Methodology
How we work
Every process at Nepal Next is designed to be challenged, audited, and improved. This document explains exactly how decisions are made.
Verification Framework (Levels 1–5)
Every piece of information published on Nepal Next is assigned a verification level. Levels 1–4 may be published with a visible limitations badge. Level 5 is the full standard.
Initial check
Source URL confirmed to exist. Basic plausibility check. No deep cross-referencing.
Secondary source confirmed
At least one independent source corroborates the core claim. Minor details may vary.
Official documentation
Official documents, ministry records, or court filings reviewed. Claims grounded in primary sources.
Primary source verified
Two or more primary sources. On-the-record statements. Officer sign-off required.
Full verification
Minimum two independent primary sources plus editor approval. Right of reply offered where named.
Level 5 requires: minimum two independent primary sources + officer sign-off + editor approval. Right of reply must be offered when any named person is described negatively.
Leader Tracker — Scoring Model
Every score is based on verifiable evidence, parliamentary records, and public data. Scores change only after advisory panel approval. Every version is preserved permanently. Nepal Next does not glorify power — we track relevance, trust, performance, and future value.
Parliamentary attendance, declared assets, court records, party discipline.
Delivery on promises, budget utilisation, completed projects vs. announced.
Subject knowledge, quality of statements, external assessments.
Policy positions, long-term planning, evidence of strategic thinking.
Polling data, constituent reports, media coverage sentiment.
Engagement with youth issues, generational relevance, future-facing positions.
Quality of proposed legislation, engagement with technical experts, committee work.
Scope of influence, institutional reach, cross-party or cross-sector impact.
Composite = (Integrity × 0.20) + (Execution × 0.20) + (Competence × 0.15) + (Vision × 0.15) + (Public Trust × 0.10) + (Youth Resonance × 0.10) + (Policy Seriousness × 0.05) + (National Relevance × 0.05)
Leader Tracker — Editorial Layers
Citizen Complaint Workflow
1 — Intake
System validates email, generates case ID, stores complaint.
2 — Classified
AI classifies category and province, assigns to department. Officer confirms.
3 — Verification
Officer checks sources at each level (1 to 5). Multiple verification records may exist.
4 — Verdict
Officer writes verdict from a typed enum, writes bilingual summary, submits for editor approval.
5 — Published
Chief editor approves. Public page goes live with verdict and status badge.
6 — Archived
Closed matters. Public page remains permanently. Nothing is deleted.
Verdicts are typed from a fixed enum: verified | likely_true | partially_verified | insufficient_evidence | false | needs_response | escalated. No free-form verdict labels — this keeps the public record consistent.
Suchana Board — Government Notice Ingestion
The Suchana Board ingestion pipeline runs every 15 minutes on high-priority sources (NEA, OPMCM, Supreme Court) and every 1–2 hours on standard ministry sources.
- Fetch source URL. Download documents not yet in the database.
- If PDF: run OCR via Claude Vision. Extract text.
- Classify into category and province using OpenAI classifier.
- Extract: ministry, subject, date, affected area.
- Draft plain-language summary in Nepali and English using Claude.
- Insert row with status = 'queued'. Do not auto-publish.
- Department Officer approves or rejects. Only then does it go live.
Automation level 1 (low-risk notices like road alerts) may be published without individual officer approval. Levels 2 and 3 require officer or chief editor sign-off.
Nepal Future Index — Scoring Methodology
The composite score is a weighted average of 10 category scores (0–100 each). Each category score is derived from its indicators, normalised against a developed-country benchmark.
Projections use CAGR (Compound Annual Growth Rate) from the most recent 10 years of data, applied forward to 2040. Four scenarios: current trend, moderate reform, aggressive reform, and stagnation. All projected values are clamped to zero or above.
Data quality is rated as: high confidence (≥7 data points, consistent sourcing), partial (3–6 points), or building (fewer than 3 points). Sparse categories are excluded from the composite score.
AI Attribution Standard
AI-assisted civic research | Reviewed by [Name], [Department Officer Title]
No AI output is labelled as coming from a human.
Every AI call stores: model name, prompt version, input hash, and the ID of the officer who reviewed it. Any published output can be traced to its source.
Task routing: summarisation, translation, and sensitive political content use Claude. Classification, embedding, and bulk non-sensitive processing use OpenAI.
Data Sources
Each data point on the Nepal Future Index shows its source organisation, year, and data type. Sources are checked for methodology, sample size, and independence before being registered.
This methodology document is version-controlled. Changes are logged. The current version applies from 28 May 2026. Read our corrections policy →