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

# Maintain and review context

> Keep shared knowledge accurate, traceable, and useful.

Good context answers a concrete question: what should Dawn know, where did that information come from, and when does it apply?

## Add a fact while you work

1. Open **Context → Overview**.
2. Enter the fact in the context input and choose **Add**.
3. Review the conversation Dawn prepares to help classify and record it.
4. Inspect the resulting entry in the relevant Context section.
5. Check whether it is a proposal or an approved entry before relying on it as shared knowledge.

The Overview action helps prepare an update. It is not a guarantee that every sentence entered has become approved context.

Include the owner or source, applicable dates, definitions, and exceptions. Avoid broad claims such as “our sales data is reliable” when a narrower fact is available, such as “the completed-orders table is reconciled by Finance each month.”

## Research missing context

Use **Ask Dawn to research context** on the Overview page. Enter a focused question and select **Prepare a research draft**.

```text theme={null}
Which available datasets contain completed sales, and what evidence links
those datasets to our monthly revenue reporting? Identify conflicting
amount definitions and propose the smallest useful context updates.
```

This flow asks Dawn to use available read-only sources and prepare findings with evidence, confidence, assumptions, and priorities. Findings should remain reviewable proposals until a person reviews them.

## Review proposed knowledge

Inspect the entry and its supporting evidence in the relevant context or review surface. Use **Evaluations** to inspect the reviews available to your role. Review controls differ by context type; do not assume every kind of context follows the same approval process.

Before approving a proposal, check:

* Does the source actually support the statement?
* Is it a fact, an assumption, or a recommendation?
* Does it conflict with an existing definition?
* Which team, period, or dataset does it apply to?
* Who can resolve uncertainty?

Edit an incomplete proposal or dismiss an incorrect one using the controls available for that item. Ask an authorised reviewer if you cannot approve it.

## Correct an outdated definition

Find the existing entry first. Explain the old and new definitions, the effective date, and the source of the change. Review related entries in Data, Catalogue, Conventions, and Skills so one stale copy does not contradict the correction.

```text theme={null}
The definition of active customer changed on 1 September. Find the existing
context and show where it is used. Prepare an update that preserves the old
definition for historical reporting and clearly states the new effective date.
```

Changing context does not automatically recalculate stored datasets or rewrite existing pipelines and reports. Review affected assets separately.

## Diagnose an answer using the wrong context

Attach the relevant asset and ask Dawn to identify the data and definitions it used. Compare those with Context. Then correct the smallest underlying issue: the selected dataset, the stored definition, or the instruction in the current conversation.

Use **Ask Dawn to check for gaps** when the available knowledge is incomplete. A small, verified update is easier to maintain than a large collection of overlapping notes.
