Admission Dataset Export
Every guide/context decision can become a training and audit row.
Admission Dataset Export is the product path for a future learned admission policy: a model or policy layer that predicts whether memory is useful, risky, or irrelevant in a specific Agent state.
What Gets Exported
| Field | Meaning |
|---|---|
| candidate memory | The memory considered by a guide or firewall pass. |
| admission action | use_now, inspect_before_use, do_not_use, or rehydrate. |
| reason | The Runtime explanation for that action. |
| prompt inclusion | Whether any part of the memory reached Agent context. |
| host use | Whether the Agent actually used it. |
| outcome | Whether the run succeeded, failed, or remained inconclusive. |
| feedback attribution | The link between memory exposure and outcome evidence. |
Why It Matters
Most memory systems know what they stored and what they retrieved. Aionis can also know what was admitted, what was blocked, whether the Agent used it, and what happened afterward.
That creates a better data asset:
candidate -> admission decision -> prompt exposure -> host use -> outcomeSDK
import {
memoryAdmissionDatasetJsonlFromGuide,
} from "@aionis/sdk";
const jsonl = memoryAdmissionDatasetJsonlFromGuide(guide);
await fs.promises.writeFile("admission-dataset.jsonl", jsonl);What Not To Export
Do not export full prompt payloads unless you have a clear privacy and retention policy. The default export path should prefer decision records, IDs, reasons, and bounded context summaries.