> 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/improve-your-results/choose-good-examples.md).

# Choose good examples

Pick labels that teach Earth Index what you want, using variety, edge cases, spread and a simple label-run-check loop.

A few well-chosen labels beat a large pile of similar ones. This page shows how to pick labels that teach Earth Index the full range of what you are looking for.

## The short version

* Variety beats volume. Label different soils, light, seasons and shapes of your target.
* Spread labels across the whole area, and keep one project for each distinct look.
* Work in a loop: label, run, check the top results, fix the worst mistakes, run again.

## Vary your examples

Earth Index finds tiles that are near your examples. If all your examples look the same, it finds only that one look and misses the rest.

* Pick examples that differ in ground colour, light, season and shape.
* Include edge cases that you want found, such as a half-overgrown road or a small clearing.
* Do not label ten tiles from one spot. One tile from each of ten places teaches more.

## Spread labels and split projects

Label in several parts of your area of interest. If all labels sit in one corner, all results tend to come from that corner.

If your target has distinct looks, make one project for each look. Single logging roads and tangled logging roads, or mines and small river diggings, behave differently and give cleaner results as separate projects. See [Plan your search](https://docs.earthindex.ai/search/plan-your-search.md).

## Work in a loop

1. Label a handful of positive tiles, and at least one negative.
2. Run a search.
3. Check the top results. Count how many are right.
4. Mark the worst mistakes as negatives and add a few positives that were missed.
5. Run again.

A short loop with a few good corrections works better than one long labelling session. See [Why negatives matter](https://docs.earthindex.ai/improve-your-results/why-negatives-matter.md) for which mistakes to fix first.

## How many labels you need

The app sets a minimum for each kind of search:

* **Quick Search:** 1 positive.
* **Deep Search:** 10 positives and 1 negative.
* **Monthly Search:** 10 positives and 100 negatives.

In Monthly Search, the label meter shows a positive guide of 10, 50 and 100. These are minimums, not targets. More varied labels usually help more than more labels of the same look. For the full list of limits, see [Limits and quotas](https://docs.earthindex.ai/reference/limits-and-quotas.md).

{% hint style="success" %}
**What to do**

Start with a few varied positives and one or two look-alikes as negatives. Run a Quick Search, fix the worst results, and repeat until the top results are right.
{% endhint %}

<details>

<summary>Go deeper</summary>

Variety matters because Earth Index compares tiles by their embeddings, lists of 384 numbers. Examples that look alike sit close together in that space. Examples that differ widen the area that counts as a match. See [How Earth Index finds matches](https://docs.earthindex.ai/improve-your-results/how-search-works.md).

</details>

## Related

* [Plan your search](https://docs.earthindex.ai/search/plan-your-search.md)
* [Label tiles](https://docs.earthindex.ai/search/label-tiles.md)
* [Fix poor results](https://docs.earthindex.ai/improve-your-results/fix-bad-results.md)
