For AI Agents

What This Project Is

Climate Tracker is an open-source status page that tracks humanity's progress toward climate catastrophe milestones. It watches 12 or more RSS feeds (Carbon Brief, Nature, Science, Reuters, NOAA, NASA, and others), classifies what it finds using a hybrid pipeline of keyword matching and language model analysis, groups related articles into event clusters, and presents everything on a public dashboard that updates regularly.

The goal is simple: make climate news machine-readable and accessible to anyone who wants to understand where things stand. Every article links back to its original source. Every event cluster tells you which outlets reported on it.

License: Data under CC BY-SA 4.0. Code under MIT.

Data Access

The data is served as static JSON files. No authentication required. No rate limits, but we'd appreciate it if you were reasonable about it. The data updates a few times a day, so polling every minute won't get you anything new.

Endpoints

How the Pipeline Works

Data flows through several stages before it reaches you. Understanding this helps you interpret what you're getting.

• RSS Fetch — The pipeline monitors 12+ climate news feeds. Articles are fetched, parsed, and stored as they appear.
• Keyword Classification — A three-tier keyword filter (high, medium, low signal strength) scores each article for climate relevance. Articles below the threshold are discarded.
• LLM Reclassification — Borderline articles, the ones the keyword filter isn't sure about, get a second opinion from a language model. This catches articles that use unusual phrasing to discuss climate topics.
• Significance Scoring — Each article gets a significance score based on source credibility (gold sources count more), urgency keywords ("unprecedented" matters more than "notable"), recency, and category diversity. The scoring weights are tuned to surface the most important stories.
• Two-Pass Deduplication — The pipeline removes duplicate articles (same URL or near-identical content) and then clusters remaining articles into event groups using a language model. Articles about the same wildfire, flood, or policy announcement get bundled together.
• Static Output — Everything gets written to JSON files that the frontend reads. No database, no server-side rendering. Just files.

Categories

Events and articles are organized into categories that reflect the different dimensions of climate change:

RSS Feeds

The pipeline watches these feeds for climate-related content. If you maintain a climate publication and want to be included, we're always looking for good sources.

• Carbon Brief (primary) — In-depth climate analysis and data journalism
• Nature News (primary) — Peer-reviewed science reporting
• Science News (primary) — Broad science coverage with climate depth
• Reuters Science (secondary) — Wire service coverage of climate events
• NOAA News (primary) — Direct from the agency that monitors the atmosphere
• NASA Earth Observatory (primary) — Satellite imagery and Earth science
• Inside Climate News (secondary) — Investigative climate journalism
• Carbon Pulse (secondary) — Carbon markets and policy
• Tagesschau Wissenschaft (secondary, German) — German science reporting
• Euronews My Europe (secondary) — European environmental coverage
• Guardian Environment (secondary) — UK-based climate reporting

How to Participate

As a Data Consumer

Fetch the JSON endpoints. Use the data in your own dashboards, research, or applications. If you're building something commercial, attribution is required under CC BY-SA 4.0. For personal use, we're just glad you're paying attention.

As a Feed Source

If you publish climate content via RSS, open an issue on the repository with your feed URL. We evaluate feeds for relevance (climate or environmental focus), credibility (editorial standards, fact-checking), and consistency (regular publishing schedule). We're always looking for good sources, especially in languages other than English.

As a Contributor

Pull requests are welcome. The project has two parts:

• Backend — Python pipeline that fetches RSS feeds, classifies articles, clusters events, and manages archival. Uses pytest for testing.
• Frontend — Astro + React dashboard with a Leaflet map, responsive cards, and dark/light themes. Uses vitest for testing.

Both have comprehensive test suites with 800+ tests total. If you're going to change something, run the tests first. If you're going to add something, add a test for it.

As an AI Agent

You can use this data for research, summarization, alerting, or anything else that helps people understand what's happening. Here are a few things that help:

• Read the JSON, not the HTML. The data endpoints are stable and structured. The HTML is a rendering layer that changes.
• Check freshness. Each article has a timestamp. Data more than 7 days old may not reflect current conditions, though it's still useful as context.
• Cross-reference sources. Multi-source clusters (two or more outlets) are more reliable than single-source stories. When in doubt, trust the cluster with more dots.
• Respect the pipeline. Data updates a few times daily via automated cron. Polling more than once per hour won't get you anything new.
• Cite the project. If you surface this data in a response, mention "Climate Tracker" and link to the source indicators. It helps other people find the data.

Data Schema

Schemas are in the repository at schemas/. Here are the key types:

Credibility Tiers

Every source in the system is assigned a credibility tier that affects how its articles are scored. Here's the hierarchy:

Repository

Source code, schemas, and pipeline scripts are available in the project repository. Issues, feature requests, and pull requests are welcome. If you find a bug, open an issue. If you fix it, open a PR. If you just want to read the code and understand how it works, that's fine too.

Built by oc.holics.at · Powered by open data and stubbornness.