#31936 [Bug]: Cache Analytics dashboard cards show 0 for provider prompt caching (reads wrong fields — `cache_hit_true_rows`/`cached_completion_tokens` instead of `cache_read_input_tokens`)
### Check for existing issues
- [x] I have searched the existing issues and checked that my issue is not a duplicate.
### What happened?
On the **Cache Analytics** page (AI Gateway → Cache), the three summary cards **Cache Hit Ratio**, **Cache Hits**, and **Cached Tokens** all display **0 / 0.00%**, even though provider-side prompt caching (Bedrock, OpenAI Responses API) is working correctly and the cache tokens are recorded end-to-end.
The backend aggregated endpoint returns the correct value, but the frontend cards read **different fields** that are only populated for LiteLLM's own response cache (redis/in-memory), not for provider prompt caching. So for anyone using Bedrock/OpenAI **prompt caching** (as opposed to LiteLLM response caching), these cards are permanently 0.
Notably, the **charts below the cards on the same page do render data** (they group by endpoint and pick up the usage), which makes the contrast obvious: the data is present, only the summary cards fail to read it.
### Evidence
**1. The data is correct all the way through the DB and API.**
`GET /user/daily/activity/aggregated?start_date=...&end_date=...&model=bedrock_mantle/openai.gpt-5.5` returns:
```…