# Agent memory benchmarks all measure recall. Almost none measure provenance — and that's the part that bites in production. — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/bfcf5513a46dc199

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-11-17 18:13 ingested 2026-11-18 02:36

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1xyrr1u/agent_memory_benchmarks_all_measure_recall_almost/)

## retained text



```
I've been building an agent that runs continuously and keeps memory across sessions. The thing that surprised me wasn't retrieval quality — it was that I had no way to answer "what did it believe last Tuesday, and who put that there."

Every memory benchmark I can find measures recall. Recall is easy to measure and easy to demo. Provenance is neither, so it gets skipped.

The failure mode I hit: the agent stored an assumption early on, acted on it weeks later, and by then the original context was gone. Nothing was wrong with the retrieval. The store returned exactly what it had been given. The problem was that nothing recorded *when* it learned that, *from what source*, or whether it had since been contradicted.

Three things I'd now build in from day one:

1. Every memory write carries a source and a timestamp, and is append-only. Corrections are new rows, not overwrites.
2. Retrieval returns provenance alongside the fact, so the agent can say "I believe X, recorded on date Y from source Z" instead of asserting X flatly.
3. A way to see what changed between two points in time — not just current state.

Curious how others handle this. Are you storing provenance with your memories, or is it just content plus an embedding? And if you do store it, does it actually reach the model at inference time, or does it get dropped in the context assembly step?

Genuinely asking — I haven't found a clean pattern for the third one.
```



document fetched 2026-11-18 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1xyrr1u/agent_memory_benchmarks_all_measure_recall_almost
