Purpose
The Labor Impact Dashboard tracks reported news about artificial intelligence and its effects on work: layoffs, automation, platform labor, creative rights, unions, policy, and sector change. It is a monitoring tool, not a complete census. Each row in Explore articles is one news URL that matched discovery and was parsed into structured fields.
Timespan
- Default: January 1, 2020 through today.
- Floor: Incidents before 2020 are excluded.
- On the dashboard: Adjust the date range or use presets (last week, Jan 2020 – now, etc.).
- On morethancode.org: Data is embedded when the site is built and redeployed; use
python -m src.weblocally and click Refresh data for a live pull.
What counts as an incident
An incident is a single news article that:
- Was found by an enabled source (below).
- Has a publication date inside your selected window.
- Is deduplicated by URL (one row per story).
There is no manual editorial review on the dashboard pipeline. Tags come from headlines, snippets, Event Registry concepts, and keyword rules.
Data sources
Discovery merges several batches in round-robin order so regional stories are not crowded out.
Event Registry (primary)
Requires an API key at build time. Queries combine Artificial intelligence with labor concepts (employment, layoffs, workforce, automation, gig economy, etc.). Additional pulls target Africa, Latin America, and the Middle East by publisher country. Long histories use a per-year cache plus a recent-year live query.
NewsAPI.org (supplement)
When configured, searches regional outlet bundles: Africa, Latin America, Middle East, global-majority publishers, and Asia. Subject to daily rate limits on the free tier.
DuckDuckGo, RSS, outlets, Reddit (supplement)
Regional DuckDuckGo news, Google News RSS, 20+ major outlets (Reuters, BBC, Bloomberg, ILO, etc.), and labor-related subreddits. Supplements focus on the last ~90 days when the full range is multi-year.
Regional focus
Default discovery and display bias: Africa, Latin America, and the Middle East. Geographic balancing reserves a minimum number of rows per focus region before filling the rest of the table (cap ~400 incidents on the public site).
Parsed fields
| Field | Derivation |
|---|---|
| Region | Thematic label from country, source bucket, or keywords |
| Country | Event Registry metadata, publisher, or domain inference |
| US state | Inferred from text when country is United States |
| Industry | Concept tags + keyword rules (Technology, Entertainment, …) |
| AI incident type | Layoffs, automation, gig work, policy, creative rights, … |
| Headline / source | Original article metadata; headline links to source URL |
Limitations
- English-heavy queries; global coverage depends on indexed outlets.
- Rule-based tags can misclassify; verify against the linked article.
- Static production site updates on deploy, not continuously.
- API quotas can limit any single build.
Full technical detail: docs/METHODOLOGY.md in the project repository.