Every number, defined.

What Form 26 contains

Every candidate in an Indian election must file a sworn affidavit — Form 26 — with their nomination papers, under the Conduct of Elections Rules and Election Commission directions. It requires disclosure of: pending criminal cases (with FIR/case numbers, police station, court, the Acts and sections, and whether a court has taken cognizance or framed charges), past convictions with sentence, plus assets, liabilities and education. Candidates sign it under oath; filing a false affidavit is itself an offence. The ECI publishes these affidavits on its affidavit portal.

Why a declared case is not a conviction

A criminal case moves through stages, and most entries in affidavits are at an early one. On this site, every case is shown at the stage the candidate declared, and nothing more:

• FIR registered — the police recorded an allegation. Nobody has weighed evidence yet.
• Investigation — the police are collecting evidence; it can end in a chargesheet or a closure report.
• Cognizance taken — a magistrate has formally taken notice of the case for judicial process.
• Charges framed — a court found enough material to put specific charges to trial. Still not a finding of guilt.
• Trial — evidence is heard. Ends in conviction, acquittal, or discharge.
• Conviction — a court found guilt. Can be appealed and stayed or reversed.
• Acquittal — the court found the person not guilty.
• Discharge — the court found insufficient grounds even to frame charges.
• Quashed — a higher court annulled the proceedings.
• Closure report — police concluded there was no case to pursue (court may accept or reject).
• Appeal — a higher court is re-examining a verdict; a conviction under appeal is shown exactly as declared.

Presumption of innocence applies throughout. This site never calls a person a criminal. Declared pending cases and disclosed convictions are presented separately and labelled as declared.

The data actually used in this release

• Winners and constituencies: the official “State/UT-wise List of Successful Candidate during 2024” dataset on data.gov.in (Government Open Data License – India). It contains 542 rows — Surat’s uncontested seat is excluded at source and is recorded here from the analysis report below, with a note.
• Party at election and all case data: the Association for Democratic Reforms / National Election Watch report “Lok Sabha Elections 2024: Analysis of Criminal Background…of Winning Candidates” (6 June 2024), which digests all 543 winners’ sworn affidavits obtained from the ECI portal. We downloaded this published report once and extracted its tables; we did notscrape MyNeta or ADR pages (their terms prohibit systematic collection). Party attribution was cross-checked: our extracted seat totals equal the report’s own printed party table exactly (240 BJP, 99 INC, 37 SP, 29 AITC, 22 DMK, 16 TDP, 7 SHS among the anchors).
• Why not the official winner-party column? The machine-readable copy of the official results file collapsed duplicate winner/runner-up columns during the portal’s ingestion, corrupting the winner-party field (documented in the coverage page fallbacks). Names, constituencies and states are used from the official file; party comes from the report above, reconciled as described.
• State enforcement: NCRB “Crime in India 2023” tables 8C.2–8C.4 via data.gov.in (GODL) — data year 2023, published later (both years shown). Sums of states were verified against the printed All-India rows for every column.

Every record carries a review state. In this release everything is machine-checked(validated against the source’s own printed totals) and nothing is human-verified yet — profiles with adverse records are excluded from search-engine indexing until a human review pass is done.

Exact metric definitions

• reps_covered — Elected representatives in this cohort with affidavit-derived case data (parsed case rows or a cited published summary), out of expected seats.
• reps_with_declared_cases_pct — Covered representatives who declared one or more criminal cases in their election affidavit ÷ covered representatives. Counts people, not cases. A declared case is an accusation, not a conviction.
• reps_with_serious_cases_pct — Covered representatives with one or more 'serious' declared cases ÷ covered representatives with serious-case data, using the source report's own definition of serious (quoted on the methodology page).
• reps_with_convictions — Covered representatives whose affidavit-derived records disclose at least one conviction. Null when no conviction-level data has been imported for this cohort.
• total_declared_cases — Sum of declared-case counts across covered representatives, as counted by the underlying source. One person may account for several cases; do not read this as unique court cases.
• affidavits_linked — Representatives whose profile links affidavit-derived data (parsed rows or published summary), out of all representatives identified in the cohort.
• coverage_pct — Covered representatives ÷ expected seats for this cohort. Anything below 100% is listed on the coverage page with reasons.

Additional derived figures shown in the corruption dashboard use NCRB’s own published rate columns, whose formulas appear in the column headings (e.g. Conviction Rate = cases convicted ÷ trials completed × 100).

“Serious” cases

The serious-case flag uses the source report’s printed criteria, quoted: offences with maximum punishment of five years or more; non-bailable offences; electoral offences (e.g. bribery); offences causing loss to the exchequer; offences of assault, murder, kidnapping, or crimes against women. It is acharge-level flag (a representative meets it if any declared charge does), not a case count.

Party denominator rules

• Party is party at election; later defections and mergers are not reflected until verified separately.
• Denominators are covered representatives (those with affidavit-derived data), shown on every bar.
• Parties with fewer than 5 covered representatives are displayed but never ranked; their raw numerator/denominator stays visible.
• One person with several cases counts once in any “people” metric. Rows are seat-winners: a person who won two seats appears once per seat, matching the source report’s unit.

State-comparison limitations

NCRB counts are reported enforcement activity. More registered cases can reflect more underlying corruption, more reporting, stronger agencies, or all three — so raw counts are never presented as a corruption ranking, and no normalized rate is invented here: only NCRB’s own published rates are shown, with their formulas. Affidavit-based counts and NCRB counts are different datasets with different units and are never joined or added.

Statute normalization

Raw statute text from declarations is preserved verbatim on every case card. A versioned, reviewed dictionary maps known acts (IPC, BNS, PC Act, Arms Act, …) for filtering and category views; strings that don’t match stay visibly unmapped and are excluded from category rankings. A case counts as acorruption case only if it cites the Prevention of Corruption Act 1988 (or its mapped 1947 predecessor) under rule ids recorded on the record. PMLA, cheating (IPC 415–420), criminal breach of trust (IPC 405–409) and electoral bribery are never auto-classified — they are marked “needs review”. The IPC↔BNS crosswalk is display-only and marks each pair’s review status; differently numbered provisions are never silently equated.

Identity matching

Records are joined on state + constituency (unique per election), with names cross-checked and both spellings kept (the official result spelling appears on each profile). People are notmerged across elections or houses by name similarity; cross-election history will appear only after human review. Two same-name winners are distinguished by constituency in ids and URLs.

Data freshness

Case information reflects the affidavit digest dated 6 June 2024; each card repeats “the case may have changed since then”. Membership status is the election result as of 4 June 2024; deaths, resignations, by-elections and party changes since are not yet imported and profiles say so. Post-affidavit case events will be added only from primary sources, each with its own date. Current dataset: db03cf5e4a67a1b1, built 2026-07-24.

Corrections and versioning

Anyone — especially a person named on this site — can request a correction via thecorrections page. Corrections are applied in data, not by silent edits: the public log records date, old value, new value, reason and source, and the dataset version changes. The full policy lives in docs/corrections-policy.md.

Why there is no corruption score

Any single “corruption score” would have to mix accusations with convictions, enforcement activity with underlying behaviour, and self-declared data with agency statistics — hiding every judgment inside one opaque number. This site refuses that: it shows each dataset separately, with its unit, denominator, source and limits. If you see a composite corruption index built from these numbers elsewhere, treat it with suspicion.

About the map

The state view uses the 2024 Local Government Directory state and Union Territory layer, distributed by Bharatlas under CC0-1.0 / CC-BY-4.0. Its 36 polygons are projected and simplified only for screen rendering; the underlying values are joined by state code. The map is an orientation and filtering device, not a statement about comparative risk: every value is repeated in an adjacent index and accessible table so no conclusion depends on boundary size, color perception, or pointer precision. See theregistered boundary source.

Editorial guardrails

The site publishes sworn self-declarations and official statistics with attribution — it adds no allegations of its own. Statutory context considered in the editorial policy includes the Bharatiya Nyaya Sanhita 2023 (s.356, defamation) and the Digital Personal Data Protection Act 2023; the launch checklist requires a pre-publication review by an Indian lawyer, which has not yet been done — seedocs/editorial-policy.md. This site is not legal advice.