure. The dashed line is the ceiling each corporation said it never crossed; four of the eight sampled wards sit above it. NewsDunia illustration.
Civic Infrastructure — Investigation
The ward water files: eight months inside a municipal testing regime
Three municipal corporations published monthly water-quality summaries that their own laboratory sheets did not support. After we put the discrepancies to them, 214 wards republished their reports with raw readings attached.
What we found
The story began with a routine request. A reader in an eastern ward wrote to ask why her building’s supply tasted of chlorine on some mornings and of nothing at all on others, and whether the corporation’s cheerful monthly summary — “all samples within specification” — could really be true for every one of the 4,600 samples it claimed to have drawn.
Over eight months we obtained the underlying sampling registers for three corporations under transparency law: the collection slips, the laboratory bench sheets and the ward-level roll-ups that feed the published summary. In 44 of the 214 wards covered, the roll-up could not be reconstructed from the bench sheets beneath it. Samples logged as drawn on public holidays had no collection slip. Twelve wards reported identical residual-chlorine values for six consecutive months, a pattern the two independent laboratories we consulted described as implausible in a live distribution network.
Crucially, the failure was not one of contamination. Nothing we found suggested the water was unsafe. The failure was of record-keeping and disclosure: readers were being handed a summary that could not be checked, by systems that had quietly stopped checking themselves.
How we verified it
We treated every claim as false until a document made it true. Each of the 214 ward roll-ups was re-entered by hand into a spreadsheet by two reporters working independently; only rows where both entries matched were used. Where they disagreed, we went back to the scan.
We then commissioned parallel sampling. Over five weeks a technician from an accredited laboratory, working to a protocol published in advance on our website, drew 96 samples from taps in wards we had flagged and 40 from wards we had not. The control wards matched their published figures closely. The flagged wards did not.
Before publication we wrote to all three corporations with a complete schedule of the discrepancies, ward by ward, and gave them fourteen days to respond. Two replied in detail and both responses are reproduced in full alongside the series. A spokesperson for the Metropolitan Water Systems Board told us the roll-up template had been “carried forward from a legacy reporting format” and that the board would review it — a statement we published without adornment.
We did not publish a single figure we could not point to on a page. Where the record was silent, we said so in the copy instead of filling the gap with an inference.
NewsDunia standards note, filed with the seriesWhat changed afterwards
Within six weeks of publication all three corporations moved to monthly disclosure with raw bench sheets attached, and 214 ward reports were republished in the new format. The state’s urban development department issued a circular requiring collection slips to be scanned and retained for three years. An independent oversight panel — the State Water Quality Assurance Cell — was constituted in October with a published terms of reference that cites the series in its preamble.
Two further outcomes matter more to us than any of these. The parallel-sampling protocol we wrote has since been adopted by three university civic-data programmes as a teaching template. And the reader whose letter began the whole thing now receives, like everyone else in her ward, a report she can actually read.
Six more that moved something
Public health, education, environment, consumer rights, civic infrastructure and rural connectivity.
Public Health
The line item nobody could explain
Four thousand one hundred discharge summaries from empanelled hospitals carried an unlabelled “facility charge” that no scheme document defined. Ninety days later, the schedule of chargeable heads was published for the first time.
Education
Sealed and signed
We reconstructed the custody trail of question papers across three examination districts from signature registers alone. Two of eleven handovers had no counter-signature at all. Sealed-route tracking is now recommended state-wide.
Environment
Downwind of the kiln belt
Two winters of independent particulate sampling at fourteen village sites showed the nearest official monitor sat upwind of every kiln in the cluster. Three new monitors were sited downwind before the following season.
Consumer Rights
The refund that took eleven months
Reader complaints let us map a redress backlog nobody was publishing: median 47 days at one district forum, 311 at another. Quarterly disposal statistics are now posted by every forum in the state.
Civic Infrastructure
Nineteen hundred black spots
The crash register recorded location as a police station name, not a coordinate. Re-geocoding nine years of entries revealed 1,900 recurring sites; 62 were taken into a remedial works programme.
Rural Connectivity
Signal on paper
Four thousand two hundred village Wi-Fi points were reported as commissioned. Volunteer reporters ran a standard reachability test at 340 of them over one month; a third answered for less than half the working day.
Three, in full
The reporting, the verification and the sidebar that tells you what it cost.
Education — Examination logistics
Sealed and signed: reconstructing the custody trail of a board question paper
A question paper travels from a printing press to a strongroom to a district treasury to a school, and every leg is supposed to be counter-signed. We asked a simple question: could an outsider, using only the signature registers, say where a given packet had been at any hour? For two of eleven handovers, the answer was no.
New Delhi — The examination board publishes an elaborate custody protocol and it is, on paper, a good one. Each packet of sealed question papers carries a serial; each transfer requires two signatures, a time and a seal number. The protocol has eleven defined handover points between press and desk.
We obtained photocopies of the custody registers for 38 packets across three examination districts for a single sitting, and asked two reporters to reconstruct each journey independently. Nine of the eleven handovers reconciled cleanly. Two did not: the transfer from district treasury to school, and the return of unused packets, were frequently recorded with a single signature, no seal number, and in fourteen instances a time entered as a date only.
We want to be precise about what this does and does not show. We found no evidence that any paper leaked, and we said so in the second paragraph of the published story. What we found was that the record was not capable of proving the negative — and a custody protocol whose records cannot be audited is a protocol that offers reassurance rather than security.
How we verified it
Every register page was transcribed twice, by different reporters, into a structured sheet with one row per handover. Disagreements were resolved against the scan by a third person. We shared our full schedule of gaps with the board and with two district education offices, and gave twenty-one days for reply. The board’s controller of examinations responded with a detailed note, published in full, agreeing that the treasury-to-school leg was “the weakest documented segment” and describing a pilot of barcode scanning already under consideration.
We also consulted two retired examination administrators, neither of whom had worked in the districts concerned, to check that we had understood the protocol correctly before we accused anyone of departing from it.
What changed afterwards
The board’s logistics review, published four months later, recommended sealed-route tracking with scanned handovers at every leg, retention of custody records for five years, and a standing quarterly audit of a random ten per cent of packets. The recommendations were adopted for the following academic year in the three districts we examined and extended to eleven more in April.
Public health — Hospital billing
The line item nobody could explain: inside 4,100 discharge bills
Readers sent us their bills. We asked every institution in the chain to tell us what a particular charge was for. Nobody could point to a document that defined it — because, it turned out, no such document had ever been published.
New Delhi — The charge appeared under a dozen names: facility charge, service component, establishment head. It ranged from a few hundred rupees to a fifth of the bill. It appeared on discharge summaries from empanelled hospitals across four states, and patients were routinely told it was “as per scheme”.
We built a corpus of 4,100 anonymised bills submitted by readers over eleven weeks, stripped of every patient identifier before a reporter ever saw them, and coded each line item against the published schedule of chargeable heads. Roughly one rupee in nine sat under a head the schedule did not contain.
How we verified it
Anonymisation came first: an intake script removed names, registration numbers and dates of birth at upload, and the raw files were never stored on a reporter’s machine. We then had a health-finance researcher at an independent institute re-code a random 400-bill sample without seeing our coding; agreement was 94 per cent, and every disagreement was reviewed line by line.
We approached the scheme administrator, two hospital associations and the State Health Systems Resource Centre. A spokesperson for the administrator confirmed that no consolidated schedule of chargeable heads had been published since the scheme’s second revision, and that hospitals had been working from circulars issued piecemeal.
What changed afterwards
Ninety days after publication, a consolidated schedule of chargeable heads was published for the first time, with a mandatory plain-language line description on every discharge summary. A grievance route specific to billing disputes was opened, and the administrator now publishes quarterly counts of disputes raised and resolved. We continue to track that series; when the numbers stop being published, we will report that too.
Civic infrastructure — Road safety
Nineteen hundred black spots: what the highway crash register was not saying
The data existed. It had simply never been joined up, because location was recorded as the name of the reporting police station rather than a point on a map. Re-geocoding nine years of entries changed what the register could tell anyone.
Chennai — A black spot is a place where crashes recur. Identifying one requires knowing where crashes happened, to a resolution finer than a police jurisdiction that may span forty kilometres of highway. For nine years the register had recorded jurisdiction, chainage occasionally, and a free-text landmark that ranged from a milestone number to “near the water tank”.
We took 214,000 anonymised entries covering nine years and two states, parsed the free-text landmark field, and matched it against milestone gazetteers, toll-plaza schedules and open road-network data. Sixty-one per cent of entries resolved to a point within 500 metres. Those points clustered: 1,900 locations accounted for a share of recurring incidents far above what a uniform spread would produce.
How we verified it
Geocoding is guesswork unless it is checked, so we checked it. A road-safety engineering group at a state technical university, working with our method note but not our results, re-geocoded a random 2,000-entry sample; their points fell within 500 metres of ours in 88 per cent of cases. We published the disagreement rate as prominently as the agreement rate.
We then visited 24 of the highest-ranked sites with a road-safety auditor and photographed the geometry: missing chevrons at a curve, a median opening opposite a school gate, a service road merging without a taper. Where our data said a site was dangerous and the site itself did not obviously agree, we said so.
What changed afterwards
The state highways department added a mandatory latitude-longitude field to the crash entry form within the year, backfilled it for the previous eighteen months, and took 62 of our sites into a remedial works programme. An inter-agency Road Safety Data Cell now publishes an annual black-spot list; the first edition overlapped with ours at 71 per cent, and where it did not, it explained why.
The record, in one table
Reach is measured as unique readers in the ninety days after first publication.
| Story | Beat | Reach | Duration | Documented outcome |
|---|---|---|---|---|
| The ward water files | Civic infrastructure | 1.42 m | 8 months | 214 ward reports republished with raw readings; oversight cell constituted |
| Sealed and signed | Education | 910 k | 5 months | Sealed-route tracking adopted across 14 examination districts |
| The line item nobody could explain | Public health | 1.06 m | 6 months | Consolidated schedule of chargeable heads published; billing grievance route opened |
| Nineteen hundred black spots | Road safety | 780 k | 9 months | Coordinates made mandatory in the crash register; 62 sites into remedial works |
| Signal on paper | Rural connectivity | 640 k | 4 months | Uptime reporting added to the commissioning certificate for all new sites |
| The refund that took eleven months | Consumer rights | 520 k | 3 months | Quarterly disposal statistics now published by every district forum in the state |
| Downwind of the kiln belt | Environment | 465 k | 14 months | Three ambient monitors resited downwind of the cluster before the next season |
Reach figures are rounded and drawn from our own analytics; they count unique readers, not page views. Outcomes are listed only where an institution has published a decision that we can link to.
What readers and partners said
Reproduced with permission. Institutional partners have no editorial involvement in our reporting.
I wrote in about the taste of the water in my building. Nobody had ever asked me for the evidence before, and nobody had ever sent me the answer afterwards. This newsroom did both.
They sent us their full schedule of discrepancies fourteen days before publishing, ward by ward. That is not a courtesy. It is what allowed us to fix the template rather than argue about the story.
We now teach the parallel-sampling protocol to second-year students. It is the clearest worked example we have of a newsroom publishing its method before it publishes its findings.
Three and a half years of investigative work
Every figure below is one we can evidence from our own records.
41
Investigations published
286
Right-of-reply letters sent
31
Corrections issued in the open
7.4 m
Readers reached
A correction rate we publish rather than hide: thirty-one corrections across forty-one investigations, each carrying the date, the error and the person who caught it.
Recognition and membership
Independent bodies whose standards we submit to, and juries that have cited this work.
Membership of a standards body does not confer editorial approval of any individual story. Complaints about our reporting are handled first by our own Grievance Officer, whose details appear on our Disclaimer page.
The next one usually starts with a letter
Every investigation on this page began with a document somebody decided to send us, or a reader who noticed that two numbers did not agree. We confirm receipt within one working day, we protect sources as a matter of policy rather than favour, and we will tell you honestly if we cannot take it on.