← Watchboard

How Watchboard Works

What is Watchboard?

Watchboard is an open-source intelligence dashboard that tracks world events across multiple topics—from active military conflicts to political developments, economic crises, and cultural events. Each tracker is an independent dashboard with its own data, maps, 3D globe, and analysis.

The platform is designed for anyone who wants structured, sourced, and transparent reporting on complex ongoing situations. Rather than presenting a single narrative, Watchboard surfaces data from multiple perspectives and lets readers evaluate credibility through explicit source tier classifications.

Automated Data Pipeline

Every day at 14:00 UTC, an AI-powered pipeline updates all active trackers. The process runs through six stages, each with built-in safeguards to prevent bad data from reaching the published site.

1
Resolve

Determines which trackers are due for update based on their configured interval. A tracker set to update every 3 days will not run again until its window opens.

2
Review

Generates a manifest of event gaps—missing days in the timeline—and cross-references related trackers to avoid duplication across overlapping topics.

3
Update

Each tracker runs in parallel with its own AI agent that searches the web, verifies sources against known outlets, and updates the data files with new events, KPIs, and analysis.

4
Validate

All data passes through Zod schema validation. If structural errors are found, a fix agent automatically corrects them before the data is accepted.

5
Build Gate

The entire site is built from scratch to catch any runtime errors before committing. If the build fails, the update is rejected and no data reaches production.

6
Metrics

Every run—pass or fail—is recorded to the ingestion metrics dashboard, providing full auditability of the pipeline’s health over time.

Data Quality & Source Tiers

Every data point in Watchboard carries a source tier classification. This system is the backbone of the platform’s commitment to transparency: readers should always know where information comes from and how reliable it is likely to be.

Tier 1
Official / Primary Government statements, UN reports, IAEA assessments, court filings, official military communications
Tier 2
Major Outlets Reuters, AP, BBC, CNN, Al Jazeera, AFP—wire services and major international news organizations
Tier 3
Institutional Think tanks (CSIS, Brookings), NGOs (HRW, MSF), academic institutions, research organizations
Tier 4
Unverified Social media posts, anonymous sources, unattributed claims, Telegram channels, unverified imagery

OSINT confirmation rule: an event is marked “confirmed” when it has at least 1 Tier-1 source or 2 or more independent Tier-2 sources. Single-source Tier-3 or Tier-4 reports remain flagged as unconfirmed.

Casualty figures include a contested classification—yes, no, evolving, or heavily—to explicitly flag disputed numbers rather than presenting any single estimate as authoritative.

Distribution

Watchboard data is available through multiple channels beyond the web dashboard, making it easy to integrate into your own tools, agents, and workflows.

🔗
Public JSON API

Free, no authentication, no rate limits. Static JSON files served from CDN with full CORS support. Endpoints for tracker metadata, events, KPIs, breaking news, and search. Documentation →

🤖
MCP Server

Model Context Protocol server for AI agents (Claude, ChatGPT, Cursor, and others). Provides 7 tools and 3 resources for querying tracker data programmatically. Setup guide →

🧠
Claude Skill

Install the Watchboard skill to let Claude query trackers, interpret KPIs, and surface breaking news automatically. Available in the Anthropic skills registry. Installation guide →

✉️
Telegram Channel

Automated breaking news posts triggered by an hourly scan. High-priority events pushed in real time. Join @watchboard_dev →

🥋
Bluesky

Automated social posting from the curated social queue. Breaking news, daily digests, and video briefs. Follow @watchboard.bsky.social →

📨
Newsletter

Weekly digest email summarizing the most significant events across all active trackers, with subscriber management and one-click unsubscribe.

Cross-Tracker Intelligence

Related trackers are aware of each other. Before updating, each tracker receives a “sibling brief”—a summary of what related trackers already cover. This prevents event duplication across overlapping topics. For example, a drone strike reported in both the Iran Conflict and a regional theater tracker will only appear in whichever topic it most directly belongs to, with a cross-reference in the other.

This sibling awareness system also enables richer context: when a political development in one tracker has downstream effects on another, the pipeline can surface that connection without duplicating the underlying data.

Schema Enforcement

All data is validated against strict Zod schemas before it enters the system. These schemas enforce structural and semantic constraints that prevent common data quality issues:

  • No future dates allowed—events cannot be dated ahead of the current day
  • Strike and retaliation map entries require weapon type and timestamp
  • Source poles must be one of: western, middle_eastern, eastern, or international
  • Economic indicators must include directional trends (up or down)
  • Every event must have an ID, year, title, detail text, and at least one source
  • Map coordinates are validated against the tracker’s configured geographic bounds

When the AI updater produces data that fails validation, a dedicated fix agent reads the error messages, identifies the root cause, and corrects the data automatically. If the fix agent also fails, the entire update is rejected.

Roadmap

Watchboard's product roadmap is public and live. View the roadmap → for what's shipped, in progress, planned, and captured as ideas — broken down by area (performance, growth, content, accessibility, infrastructure) with priority and effort estimates. The page is fed directly from the codebase, so it's never out of date with what's deployed.

The breaking-news audit page shows every triage decision the hourly + 6-hourly scans made in the last 14 days — useful when calibrating which headlines should reach Telegram.

Transparency

Watchboard is built on the principle that intelligence tools are only useful if you can verify how they work. Everything about this platform is open and auditable.

  • Ingestion metrics—View run history, success rates, and validation errors on the Status dashboard
  • Open source—Full source code, data files, and update history available on GitHub
  • RSS feeds—Subscribe to global, breaking, light-scan triage, or per-tracker feeds. Browse all at /feeds (HTML index, JSON, and OPML bundle for one-click reader import)
  • Changelog—Every data update is committed to git with a full diff, so you can see exactly what changed and when

Social Media

Watchboard publishes curated updates to X/Twitter through an AI-powered content strategy system. Rather than auto-posting every update, an AI content strategist reviews all tracker data each day and decides what is worth posting—optimizing for impact within a configurable monthly budget.

1
Curate

The AI receives all tracker digests, KPIs, and events for the day along with the remaining monthly budget and historic tweet performance. It selects the most newsworthy items and chooses the best format: digest, breaking news, thread, hot take, data visualization, or meme.

2
Judge

Every draft tweet is scored and fact-checked against tracker data. Numbers, claims, and quotes are verified. Each tweet receives a verdict: PUBLISH (auto-approve), REVIEW (needs human approval), HOLD (defer), or KILL (reject). Tweets with failed fact checks are automatically rejected.

3
Review

The Social Command Center dashboard shows all queued tweets with pixel-accurate X previews, judge assessments, and fact-check results. Tweets can be approved, edited, or rejected individually or in batch.

4
Schedule

Approved tweets are distributed across four daily time slots (08:00, 13:00, 18:00, 22:00 UTC) to reach audiences across time zones. A cron workflow posts them automatically at their scheduled time.

Tweets are published in up to four languages (English, Spanish, French, Portuguese) and use a mixed persona system—analyst voice for data-driven updates, journalist voice for breaking news, and witty voice for memes. The system targets a monthly budget of $1.00 using X’s pay-per-use API pricing.

Built With

Astro 5Static site generator
ReactInteractive islands
CesiumJS3D globe visualization
Leaflet2D interactive maps
ZodSchema validation
Claude Code ActionAI data pipeline
GitHub ActionsCI/CD orchestration
GitHub PagesStatic hosting
globe.glWebGL globe rendering
Solar System ScopeEarth night texture (CC BY 4.0)