Pool info
Scope, cadence, evidence boundary, and interpretation rules for this pool.
Purpose
This pool maintains a public answer to whether popular AI models are getting worse. It researches the internet repeatedly and preserves the evidence behind each update.
Interpretation rules
- An outage is not capability degradation.
- A product-default or routing change is not automatically a base-model change.
- A cross-model leaderboard snapshot is not a longitudinal trend.
- Unsupported anecdotes can motivate research but cannot settle the answer.
Cadence
The intended schedule is every 12 hours. The current public release is stale until the cloud runner completes a new verified cycle.