Abu Dhabi Power Network — V3 · Intelligence Map

Full-scale, no-cap network map of Abu Dhabi's key industries, institutions and key people, organized in four layers — now with an automation pipeline for news and your LinkedIn network.

Files

File	What it is
abu_dhabi_power_hub.html	The map. Open in any browser (keep it in this folder — it loads the three .js files below).
network_data.js	Source of truth. All sectors, institutions, people, ownership, family ties, aliases. Edit → refresh browser. No build step.
news_data.js	Auto-generated news signals (tools/match_news.py).
my_network.js	Auto-generated LinkedIn overlay (tools/import_linkedin.py). Private — keep local.
news_inbox/	Drop news articles here as .md/.txt (two [SAMPLE] items included — delete them).
tools/	Automation scripts (Python 3, stdlib only).
data_dictionary.md	Data model and operating rules.
abu_dhabi_power_hub_v1_plotly_backup.html	Old V1 Plotly map, kept for reference.

The layer model

Layer	Contents
L0 · Ruling Core	Al Nahyan ruling family, Presidential Court, Executive Council, SCFEA, UAE Cabinet
L1 · Sovereign Capital & State	ADIA, Mubadala, ADQ, IHC/Royal Group, department chairs, ministers, super-connectors
L2 · Operators & Sector Leaders	Operating companies and their CEOs across every sector
L3 · Private & International Partners	Energy majors (TotalEnergies, bp, Shell, Exxon…), Big Tech (Microsoft, OpenAI, NVIDIA), global finance (BlackRock, Goldman), consulting/audit/law gatekeepers, local private groups (Waha, Trojan, Rotana…)

Local-people policy: global HQ executives (Sam Altman, Jensen Huang, Larry Fink, the energy-major CEOs, …) are deliberately excluded — the people layer is local. L3 companies are anchor nodes; the people who serve them in the UAE surface through the LinkedIn overlay, and get promoted into PEOPLE once their local role (country manager, UAE managing partner, Gulf CEO) is verified.

Sectors = GICS 11 + government-side, identical across every region: Energy 能源 · Materials 原材料 · Industrials 工业 (incl. transport, defense companies, professional services) · Consumer Discretionary 可选消费 · Consumer Staples 必选消费 · Health Care 医疗保健 · Financials 金融 · Information Technology 信息技术 · Communication Services 通信服务 (telecom + media + entertainment/sport) · Utilities 公用事业 · Real Estate 房地产 — plus Government & Political, Sovereign Capital, Education & Research, Family Conglomerates. Every region covers all 11 GICS sectors (checked by tools/gics_migrate.py conventions; keys: energy, materials, industry, consumer_disc, consumer_stap, health, finance, tech, comm, utilities, realestate, gov, sovereign, education, conglomerate).

3 edge types (roles, ownership/control, family) · power score drives node size. Current scale: ~100 people · ~143 institutions · ~298 edges. No cap.

Automation pipeline

0. Fully automated loop (no human needed)

.github/workflows/update.yml runs every 6 hours on GitHub Actions: fetch fresh news (tools/fetch_news.py, Google News RSS for 20 entity queries) → match to network nodes → detect co-mentions → commit. The Alicloud server pulls every 5 minutes, so new signals appear on the live map unattended. Candidate relationships accumulate in data/suggested_edges.csv — entity pairs that keep sharing articles but have no edge yet; review and promote the real ones into ownership.csv/roles.csv.

For manual data edits, one command does everything:

python tools/update.py "describe the change"

(validates + imports CSVs, re-matches news, commits, pushes → live in ~5 min)

1. News → map signals

python tools/match_news.py

Drop articles into news_inbox/ (2026-08-25_title.md; first line = title; optional source: / url: lines). The matcher scans them against every node name + alias (AKA block in network_data.js), handles line-broken names, suppresses fragment false-positives ("Al Jaber" inside "Sultan Al Jaber"), and regenerates news_data.js. Matched nodes get a gold badge on the map and a Recent signals section in their panel. Idempotent — re-run any time.

2. LinkedIn → warm-intro overlay

python tools/import_linkedin.py

Export your connections first: LinkedIn → Settings & Privacy → Data privacy → Get a copy of your data → Connections → put Connections.csv in this folder. The importer fuzzy-matches employers against all institutions (suffix-stripping, alias-aware) and writes my_network.js. Institutions you can reach get a green dashed ring and your contacts appear in their detail panel — so the path you → contact → institution → key person is visible on the map. Privacy: my_network.js contains your personal contacts — never share it with the map.

3. Rolling growth — the network gets larger every day

The database is not static. Growth runs on three streams, executed by the daily 9 AM enrichment agent (a scheduled Claude session on this machine):

• Leadership backfill: data/backfill_queue.csv (regenerated by tools/backfill_queue.py) lists every institution still thin on leadership, priority-ordered. The agent works the queue daily — researching each company's current chairman/CEO/board with sources and adding them.
• News-driven adds: new companies/people revealed by the tier-1 + Google News pipelines; suggested co-mention edges promoted after verification.
• Verification rotation: ns flags re-checked against official sources and flipped to v (or corrected), rotating through the regions weekly. Every dossier also carries a Search on LinkedIn ↗ button (people-search prefilled with the entity name) for manual network digging; automated LinkedIn scraping is deliberately not part of the system (ToS) — use the official Connections.csv export + tools/import_linkedin.py for your own network.

Adding more later

The same pattern extends to any content: put items in an inbox, match on the alias table, emit a *_data.js overlay. Candidates: press releases, tender awards, event attendee lists, podcast transcripts.

Design system (V5 — "the park map": Navigate → Locate yourself → Find your path)

The engine lives in map_template.html (stamped into every region page by tools/build_regions.py); the landing page is the hand-editable index.html.

• Deep navy ground (#0A1120), ink tokens for all text, white for emphasis (selection, routes) and one luminous accent (#7C8CFF) for you and your links. Fonts: Sora (display) and Inter (UI, canvas labels).
• Industries = the site's 10 existing sector groups (15 sector keys folded, nothing invented), each with a fixed colour, an icon and a name, used identically in the map legend, the phone chip strip, filters, profiles, district tags and the landing page. Hues are the dataviz reference palette's dark steps, validated against the navy surface; identity is never colour-alone (icons + labels everywhere).
• Legend: click/tap industries to highlight (multi-select, Reset; others fade but stay in context) · separate connection key (role, ownership, family, unverified, potential, former, your connection, demo, route) · shapes · latest signals. Desktop: floating sidebar; phone: chip strip under the header + "Map key" sheet.
• Views: Network (tier rings) · Industries (park districts with name tags) · Focus (ego rings) · Ownership (control hierarchy).
• You on the map: people you add (by name on the map, LinkedIn Connections.csv export, a profile link, or manually) appear as a "You" pin linked to them; organizations where unmatched contacts work become possible ways in. Updates live, no reload.
• Routes: Dijkstra from the target over public relationships; each entry point yields one route; ranked by hops, verified vs unverified records, how broad a shared organization is, and (only if you rated it) your relationship strength. Every hop shows its basis and class: Confirmed (by you / public record) · Unverified record · Potential (shared organization, contacts at an organization) · Demo data. Routes are presented as possible introductions, never guarantees.
• Path mode: unrelated nodes fade, segments light up in sequence, synced step list with Back/Next (and ←/→), alternatives as chips; no-route and "where do you start" states explain what to do next.
• Guided tour (first visit only; "Guide" replays it): 3 steps that drive the real map, with Next / Back / Skip / Replay and Try a demo (?demo=1) — sample contacts that are clearly labeled and never mixed with real data.
• Phones: map-first; bottom sheets (peek / half / full, swipe to dismiss) for profiles, routes, tour, My Network and the key; map controls (recenter, fit, full screen, key); ~24 key nodes and ≤7 labels by default with collision culling, more on select/expand/ zoom; 44px targets; safe areas, dvh and visualViewport; reduced-motion respected.

EN / 中文 switches UI chrome only; all data stays English. window.__map exposes a debug/automation handle.

Deployment

Repo: https://github.com/qizhangumich/power_network — index.html is the site entry (identical to abu_dhabi_power_hub.html); zero-config static hosting.

• Vercel: import the repo at https://vercel.com/new (framework preset: Other, no build command, output dir default). Every git push then auto-deploys. Or CLI: vercel login once, then vercel --prod in this folder.
• Privacy: .gitignore keeps my_network.js (your LinkedIn contacts) and Connections.csv out of the repo — the public site simply shows no LinkedIn overlay. Never commit these files.

Data quality

Facts vs. analysis stay separate. Roles/edges marked needs source are public-record but not yet re-verified against an official page. L3 "advisory ecosystem" edges (consultancies → ADDED etc.) are deliberately generic — replace with specific sourced engagements as you learn them. Parked V1 names pending role verification: Etienne Petit, Sherif Tawfik, Mohamed Almarzooqi.

Regions — the GCC

Eight maps on one engine, linked by the folded region menu in the top bar:

Path	Region	Core
/	Abu Dhabi	Al Nahyan · ADIA/Mubadala/ADQ/IHC · full news pipeline
/dubai/	Dubai	Al Maktoum · ICD/Dubai Holding · Emirates/DP World/Emaar
/northern/	Sharjah · RAK · Fujairah · Ajman · UAQ	Al Qasimi/Al Sharqi/Al Nuaimi/Al Mualla · Crescent/BEEAH/Wynn
/saudi/	Saudi Arabia	Al Saud (MBS) · PIF/Aramco · giga-projects
/qatar/	Qatar	Al Thani · QIA/QatarEnergy · LNG partners
/bahrain/	Bahrain	Al Khalifa · Mumtalakat/Bapco/Investcorp
/oman/	Oman	Al Said · OIA/OQ/PDO
/kuwait/	Kuwait	Al Sabah · KIA/KPC · merchant families

Each region: <region>/network_data.js (same schema). tools/build_regions.py stamps every region page from the root index.html, so design changes propagate everywhere (update.py runs it; a region ships once its network_data.js exists — add a new one by adding a row to REGIONS and a dataset). News automation + CSV layer currently cover Abu Dhabi only; per-region parameterization is the next step. All non-AD datasets are V1 backbones — review the ns flags.