> For the complete documentation index, see [llms.txt](https://docs.controltheory.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.controltheory.com/controltheory-documentation/dstl8-docs/incidents/forecasts.md).

# Forecasts

Incidents Möbius opens before anything fails, when a metric is heading for a limit

A forecast is an incident Möbius opens before anything has failed. When a metric is climbing steadily toward a hard limit, such as a container's memory toward its pod limit or a disk filling up, Möbius opens an incident while there's still time to act, instead of waiting for the out-of-memory kill or the full disk.

Forecasts live alongside your other incidents, marked so you can tell them apart, and they close themselves if the metric turns around.

## What Möbius forecasts

Möbius forecasts a metric only when it knows the metric's limit:

| Limit           | Applies to                                                                                                                                                                                                                  |
| --------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Full (100%)** | Utilization and ratio metrics, such as disk or filesystem utilization.                                                                                                                                                      |
| **Pod limit**   | Container memory on Kubernetes, when the container has a memory limit set and Dstl8 collects it from your [Kubernetes source](https://docs.controltheory.com/controltheory-documentation/dstl8-docs/sources/kubernetes.md). |

Metrics with no known limit, like request counts, latencies and counters, aren't forecast. Möbius still watches them for values outside their usual range; see [Incidents](https://docs.controltheory.com/controltheory-documentation/dstl8-docs/incidents.md).

A forecast incident is opened when the metric is expected to reach its limit within about the next six hours. Further out than that, you'll see it on the [Metrics page](https://docs.controltheory.com/controltheory-documentation/dstl8-docs/metrics/explore.md#capacity-horizon) under **Capacity horizon**, which lists everything heading for a limit, soonest first, without opening an incident.

Möbius is careful about what it calls a forecast:

* **It has to hold.** A trend must hold over several checks before Möbius opens a forecast, so a brief ramp, like a burst of traffic, doesn't raise one.
* **It needs history.** A new metric needs time for Möbius to learn its normal behavior before it's forecast.
* **Memory is judged by its low points,** not its peaks, so a leak stands out from a busy day. See [What a Forecast looks for](https://docs.controltheory.com/controltheory-documentation/dstl8-docs/metrics/insights.md#what-a-forecast-looks-for).

## Where you see forecasts

* **The Forecast badge.** Forecast incidents carry a **Forecast** badge in the incidents list, on workspace pages and on the incident page. Its tooltip reads *Opened on a forecast: nothing has failed yet*.
* **The Forecasts filter.** On the **Incidents** page, the filter that reads **Forecasts and failures** narrows the list to **Forecasts only** or **Failures only**.
* **The incident page.** A forecast's [Signals](https://docs.controltheory.com/controltheory-documentation/dstl8-docs/incidents.md#signals) section charts the metric with its limit line, its usual level and where the forecast opened, and the summary above the chart gives the value now, its peak and the **limit**.

A forecast opens as **Investigating**, usually with **Minor** impact. It can be Major when the limit is less than an hour away on something your users depend on. Its write-up describes the trend and the limit, rather than a countdown to the minute.

## How a forecast ends

A forecast resolves itself one of three ways, and its **Timeline** says which:

| What happened                                                | What Möbius does                                                                                                                          | Timeline note                                                 |
| ------------------------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------- |
| The metric reached its limit, or started behaving abnormally | The forecast came true. The incident becomes **Active**, and its badge changes to **Was a forecast** so you can see Möbius saw it coming. | *The forecast came true: … reached its limit.*                |
| The metric stopped heading for its limit                     | The incident is resolved automatically, after the trend has ended for a while.                                                            | *The forecast cleared: … is no longer heading for its limit.* |
| The metric stopped reporting                                 | The incident is resolved once Möbius can no longer follow it.                                                                             | *The forecast ended: it can no longer be followed …*          |

If you set a forecast to **Active** yourself, it's treated as a real problem from then on, and the badge changes to **Was a forecast**. If someone on your team changed the incident recently, Möbius leaves it for you rather than changing its status.

## Acting on a forecast

1. **Open the incident** and look at the metric's card under **Signals**. The chart shows how fast it's climbing, how close the limit is, and any deploys in the window.
2. **Check for a recent change.** A deploy marker near the start of the climb is a strong lead. The **Events** tab lists the deploys and configuration changes around the incident.
3. **Use Open in Metrics** to see the metric over a longer range, or compared across pods.
4. **Ask Möbius** in the chat sidebar, for example "what changed before memory started climbing?" or "is this a leak or more traffic?".
5. **Fix it or buy time.** Raise the limit, scale out, clean up the disk, or fix the leak. A restart resets memory and buys time, but if the leak is still there the climb starts again and Möbius can forecast it again.

Once the metric turns around, the forecast clears on its own; you don't need to close it.

## Notifications for forecasts

Forecasts go through the same [alert rules](https://docs.controltheory.com/controltheory-documentation/dstl8-docs/incidents/alerts-and-notifications.md) as other incidents. Their notifications are marked so you can tell them apart at a glance: the title starts with **Forecast:**, and they show when the limit will be reached. See [What notifications look like](https://docs.controltheory.com/controltheory-documentation/dstl8-docs/incidents/alerts-and-notifications.md#what-notifications-look-like).

Every alert rule has a **Forecasts** setting: **Forecasts and failures**, **Failures only** or **Forecasts only**. A common setup is two rules, one that sends **Failures only** to an urgent channel and one that sends **Forecasts only** to email or a quieter channel.

{% hint style="info" %}
The **Default incident rule** created with your account only sends Major and Critical incidents. Most forecasts open as Minor, so to hear about them, add a rule with **Forecasts** set to **Forecasts only**, **Statuses** including Investigating, and **Impact** left empty or including Minor.
{% endhint %}

A forecast that comes true becomes an ordinary incident. A **Failures only** rule with **Trigger Mode** set to **Status Change** or **Any** picks it up at that point; one set to **Creation** doesn't, because the incident was created earlier as a forecast.

## Tips and troubleshooting

* **"Why open an incident when nothing is broken?"** That's the point of a forecast: it gives you time to act before your users notice. If you'd rather not be notified, adjust your alert rules' **Forecasts** setting; forecasts still appear in the incidents list.
* **"Why did it resolve on its own?"** The metric stopped heading for its limit, or stopped reporting. The **Timeline** says which.
* **"This metric always runs close to its limit by design."** Some workloads, like caches, are meant to fill up. Close the forecast as **Not an issue**, explain why in the note, and tick **Remember this as expected noise** so Möbius keeps it in mind. See [Closing an incident](https://docs.controltheory.com/controltheory-documentation/dstl8-docs/incidents.md#closing-an-incident).
* **"Why wasn't my metric forecast?"** It may have no known limit (see the table above), it may still be new, or its climb may not have held long enough. The **Capacity horizon** list on the [Metrics page](https://docs.controltheory.com/controltheory-documentation/dstl8-docs/metrics/explore.md#capacity-horizon) shows what Möbius currently sees heading for a limit.
* **Set memory limits on your containers.** Without a memory limit, there's nothing for container memory to be forecast against.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.controltheory.com/controltheory-documentation/dstl8-docs/incidents/forecasts.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
