> For the complete documentation index, see [llms.txt](https://docs.earthindex.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.earthindex.ai/monthly-search-beta/overview.md).

# Overview

What Monthly Search (monitoring) is, how the detector and monthly runs fit together, and why access is currently by partner arrangement.

{% hint style="warning" %}
Monthly Search is in beta.
{% endhint %}

Monthly Search (monitoring) is a project type for watching one area over time. You build a detector from examples you label, test it, then run it on any complete month of satellite imagery and compare months to see what is changing.

## Who can use it today

Monthly Search is partner access, not self-serve. Earth Genome runs it as a supported program with partner organizations. On the **New Project** page it appears in the **Project type** list next to **Standard Search**. You can select it only when both of these are true:

* Your account has Full Access, not Open access.
* The Earth Index team has turned Monthly Search on for your account.

Requesting Full Access on its own does not turn Monthly Search on. If you cannot select it, the option is disabled for one of two reasons:

* With Open access, hovering the option shows a tooltip about requesting Full Access.
* With Full Access but without Monthly Search turned on, the option carries a **Partner access** badge. Hover it to read **Why partner access?**.

To discuss access, email <info@earthgenome.org>. See also [Create an account](/get-started/account.md) and [Start a project](/get-started/starting-a-new-project.md).

## How it works

A Monthly Search project has two tabs at the top of the sidebar:

| Tab                  | What you do there                                                                                                                                              |
| -------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Build a detector** | Three steps: **Label** (mark examples on the map), **Train** (the app trains the detector from your labels), **Test** (you review a sample of its detections). |
| **Monthly Search**   | Choose a month, run the detector on it, view the detections on the map, compare months, and optionally turn on automatic runs.                                 |

The **Monthly Search** tab opens when the detector leaves the **Test** step: after you pass the test, or after you skip or bypass it.

A positive label is a confirmed example of what you are looking for. A negative label is a square that does not include it. The detector learns from both. It is trained on the project's annual imagery, then scores the monthly imagery you run it on. The map is divided into squares (tiles), and a detection is a tile the detector thinks matches your examples.

You can run any complete calendar month from January 2024 up to the latest month that has ended. The current month is not available until it ends. Each month is processed on its own, and its results stay in the project until you delete them.

## Confidence and sensitivity

Every detection has a confidence score, the detector's estimate of how likely the tile is a true match. Sensitivity sets the lowest score that counts as a detection:

| Setting                | Minimum confidence       |
| ---------------------- | ------------------------ |
| **Broad**              | 70%                      |
| **Balanced** (default) | 85%                      |
| **Precise**            | 95%                      |
| **Custom**             | Any value from 1% to 99% |

Higher values return fewer, more confident detections.

## Keep in mind

* Label before you train. The app asks for at least 10 positive and 100 negative labels. The best negatives look like your target but are not what you want to find.
* The test is a quality check, not a guarantee. The detector passes when at least 60% of the sample detections you review are correct.
* Detections are candidates. The app does not ask you to validate every detection of a monthly run.
* Seasons move the numbers. Season, cloud and haze change the imagery, so counts can change without real change on the ground. Compare months from a similar time of year where you can.
* It is a beta. Screens, limits and behavior can change.

## Related

* [Build a detector](/monthly-search-beta/set-up-monthly-search.md)
* [Run and compare months](/monthly-search-beta/monthly-monitor.md)
* [Auto-runs and limits](/monthly-search-beta/notifications-and-limits.md)