Research · Corruption Accountability
Where Nepal's corruption accountability actually leaks
How many complaints become cases, how those cases resolve, and which charges the system convicts — drawn from court records ingested from Nepal's judiciary and the CIAA's own annual reports. Every figure links to the record behind it.
Based on a 14-fiscal-year analysis of court records (FY2069/70–2082/83) and the CIAA's annual reports through FY2081/82, the most recent published. How we built and cross-checked these numbers →
यो प्रतिवेदन हाल अङ्ग्रेजीमा मात्र उपलब्ध छ।
The funnel
Between 0.2% and 0.3% of complaints end in a conviction
In FY2081/82 the CIAA registered 28,554 new complaints. Most were screened out at intake — only 947 (3.3%) went to a full investigation — and of those it prosecuted 137, roughly one in seven. Apply this archive's measured conviction rates and somewhere between 62 and 84 of them end in a conviction: the low end counts only cases where the charge was upheld outright, the high end also counts the partial verdicts. Both are defensible readings, so we publish the range rather than pick one.
- Complaints to the CIAA28,554100% of complaints
- only 3.3% go to a full investigationComplaints fully investigated9473.3% of complaints
- ≈1 in 7 investigated are prosecutedProsecutions filed1370.5% of complaints
- ≈45% of prosecuted cases convict fully, ≈61% including partialConvictions (est., full → incl. partial)62–840.2–0.3% of complaints
Why this starts at 28,554 and not 37,026
You will more often see 37,026 complaints quoted for FY2081/82, and that number is also the CIAA's own. It is the Commission's total workload: 28,554 complaints newly registered during the year plus 8,472 still open from previous years. Both appear in the same table of the 35th report. A funnel that starts at 37,026 would count those 8,472 a second time, having already counted them in an earlier year, so this one starts at new intake — which also makes the percentage at each stage smaller and more conservative than the headline version.
Complaint, investigation and prosecution counts: CIAA 35th annual report (FY 2081/82), cross-checked against the court records. The steep drop is at intake screening, not the courtroom — most complaints never warrant a full investigation (many are outside the CIAA's jurisdiction or evidence-free); of those it does investigate, it prosecutes about 1 in 7. The conviction stage is a projection and a range, not a count: it applies this archive's measured 45% full-conviction rate across all 14 years to the filed count for the floor, and its 61% including-partial rate for the ceiling. It is what cases like these have historically done — not a tally of how those particular 137 were decided. CIAA 35th annual report
Outcomes
Fewer than half of decided prosecutions end in a clean conviction
Of 2,728 decided cases with a clear verdict, 45% end in a full conviction, 39% in outright acquittal, and 16% in partial conviction — so 55% end in something less than a clean conviction.
The CIAA puts its own “success” rate at 52.67% because it counts partial convictions as successes. That figure is comparable to the top of our range, not the bottom: on the CIAA's own definition this archive gives 61%, higher than the Commission's number rather than lower. Set 52.67% against our 45% and you are comparing two different definitions of the word. Browse the court records
What these numbers count, and what “partial” covers
These are cases, not people. The court records one verdict per case and publishes no outcome for each accused separately, so nothing here is a per-person conviction rate and none of it should be read as the share of individuals convicted. A single case may have one accused or a dozen.
That is also why the middle category does two jobs at once. A partial verdict (आंशिक ठहर) covers a case where one accused was convicted on some charges and cleared of others, AND a case with several accused where some were convicted and others acquitted. The court's record does not distinguish the two, so neither can we. A partial verdict tells you the prosecution did not fail outright; it does not tell you how many people were convicted, or of what.
“Full conviction” (ठहर) is the strict reading: the charge was upheld against the case as the court framed it. It is the one category that cannot be read two ways, which is why it anchors the low end of every range on this page — 45% here, 61% once partial verdicts are counted as convictions too. Neither endpoint is the whole truth. Reporting only 45% treats every mixed verdict as a failure; reporting only 61% treats a case where one junior official was convicted and everyone senior walked as a win. The court's record cannot tell those apart, so we give you both numbers and never a single one on its own.
The docket
What the court is actually asked to try
Cases filed vs. decided, by year
Fiscal year. The filing peak (FY2076/77) precedes the decided peak (FY2080/81) by ~4 years.
How the charge mix shifted, by year
Register cases by fiscal filing year — 2,852 of the 2,949, money laundering excluded. “Other” folds together seven smaller charge families and the matters whose charge text could not be classified. Fake-credential cases (crimson) dominated the early docket — about 70% in FY2069/70 — then fell to single digits by FY2077/78–2079/80, with a rebound in FY2080/81. The illegal-benefit charge was barely used before FY2078/79 (a single earlier case, in FY2069/70) and loss to government grew.
- Bribery
- Fake credential
- Embezzlement
- Illegal benefit
- Loss to government
- Other
What actually sticks
Conviction depends overwhelmingly on what was charged
Two charges carry the court. Fake-credential cases convict at 90% and supply 46% of every full conviction. Bribery — the largest docket on the court, 903 decided cases — convicts at 47%, the second-highest rate of any charge and a shade above the court average, and supplies another 34%. Between them that is four in five convictions. It is the rest of the docket that mostly fails: embezzlement, illicit enrichment and loss to government all convert in the low twenties or below, and illegal benefit sits at the bottom on 4.5%.
Decided cases per charge type, split by outcome — 2,702 of the 2,728 cases with a clear verdict; the rest carry charge text we could not classify. Money laundering keeps its own row here even though it sits outside the substantive corpus, because it is prosecuted under a separate statute. Every rate is per case, never per accused. Cited to the underlying charge sheets and court records. Charge sheets
Over time
The court convicts far less than it used to
In its first three years the Special Court fully convicted roughly 60% of the corruption cases it decided; across FY2079/80–2082/83 that fell to about 37%. Acquittals have drawn level rather than taken over: across those recent years they slightly outnumber full convictions, 386 to 345, but in only 3 of the 14 years did they finish a year ahead.
Outcome mix by verdict year
Share of decided cases by verdict fiscal year. Acquittals spike above full convictions in FY2078/79 and FY2080/81 — and edge ahead again in FY2082/83 — but they do not stay there.
Is the decline real? Easy wins vs. core graft
Documentary fake-credential cases — which convict at ~90% — fell from 59% of the decided docket to as little as 5%. Core financial graft converts at much the same level now as it did then: 26% across the first three years against 30% across the last four. So the headline decline is mostly that change of mix — the easy wins leaving — not the court convicting serious graft any less. Read the dashed line as a level, not a trend: it swings from 64% in FY2076/77 to 7% in FY2078/79 because in some years only a few dozen core-graft cases were decided.
Full-conviction rate by verdict fiscal year, case-grain, from Special Court records. Browse the court records
Pace
How long a prosecution takes, and when they arrive
Time to verdict, and the backlog
By filing cohort. Cohorts through FY2079/80 took a median 30 months at their slowest (FY2076/77); 206 cases filed since are still awaiting a verdict.
When cases are filed, by Nepali month
Mean cases filed per Nepali month across FY2069/70–2082/83; error bars show ±1 standard deviation. Filings peak in Ashadh — the fiscal year-end — and trough in Kartik, the Dashain/Tihar festival month.
Bars are the mean across the 14 fiscal years FY2069/70–2082/83 — all complete for filings, which is what this chart measures; the most recent cases are filed but many are still awaiting a verdict. Whiskers are ±1 standard deviation, i.e. how much each month swings from year to year. Registration date from Special Court records.
Which bench you draw
Full-conviction rates run from 78% to 21% across the court's benches
The court records one verdict per case and no individual vote, so this is a property of the panels a justice sat on rather than of the justice: every panel member is credited with the panel's outcome, and a judge who dissented is credited with it too. Dot size is scaled to caseload. Sitting on the same court, hearing the same prosecutor, benches diverge more than threefold — but descriptively, not causally: some of that spread is the charge mix and the era a bench sat in, which the two sections above set out.
Justices who sat on at least 30 decided cases — 39 of the 41 in the record; below that a rate turns on a handful of verdicts and means little. Because every member of a panel is credited with the panel's outcome, the 39 shown are credited with 7,538 decisions between them, well above the 2,728 cases those decisions came from.
Where the gap is
Attrition concentrates at the CIAA stage — then we lose the trail
Follow one complaint from the day it is filed to the day someone is sanctioned for it. Five stages, two institutions — and our record of it runs out before the end.
Complaint intake & screening
CIAAIn FY2081/82, 28,554 new complaints arrived and 137 prosecutions came out the far end — near 0.5%. Most are screened out here: shelved, referred elsewhere, or closed for want of evidence or jurisdiction, and much of that is legitimate. But it happens with no public verdict on any individual complaint, which makes this both the largest drop in the pipeline and the least visible.
Investigation & charging
CIAAOf the complaints it fully investigates the CIAA prosecutes about 1 in 7, and what it charges shapes the outcome more than anything the court does. Two charges do almost all the work: fake credentials convict at 90% and bribery at 47%, and between them they account for four in five convictions. Take those two out and the rest of the docket converts at 20%.
Trial & verdict
Special Court39% of decided cases end in outright acquittal and 55% in something short of a clean conviction, with more than a threefold spread across benches. Year to year the full-conviction rate has swung from about 86% to about 14%.
Appeal
Supreme CourtNo data yetWe can see how many appeals are filed: in FY2081/82 alone the CIAA appealed 251 Special Court verdicts to the Supreme Court, and defendants appeal their convictions too. We cannot see how they end. They clearly do end — the CIAA filed 5 review petitions that same year against Supreme Court rulings on its own appeals — but our court records hold no decision for them and we have not found the outcomes published anywhere as data. So we cannot tell you how often a Special Court verdict survives on appeal. That is a gap in what we have, not a claim that the answer is unknowable.
Recovery & sanction
StateNo data yetBillions of rupees in damages are demanded each year — Rs 6.02 billion in FY2081/82 alone. How much of it is ever collected, and whether those convicted serve a meaningful sanction, we do not know: we have not found a source that tracks either, and we hold no data of our own on it. It may be recorded somewhere we have not looked. What we can say is that we cannot follow the money past the verdict.
The dashed stages mark the limits of our own record, not a verdict on what happens inside them. We can see what goes in — appeals filed, damages demanded, both published by the CIAA. We have not been able to find what comes out. Whether that is because nobody publishes it or because we have not looked in the right place, we cannot tell from here, so we are not going to claim the stronger version. If you know of a dataset, report or registry covering appeal outcomes or amounts recovered, please tell us and we will fold it in. inquiry@jawafdehi.org · Appeal and damages figures: CIAA 35th annual report
Methodology
How this report was built and cross-checked
Two independent public records, cross-checked against each other — and both browsable on Jawafdehi. Complaint, investigation and prosecution counts come from the CIAA's own annual reports; conviction outcomes come from our mirror of Nepal's Special Court and wider judiciary. Where the two overlap — the cases filed each year — they agree to 1.2%, and the section above sets out case by case where they do not. Every figure on this page links to the record behind it.
Do the two records agree?
The CIAA's annual reports and the Special Court's register are independent accounts of the same events — the Commission deciding to prosecute, and the court opening a docket. Across 13 fiscal years they agree to 1.2%: 2,592 filings the CIAA published against 2,624 comparable cases in the register, and no single year differs by more than 7. Where they diverge, one offence dominates. In the 8 years whose reports break filings down by offence, the register runs 28 cases ahead, and 22 of those 28 are fake-credential cases alone.
For the 3 years where the gap is widest we checked it case by case, matching the accused named in the reports' own filing tables against the accused named in the register. The direction of the divergence turns out to be the opposite of what a missing-records story would predict: every one of the 254 cases the CIAA says it filed is in the register. What the register holds instead is 19 fake-credential cases that those years' own filing tables never list.
Why 19 register cases are missing from the CIAA's own tables
Two of these reasons are documented rather than inferred. Five cases are a difference of principle about which year a filing belongs to: the CIAA attributes it to the year the Commission decided to prosecute, the court to the year the docket was registered — and all five appear in the previous year's report, with decision dates in Ashadh and filing dates that match the register's registration date to the day. Four more are absent from one year's report but described in the next report as prosecutions the CIAA filed at the Special Court and won: one CIAA document contradicting another.
Filing counts from the CIAA's annual reports, cross-checked against the Special Court register. The register side removes 154 cases in three streams the CIAA does not file — 98 money laundering (mixed bribery-and-laundering dockets included there rather than counted separately), 33 petitions filed against the CIAA itself, and 23 offences outside its jurisdiction — an exclusion that is marginally over-broad, since the FY2081/82 report shows the CIAA filed two money-laundering cases itself. Three limits travel with this: only 3 of the 14 years and one of the 13 offence families have been checked at case level; 9 of the 19 surplus cases remain unexplained; and the reading that the annual report under-counts its own fake-credential filings is established for four specific cases and well supported for the rest, not proven for all. CIAA annual reports
Full methodology, discrepancies & limits
Corpus. Of ~12,600 Special Court records, most are procedural petitions. We isolate CIAA prosecutions as the Special Court's -CR- criminal register — 2,949 cases filed FY2069/70–2082/83 (the register is the definition; no plaintiff filter) — of which 2,795 are substantive corruption charges after removing money-laundering (a separate statute, 93 cases) and unclassified matters (61).
Grain. One verdict per case, because that is all the court publishes — there is no per-accused outcome in the record, so no figure on this page is a per-person rate. “Partial” therefore covers both a single accused convicted on some charges only and a multi-accused case split between conviction and acquittal; the two are indistinguishable in the source.
The funnel. Of the 28,554 complaints newly registered in FY2081/82, only 947 (3.3%) went to a full investigation; most of the rest were screened out at intake — shelved or referred — much of it legitimate (outside the CIAA's jurisdiction, no supporting evidence, or duplicates). Of the complaints it fully investigated, the CIAA filed charges in 137 — about 1 in 7. The 37,026 “दर्ता” headline for the same year is the Commission's total workload, which adds 8,472 complaints carried over unresolved from earlier years; starting the funnel there would count those twice. So “0.5% of all complaints reach court” and “~1 in 7 of the complaints it investigates is prosecuted” are both true and measure different stages.
Outcomes. Verdicts are coded per hearing as convicted / acquitted / partial; each case is taken at its terminal deciding hearing. The conviction rate is over the 2,728 register cases carrying an unambiguous ठहर / आंशिक / सफाई disposition. That is a different set from the 2,740 whose case status reads फैसला (which the filed-vs-decided trend counts), and neither contains the other: 2,628 cases are in both, 112 are marked decided but carry no hearing with a recorded disposition, and 100 carry a disposition without the corresponding status.
Where a verdict came from. Cases that reached the mirror without ever appearing on a published cause list carry no court-published disposition, so the only way to count them at all is to read the verdict out of the judgment text. 69 verdicts were recovered that way (37 conviction, 26 acquittal, 6 partial) and every one of them is EXCLUDED from every rate on this page — a rate that quietly mixed court-published and machine-read verdicts would misrepresent its own source. We report the number rather than filtering silently.
Dates. Verdict dates are parsed from the case status text; filings from the registration date. Bikram Sambat dates throughout; by-year charts bin by fiscal year (Shrawan–Ashadh).
Over time. Yearly rates are grouped by verdict fiscal year; the sharp rise in acquittals from FY2078/79 is a genuine surge in the record, not a coding artifact. Time-to-verdict is measured by filing cohort: cohorts through FY2079/80 are essentially fully decided, but recent cohorts are still open, so their apparent speed reflects only the cases already resolved (survivorship) and is drawn as provisional.
Per-justice. Attribution is bench-grain: every member of a panel is credited with the panel's outcome, so this describes the benches a justice sat on, not that justice's individual effect. The chart shows only justices with at least 30 decided cases — 39 of the 41 in the record — so it is a filtered view. It is descriptive, and small differences are noise.
Discrepancy with CIAA figures. The CIAA's 52.67% “success” rate for FY2081/82 counts full and partial convictions together — its 35th report states it as 87 full plus 120 partial of 393 verdicts. That is the same definition as the top of our range (61%), not the bottom (45%, full convictions only) — and on that aligned footing this archive comes out above the CIAA's figure, not below it. So the two differ by period at least as much as by definition: the CIAA's is one volatile year (its published rates range from 33% to 88%), ours is cumulative across 14. Never compare them without aligning both.
The cross-check. Matching the reports' per-case filing tables to the register is name matching, not a key lookup: the early high-divergence years print no case number at all, and a case number alone does not identify a court in any event — the same NNN-CR-NNNN format is used by the Special Court, the Supreme Court and the district courts, so a number lifted out of its column resolves to the wrong case. Names were matched after folding Devanagari spelling variants, and the residue was resolved by hand on the date: every pair we accepted matches the printed filing date to the register's registration date to the day. Three pairs were accepted this way and are flagged as such in the published data, so a reader who rejects them can re-derive the totals without them.
Identity. Only ~7% of distinct defendants (607 of 8,321) are resolved to a canonical, cross-referenced identity, so office-level and repeat-offender cuts are deferred as low-confidence.
Reading the source PDFs. The CIAA reports are Nepali-language PDFs set in legacy Devanagari fonts that ordinary tools garble. We convert them to clean, checkable Markdown with likhit — Jawafdehi's open-source universal Nepali document-to-markdown converter — then verify every figure by eye against the original page.
Limits. These are the records in the archive as of the snapshot date (2083 BS); figures update as new records are mirrored. Appellate outcomes are largely missing from what we hold, and we have not found any source for amounts actually recovered — so treat both as open questions on our side rather than settled absences. Corrections and pointers to sources we have missed are welcome.