Tyche Institute · LIMEN edge-case atlas
Does a country's AI-governance paperwork — laws, registers, standards participation — translate into AI harms actually reaching a regulator or court? This scorecard joins the public ObscureAI case dataset against the PALLAS AI-Governance Observability Index per country, so the two can be compared side by side.
Cases (n) and forum reach come from the ObscureAI dataset: how many documented AI-harm cases exist for that country, and what share of them carry a regulator/court record rather than resting on media reporting alone.
Observability score is the PALLAS confidence-adjusted score (0–100): how much public evidence infrastructure exists for AI governance in that country — laws, public registers, technical trust standards, evidence lifecycle practices, learning channels. It measures the visibility of the governance apparatus, not whether harms get resolved.
Countries with fewer than 8 documented cases are shown muted with an asterisk (*) — forum-reach percentages at that scale can swing wildly on one or two cases and should not be read as stable estimates. All countries with at least 5 cases are included.
Some countries score high on observability — plenty of public law, standards and registers — while the AI harms actually documented there rarely reach a regulator or court. That gap is the accountability question this scorecard exists to surface. Three flagged cases below:
| Country | Cases (n) | Forum reach | Observability score | Observability class | Sources |
|---|
* small n (<8 cases) — forum-reach percentage is noisy at this scale, shown for completeness but not for cross-country ranking.
Sources: ObscureAI dataset (sites/obscure-ai/data/cases.json, live build) joined by country against the PALLAS AI-Governance Observability Index (papers/ai-observability-index/data/observability-250-country-matrix.csv). Join and aggregation computed by tools/build_scorecard_data.py in this site's repo. "Forum reach" here uses the same definition as the forum-reach league: share of cases carrying a regulator/court evidence tier. Observability score measures visible governance infrastructure, not case outcomes — the two are deliberately independent measurements being compared, not two views of the same thing.