> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tensorpro.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Provenance

> Where each memory came from, and whether it can reach your model.

## Provenance-aware memory

Every record EDN stores carries a provenance class that records where the
record came from. Provenance is not a label added afterwards: it decides
whether a record can reach your model at all.

| Provenance class | Origin | Used as context |
| - | - | - |
| `USER_STATED` | Stated directly by the user | Yes, and can be cited |
| `USER_CONFIRMED` | Confirmed by the user | Yes, and can be cited |
| `DOCUMENT_VERIFIED` | Taken from an uploaded document | Yes, and can be cited |
| `ROUND` | A stored user-assistant exchange | Yes, as background for continuity; never cited |
| `AI_INFERRED` | Inferred from an assistant reply | No, not until a user confirms it |
| `AI_GENERATED` | Generated by a model and not confirmed | No, not until a user confirms it |

## Confirmed fact versus unconfirmed AI output

EDN draws a hard line between what may reach your model as context and what
may not.

* `USER_STATED`, `USER_CONFIRMED`, `DOCUMENT_VERIFIED` and `ROUND` records can
  be retrieved and placed in the SIP.
* `AI_INFERRED` and `AI_GENERATED` records are stored, but they are never
  retrieved into context. A record becomes usable only when a user confirms it,
  which stores it as a new `USER_CONFIRMED` record.

This rule is enforced when records are retrieved, not filtered out afterwards.

See [The SIP](https://docs.tensorpro.ai/concepts/sip.md) for how eligible records are chosen and cited.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.