Measure Effect
measure answers whether memory improved the run, created risk, or had no
observable effect.
Use it after a loop when you want product-level evidence rather than anecdotal success.
What To Measure
| Signal | Meaning |
|---|---|
| Context size | Did Aionis reduce prompt payload compared with raw history? |
| Wrong-history reuse | Did the Agent reuse a failed or stale branch? |
| Accepted route recognition | Did the Agent preserve current state? |
| Feedback attribution | Which admitted memories were tied to outcomes? |
| Rehydrate decisions | Did the prompt stay compact while preserving raw evidence access? |