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

> ## Agent Instructions
> Orion is a collaborative analytics platform built by Gravity (bygravity.com). It connects to data warehouses, to Looker and dbt for existing business logic, and to a small set of external-context sources, and it lets teams ask questions in natural language to get shared analyses, dashboards, reports, and slide decks. The Connecting Data Sources page is the authoritative list of supported sources. When referencing Orion features, always link to the relevant documentation page. Orion is not open-source; it is a commercial SaaS product accessed at runorion.com.

# Observatory Cards

> Act on what Orion notices after an answer

An Observatory card is a short prompt that appears under one of Orion's answers. Orion reads each finished answer and, when it spots something worth acting on, adds a card to it. A card is not part of the analysis. It offers one decision: save a preference, add a fact to Context, ask a follow-up, answer a question, or turn a suggestion into an analysis.

Cards are for analysts and admins. If you can edit a project, you see its cards. If you can only view it, you do not. Nothing happens until you click: Orion never saves, answers, or starts anything on your behalf from a card.

## Overview

* [Cards under an answer](#cards-under-an-answer)
* [Questions from Orion](#questions-from-orion)
* [Suggested analyses](#suggested-analyses)
* [Observatory evidence in answers](#observatory-evidence-in-answers)
* [Where else cards appear](#where-else-cards-appear)
* [The Observatory page](#the-observatory-page)
* [Who sees what](#who-sees-what)

## Cards under an answer

Each card sits below the answer it came from. Memory, Context, Question, and Suggested analysis cards each have a dismiss control. Follow-up cards go away when you click them, and Citation links stay in place.

| Card | What it offers |
| - | - |
| Memory | A personal working-style preference worth keeping. **Remember** saves it to [Memory](https://docs.runorion.com/chat/memory.md), **Dismiss** removes it |
| Add to Context or Update Context | A proposed new page, or a change to an existing page, in [Context](https://docs.runorion.com/knowledge-base/overview.md). **Apply** saves it, **Dismiss** removes it |
| Follow-up | A suggested next question. Click it and Orion sends it as your next message |
| Citation | A link to a dashboard, report, or slide deck Orion drew on. Click it to open the deliverable |
| Question | Something only you can answer. See [Questions from Orion](#questions-from-orion) |
| Suggested analysis | A proposed piece of work. See [Suggested analyses](#suggested-analyses) |

A few details matter when you act on a card:

* **Memory cards** show a status after you click **Remember**: **Queued**, **Processing**, **Saved**, or **Failed**. **Unconfirmed** means Orion could not confirm the save went through, and **Accepted** means it was accepted but no progress has been reported yet. Orion shows at most two memory cards per answer.
* **Context cards** show a short summary. A card that changes an existing page names the page it updates. After you click **Apply**, the card reads **Saved to Context** or **Updated Context**. When the change needs review first, it reads **Sent for approval**.
* **Follow-up cards** go away once you click them.

Orion does not show a Memory or Context card for something it already saved in the same answer, or already asked your permission to save, so you are not offered the same fact twice.

> [!TIP]
>
> A card is a proposal. Use **Dismiss** freely. A card you ignore stays where it is and does no harm.

## Questions from Orion

Some questions can only be answered by a person: what caused a move in a metric, what a term means in your business, whether a finding is right. Orion keeps a list of open questions for the project and brings one up only when your conversation is about its subject. A question about data that several projects share appears only in the projects that hold at least one of its sources. It shows at most one question or suggested analysis per answer, and never inside the conversation opened to handle one. A metric whose latest reading is 0 while its usual level is not does not raise a question: Orion treats it as a gap in the data, such as a monthly total on the first of the month, not as a move. A total that builds up through a month raises none either while it keeps rising, including the first readings of a new month after it resets. A rate that climbs steadily still does.

A question card is titled **Orion needs your input**. It states the question and says that your answer helps Orion verify a finding. When Orion has a reason on record, **Why Orion asks** explains why this question came up now. You have three choices:

* **Save answer**: type what you know in the box that reads **Share what you know…** and save it. Answers can be up to 4,000 characters. Orion uses your answer in its next review of the finding.
* **Investigate with Orion**: open a side conversation in the same project. Orion can look at your data sources while you work out the answer together. When you tell Orion the answer, it records it. If you do not know or decline, nothing is recorded.
* **Not relevant**: dismiss the question.

After you answer, the card shows **Answer recorded** with the question and every answer so far, and, while the question is still open, **Add your answer** lets you add another. Once Orion has used the answers the card reads **Answers incorporated**. A closed question shows **This question no longer accepts answers.**

## Suggested analyses

A suggested analysis is a piece of work Orion thinks is worth running because your conversation touched its subject. The card shows a title and what you would get. When there is more to say, **View details** shows why Orion suggests it.

Clicking the main button, which reads **Discuss analysis** unless the card says otherwise, does not start the analysis. It opens a short side conversation where you and Orion agree what you want. In that conversation, Orion asks which output you want: a slide deck, a dashboard, metrics, or a narrative report. Only when you confirm does Orion start the work. It runs in the background and Orion posts the result to the project's board when it is ready.

If you are not interested, click **Not now**. If a teammate has already started the same analysis, Orion tells you and does not start a second one.

> [!NOTE]
>
> Starting an analysis changes what Orion builds for the whole project, so it is available only to analysts and admins.

## Observatory evidence in answers

Orion can also use what Observatory has verified to write an answer. When you start a conversation in a project, Orion looks up verified Observatory claims that match your first message and adds a few of them to what it knows for that conversation. Orion adds at most three claims. It looks once, at the start, and later messages reuse that set. The one exception is a lookup that did not finish, below.

Orion adds none when any of these hold:

* The project's [**Observatory context**](https://docs.runorion.com/core-concepts/projects.md#observatory-context) switch is off.
* You work with viewer access, including a viewer with Extended Viewer Settings, rather than analyst access or above.
* The conversation was started by an automation, or continues an earlier turn.
* The first message gives Orion nothing to match on.

A claim is added only if Observatory has verified it. A claim checked against data needs high or medium confidence. A claim your team approved without a check can carry any confidence, including low and hearsay.

### Seeing the claims behind an answer

The reply where the claims were first added shows a **Source** or **N Sources** pill under it. That is the first reply, or the reply after a successful retry. Later answers that reuse the same claims do not list them again. Click the pill to open the **Context** tab of the [Session drawer](https://docs.runorion.com/chat/basics-navigation.md#session) for that reply. The claims are listed in an **Observatory** group, each marked **Conversation snapshot**. Click a claim to open its card:

* The claim statement.
* A badge that reads **Checked against data**, **Team approved, not checked**, or **Said by a teammate, not checked**, plus the claim's confidence. A claim said by a teammate shows no confidence.
* The dates the evidence and the verification are from.
* **Metrics**: a tile for each metric the claim cites, with its latest value, an **As of** date, and an **Open** link to the metric. A metric with no result yet reads **No result yet**.
* **What Orion noticed**: the facts behind the claim, each with where it came from. A fact that has been disabled is struck through and labeled **Disabled**.
* **Also cites**: other sources, such as Context pages, with an **Open** link where the page can be opened. If the evidence cannot be loaded, this list is titled **Sources** and holds every source the claim cites.

If you think a claim is wrong, click **This is wrong** under the card and say why. Orion shows what you wrote beside the claim under **People said this is wrong**. Use **Undo** to take back your own note.

The snapshot stays with the conversation. Turning **Observatory context** off later keeps the claims visible in the drawer but stops Orion using them in later replies. Turning it on again restores the same snapshot. It does not add claims to a conversation that started without them.

If the lookup does not finish, the drawer shows **Observatory evidence lookup did not finish**. Orion answers without the evidence. It tries once more on your next message. If that also fails, the item reads **Not loaded for this conversation**, and the conversation continues without evidence.

## Where else cards appear

Orion also brings open questions and suggestions to the project panel, so you do not have to wait for the right conversation.

### Project panel

An analyst or admin sees an **Orion needs your input** section in the project panel. While Orion has open items, the section shows their number and lists them, each row labeled **Question** or **Suggested analysis**. Click a row to open a conversation about it, or click the X at the end of the row to mark it **Not relevant**; on a wide screen the X appears when you point at the row. The panel lists at most three items. The list does not count unexplained moves. When there are no items, the section keeps the **Orion needs your input** title and says how many unexplained moves are on the timeline below, or, if there are none, reads **You’re all caught up** with **No questions need your input right now.**

The section also holds the project Timeline.

### Timeline

The last row of the section opens **What happened**, a timeline of dated events Orion has learned about your metrics. When the timeline spans more than 30 days, a **Time range** control sets how far back it looks. A project's Timeline lists only the questions about metrics that project owns. A question about a metric no project owns appears only on the [Observatory page](#the-observatory-page) in Settings.

* **Unexplained moves** come first. Each names a metric and how much it moved. Select one to answer with a quick choice (**A release or product change**, **An outage or incident**, **A campaign or promotion**, **A data or pipeline change**), to type a line in **Or say what happened in a line** and click **Answer**, or to click **Discuss in chat**. Orion adds your answer to the timeline in its next review.
* **Explained events** follow. Select one to see where it came from. **Open the conversation it came from** takes you to the chat. **Disable** stops Orion using the entry in chat and removes it from the Timeline, and **Enable** restores it.

## The Observatory page

Cards are the small part of what Orion learns. The **Observatory** page holds the rest. Every day Orion reviews the dashboards, metrics, and conversations in your workspace and in every group you can open. It checks what it can against your data and asks your team about the rest. A Tenant Admin or Gravity support staff member opens the page at **Settings → Observatory**.

> [!NOTE]
>
> The Observatory page is not available on Orion trial workspaces.

The **Group** menu at the top picks what you are looking at: a group, a project that is in no group, or the whole **Workspace**. The page shows when Orion last reviewed that scope, then three tabs. **Claims** opens first.

* **Claims**: what Orion believes about the scope. Click a claim to see **How Orion knows** it. A claim is **Checked against your data**, **Waiting for the next review**, **Unverified**, or **Didn't match the data yet**. Each row says **In chat** or **Not in chat** with the reason
* **Timeline**: the dated events Orion has learned about your metrics
* **Questions**: the **Needs your input** list, the questions only your team can answer. **Show all questions** also lists questions that are not ready for a project page

### Review claims

The **Claims** tab opens on **To review**, the claims your team has not approved. **Approved**, **All**, and **Retired** show the others, and you can search, filter by topic, or turn on **In chat only**.

* **Approve** records your team's sign-off. The row then reads **Team approved**
* **Reject** retires the claim, and **Restore** brings it back from **Retired**
* A claim expires when nobody verifies or approves it within 14 days. Orion never uses a retired or expired claim in chat

### Use in chat

Open a claim to find the **Use in chat** switch. Turn it off to keep a claim out of chat. Turn it on to release it again.

When weak evidence is the only thing keeping a claim out of chat, an analyst or admin can turn the switch on anyway. Orion asks you to confirm first, because chat then says your team approved the claim, not Orion's checks. Any other claim that is out of chat keeps the switch off, and hovering it gives the reason.

Only analysts and admins of the group can change a claim. Everyone else who opens the page sees the same controls turned off.

## Who sees what

* **Cards under an answer**: analysts and admins of the project, including people who hold that role through a group.
* **Observatory evidence in answers**: added automatically, only for people with analyst access or above, and listed on the **Context** tab of the Session drawer, which viewers do not see.
* **Project panel section**: analysts and admins of the project. Orion loads its items only for them.
* **Answering on a card under an answer**: any analyst or admin who can see the card. The same holds for **Investigate with Orion** and the side conversation it opens.
* **Answering from the project panel**: clicking a row opens a conversation, and Orion records your answer there only if you have analyst access to the project.
* **Answering from the Timeline**: the person the question was addressed to, or a Tenant Admin or Gravity support staff member who also holds analyst access on that scope. The Timeline lists an unexplained move only for someone with analyst access on its scope.
* **The Observatory page**: Tenant Admins and Gravity support staff. Changing a claim takes analyst or admin access to the group.
* **Unexplained moves on the Timeline**: analysts and admins of the project. A Tenant Admin or Gravity support staff member sees each one on the Timeline only, so the project panel does not repeat it. An analyst who is not a Tenant Admin can also see it in the project panel.

> [!TIP]
>
> Add the definitions your team keeps explaining to [Context](https://docs.runorion.com/knowledge-base/overview.md). Orion then has fewer reasons to ask.

## Next Steps

> [card] [Memory](https://docs.runorion.com/chat/memory.md)
>
> See what Orion remembers about you and how to change it

> [card] [Context](https://docs.runorion.com/knowledge-base/overview.md)
>
> Keep definitions and standard filters where the whole team can use them


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