Full Transparency

Scoring Methodology

Version 1.0 · May 2026 · Published permanently · Updated each election cycle

The Core Principle

NERVE Data produces two separate scores for every candidate. These scores are never combined into a single number. Combining them collapses two fundamentally different questions into one misleading answer.

The Integrity Score answers: what does this person's declared public record look like?

The Win Probability answers: what does historical data say about their chances of winning?

A candidate can score high on Integrity and low on Win Probability. A candidate can score low on Integrity and high on Win Probability. Both are valid, distinct, and important pieces of information. Showing them separately is the only honest approach.

Truth Engine — Public Free Layer

The Truth Engine runs on every candidate profile, free, always accessible. It powers both scores using only publicly declared data.

Score 1 — Integrity Score (0–100)

Measures the cleanliness of a candidate's declared public record. Higher = cleaner declared record. All figures labeled as declared — never implied as actual.

Criminal cases declared — serious nature, recency weightedPrimary

Declared family asset growth since last electionSecondary

Legislative Participation Score (attendance 30% + questions 40% + debates 30%)Secondary

Constituency fund utilization (MPLADS/MLA LAD)Tertiary

Asset declaration completeness flagFlag only

Score 2 — Win Probability (Range %)

Measures likelihood of electoral success based on historical patterns. Always expressed as a range — never a single number. Always shown with two scenarios.

Last election margin — strongest single predictor30%

Anti-incumbency cycle (2+ consecutive terms = elevated risk)20%

Party wave strength — overrides individual metrics in wave conditions20–50%

Demographic alignment — B2B layer only, never displayed publiclyB2B only

Local issue salience — RSS intelligence layerTertiary

Seven Stress-Test Fixes

Before launch, our methodology was stress-tested against seven known failure modes in Indian electoral data. Each was addressed before the system went live.

Never combine scores. Integrity and Win Probability are always displayed separately. A combined score produces misleading numbers for candidates where integrity and electability diverge — which is common in Indian politics.

First-time candidate handling. Missing data is flagged explicitly — never scored as zero. First-time candidates receive a Low confidence rating and a data flag. They are not penalized for the absence of historical data.

Wave election override. In strong wave conditions, party wave strength increases from 20% to 50% of Win Probability weighting. Individual performance metrics compress proportionally. Wave conditions are declared explicitly on every affected profile.

Caste alignment — B2B only. Demographic alignment scoring exists and is used in party intelligence products. It is never displayed on public candidate profiles. This is a business boundary, not a data gap.

Declared asset labeling. Every asset figure on this platform is labeled as declared. We never imply actual wealth. Candidates may under-declare; we flag declaration completeness anomalies without making character judgements.

Legislative Participation Score. Raw attendance is replaced with a composite: attendance (30%) + questions asked in session (40%) + debates participated in (30%). Attendance without participation is not credited equally.

Recency weighting on criminal cases. Cases older than 10 years with no conviction carry 30% of the weight of recent cases. Old cases without outcomes decay in scoring weight. Recent behavior matters more than distant history.

Confidence Levels

Every Win Probability output carries a confidence level: High, Medium, or Low. This is displayed on every profile.

High: Strong historical data. Incumbent with clear margin history. Multiple election cycles in database. Predictive model has solid foundation.

Medium: Partial historical data. Some gaps in constituency or candidate history. Estimate is informed but not fully validated.

Low: First-time candidate. Constituency with insufficient historical data. Wave election active. Limited basis for confident prediction.

A Low confidence rating is not a negative signal about the candidate. It is an honest statement about data availability. We never hide uncertainty behind false precision.

Data Sources

All data on NERVE Data is sourced from the following public records only:

ADR (Association for Democratic Reforms) — structured compilation of ECI affidavit data. Primary source for launch.

ECI Affidavit Archive (affidavitarchive.nic.in) — direct PDF affidavits. Being integrated for direct extraction in Month 2-3.

PRS Legislative Research — Parliament attendance, questions, debates, voting records.

MPLADS Portal — fund allocation and utilization data.

Court public records — as declared in candidate affidavits only.

Every data point is source-tagged and date-stamped. Every score change is logged with reason. The full audit trail is maintained permanently.

What This Model Cannot Do

Honest disclosure of limitations is non-negotiable for NERVE Data.

We cannot detect undeclared assets. Benami holdings, family transfers before declaration, and off-books wealth do not appear in affidavits and therefore do not appear in our scoring.

We cannot predict upset victories. Black swan local events — candidate death, last-minute alliances, community mobilization — are not in any historical dataset and will surprise our model as they surprise every analyst.

We cannot account for ground reality. Booth-level mobilization, last-mile campaign effort, and voter mood on polling day are not in public data. Our model scores declared records and historical patterns — not ground campaign quality.

Version 1 accuracy is estimated at 60–65%. This will improve with each election cycle as historical data deepens. Version 3 (Year 2) targets 75–80%. Version 4 (Year 3) targets 85%+.