Open-Source 'Global Fail Map' Turns Dead Startups and Cancelled Science Into an AI Agent Retrieval Layer

bar_raiser333 · reddit · 2026-09-15

A developer released Global Fail Map, an open-source interactive map cataloguing failed and abandoned human ambition worldwide: dead companies, cancelled or unfinished megaprojects, obsolete technologies and discontinued products, terminated scientific and clinical research programmes, and unrealised cities and future visions.

Each category has its own visual layer; clicking a marker opens a cited research report covering what was attempted, where and when, why it failed, what survived, what happened to the people/technology/infrastructure, and lessons for future builders.
Users can search for companies, places, projects, technologies or ideas not yet on the map.
The author argues failure research should be a first-class retrieval layer for AI agents: an agent planning infrastructure should find cancelled projects with similar constraints; one evaluating a technology should study precedents that failed for technical, economic or regulatory reasons; one helping someone start a company should retrieve failed analogues and explain failure modes, not just successes.
Stack: Next.js, React, Mapbox GL with globe projection, Zustand, Tailwind CSS, Valyu DeepResearch, Markdown reports with citations.
The repo is open source so others can clone it, add categories and build custom research layers. Open questions raised: what fields a structured failure record needs (failure modes, timelines, capital spent, dependencies, regulation, source confidence), and whether an MCP server or JSON/GeoJSON feed would help.

Original post →

Open-Source 'Global Fail Map' Turns Dead Startups and Cancelled Science Into an AI Agent Retrieval Layer

More from coding & agent