dickipedia

An open, sourced encyclopedia of public power. Wikipedia-like, but different: it is not a place where anyone can write anything. Every entry is a public office-holder, every figure traces to a public registry, and the platform reports facts and the arithmetic on them — it authors no accusations, and routes citizens to lawful channels.

Volume I — the 543 members of India's 18th Lok Sabha (2024). The platform is built generically: future volumes (other categories of public office-holders) extend the same engine, primitives, and sourcing rules. MPs are simply the first subject type.

Principles (non-negotiable)

1. Public-record subjects only — sitting/contesting public officials and their public conduct, never private individuals.
2. Every fact is sourced to a public registry; nothing is asserted without a citation.
3. No platform-authored accusations — facts are declared / reported / as recorded, never characterised.
4. Self-declared data is labelled as such (e.g. criminal cases from sworn affidavits: "pending ≠ convicted").
5. No re-hosting of non-open sources — see ATTRIBUTIONS.md. Photos are hotlinked under their own licence.

How it works

dickipedia is a JSON-only, statically-exported frontend — there is no database at runtime.

public registries ──(pipeline scrapers)──▶ data/raw/*.json
                                              │  (deterministic recompute)
                                              ▼
                          data/canonical/  (per-subject files + slim index + manifest)
                                              │  (read at build time)
                                              ▼
                              Next.js static export  ──▶  any CDN / static host

• The pipeline (pipeline/*.ts, run via tsx) ingests each public registry into data/raw/.
• A deterministic rebuild (canonical → merge → inferences → publish) turns raw data into the per-subject JSON the app reads, with precomputed cohort aggregates and a timestamp manifest.
• GitHub Actions crons schedule the ingestion and auto-commit refreshed JSON; a push triggers a redeploy.

Quick start

npm ci
npm run dev            # local dev server
npm run build          # static export → out/
npm run typecheck      # tsc --noEmit
npm run lint           # eslint
npm run format:check   # prettier --check

Data pipeline & refresh

The source list lives in one place — pipeline/engine/registry.ts (each source's run, cadence, and whether it feeds the accountability score). The engine runs them:

npm run ingest -- --cadence monthly      # fast feeds (parliamentary activity, news)
npm run ingest -- --cadence semiannual   # full re-crawl of every source
npm run ingest -- --sources mplads,bonds # an arbitrary subset
npm run rebuild                          # canonical → merge → inferences → publish
npm run assert:canonical                 # validate the published output

Scheduled refresh is handled by GitHub Actions (.github/workflows/data-*.yml); a manual workflow_dispatch can re-pull any single source.

Project structure

app/            Next.js App Router (routes compose components; no business logic)
components/
  ui/           generic primitives (Badge, DataSection, StatCell, Portrait, ScoreGauge, …)
  charts/       generic chart primitives
  mp/           MP-volume components (sections, browse/, affidavit/, work/)
lib/
  format.ts     generic formatters · sources.ts  generic source registry
  mp/           MP-volume data layer (types, data, loader, bio, constants, filter)
pipeline/
  engine/       registry (source-of-truth) + runner + ingest CLI + tests
  sources/mp/   one module per MP-volume source (eci, prs, mplads, …)
  build/        deterministic recompute: canonical → merge → inferences → publish (+ assert)
  lib/          generic helpers (http, csv, text, score, paths, …)
data/
  canonical/    the truth the app reads (per-subject files, index, manifest) — committed
  raw/          pipeline scraper outputs (committed; reproducible inputs)
  config/       sources + score-band config · curated/  hand-curated, sourced facts
  cache/        HTTP response cache (gitignored)
public/assets/mp/parties/   party-symbol images (served at /assets/mp/parties/…)

Generic, reusable code (primitives, charts, formatters) lives outside the volume folders; only subject-specific content lives in components/mp/ and lib/mp/. New volumes add a sibling folder.

Contributing

See CONTRIBUTING.md. Commits follow Conventional Commits (enforced by a commit hook).

Licences

• Aggregated dataset: ODbL-1.0 · Photos: each under its own licence · Code: MIT.
• Community content (edits / discussion): separate contributor terms.

See ATTRIBUTIONS.md for full source credits and the re-hosting policy.