BullDozer 🚜

Bulldoze the noise. Mine the signal.

Marketing analytics & research site: 164 public, parsed datasets — surveys, IMF & World Bank macro, markets and people. Every number sourced.

Live: azenha.ai/bulldozer

Where it lives — source: this repo (~/Projects/bulldozer, github.com/AzenhaAI/bulldozer). Deployed: npm run build → dist/ is rsynced into ~/Projects/azenha/bulldozer/ and served at azenha.ai/bulldozer via Cloudflare Pages. See docs/OVERVIEW.md for the full map.

What's inside

Header tabs: New · Top · Macro · Polls · Geo · Biz · Edu; a right-edge burger holds the rest (Quiz, Glossary, About, GitHub). Global search is on 🔍 / Cmd-K; light/dark toggle sits beside it.

Tab	Route	What
New	/	Data stories, featured dashboards and release notes
Top	/explore	Interactive explorer: bubble, correlations, beeswarm, heatmap, trajectories, index builder
Macro	/macro	104 objective statistics — IMF WEO, World Bank WDI/Findex, OECD, BIS, OWID, Maddison, Eurostat, TI CPI, Freedom House, RSF, PISA, UNHCR, Big Mac
Polls	/surveys	60 opinion datasets — WVS, WHR, V-Dem, Afrobarometer, Latinobarómetro, ESS, Reuters DNR, Hofstede and more
Biz	/markets	Largest companies & most valuable brands, with directories and metric switchers
Geo	/geo	Country profiles (name, official name, ISO codes, currency), similar-country matching, side-by-side comparison
Edu	/edu	Field guide: dashboard types, BI tools, how the site parses its own data

Plus 232 static country profiles (/country), per-dataset dashboards with KPIs / movers / maps / trends / breadcrumbs, downloadable CSVs, a glossary, and detailed source-level results dashboards.

Highlights

• Rosling bubble chart — income × life expectancy × population, animated 1800→2020 (Gapminder long series, 40k+ observations)
• Correlation matrix — Pearson r across 15 curated indicators
• Composite index builder — weight indicators into your own country ranking, live re-ranking
• Similar countries — nearest-neighbour matching on global percentile profiles
• Country heatmap — percentile grid with topic/region/search filters
• Cultural map of the world — Inglehart-Welzel dimensions from WVS Wave 7
• Global search (Cmd-K) across every dataset and country; canonical country names (common + official + ISO + currency) independent of source spelling
• Every dataset page ships schema.org/Dataset JSON-LD, its own OG image, breadcrumbs, prev/next and a CSV download

Principles

1. Only public data. Cross-country surveys and open statistical sources; no proprietary or internal data — ever.
2. Every number sourced. Each dataset carries source, licence, URL, unit and parse date, surfaced in the UI.
3. No client framework. Astro static build; interactivity is hand-written browser JS. 380+ pages, fast everywhere.

Architecture

src/
  data/
    datasets.ts        # auto-registry: import.meta.glob over surveys/*.json + macro/*.json
    surveys/  macro/    # normalised observations { meta: provenance, data: [Observation] }
    sources.ts          # catalogue of survey programmes (live + collected)
  lib/
    analytics.ts        # KPIs, movers, group rollups, period-over-period change
    countryIndex.ts     # per-country profiles: latest value + world rank per indicator
    topics.ts palette.ts worldgeo.ts …
  components/           # 25 viz components (bubble, heatmap, matrix, maps, index builder…)
  pages/                # index, explore, macro, markets, surveys, geo, country/[slug],
                        # dataset/[slug], stories/, glossary
scripts/
  parse_*.mjs           # one parser per source family → normalised JSON
  make_og_datasets.mjs  # per-dataset OG images (sharp)

Data model — one long-format Observation: { entity, iso, period, value, group? }. Surveys, macro indicators and market data all flow through the same analytics layer. Drop a normalised JSON into src/data/surveys/ or src/data/macro/ and the dataset page, CSV endpoint, country profiles and explorer pick it up automatically.

Develop

npm install
npm run dev     # http://localhost:4321/bulldozer/
npm run build   # → dist/ (static, host-agnostic)

Parsers (npm run parse:*) read curated public exports and soft-skip when a source file is absent, so a fresh clone builds out of the box.

Automation

Workflow	Schedule	Does
update-data.yml	weekly	re-runs parsers, commits data changes
telegram-digest.yml	weekly	posts a top-movers digest (env: TELEGRAM_BOT_TOKEN, TELEGRAM_CHAT_ID)

Licence

Code is MIT. The datasets remain the property of their original publishers — every dataset page credits its source and licence; check the source's terms before redistributing data.