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

# Targets

The primary payout target, the auxiliary targets, and their values.

In the tournament dataset the primary payout target is `target_everest`, and every tournament submission is scored on it. The auxiliary targets exist for ensembling and diversification; you never submit against them directly.

{% hint style="info" %}
**The graded column is named by the dataset, not by the platform.** This page describes the tournament dataset. An event or hackathon can be served a different one, whose target panel carries different names and whose graded column is therefore a different column. Call `get_dataset_schema` and read `primary_target`: that field is the column your submission is graded on, for whichever tree you were actually served. See [Events & hackathons](/events-and-hackathons.md).
{% endhint %}

## Primary target

`target_everest` is a 20-day forward-return target, rank-based and binned into five levels. The levels are not equal-sized: a bell-shaped split puts the middle half of each exped in the centre level and only a small tail in each extreme. The stored values in the served file are:

| Value  | Meaning                |
| ------ | ---------------------- |
| `0.0`  | Bottom 5% of the exped |
| `0.25` | Next 20%               |
| `0.5`  | Middle 50%             |
| `0.75` | Next 20%               |
| `1.0`  | Top 5% of the exped    |

Atlas publishes no `target` alias. Use the column name `target_everest`.

## Auxiliary targets

The dataset also ships auxiliary targets named `target_<Name>`, where `<Name>` is a codename. They are related forward-looking constructions at several horizons. The codename carries neither the horizon nor the construction, so do not infer either from the name.

The full list of auxiliary targets is in `eiq_features.json`. They are useful for ensembling: train separate models on different targets and blend the predictions. Some are near-duplicates that add little together:

| Pair                                   | Typical correlation |
| -------------------------------------- | ------------------- |
| `target_everest` vs `target_Tougroute` | \~0.84              |
| `target_Anghomar` vs `target_Bougafer` | \~0.84              |

No pair is negatively correlated. The least correlated pair is approximately 0.11, and the auxiliary target least correlated with `target_everest` sits near 0.19. Figures are measured on the `train` split.

You submit one prediction per instrument, and it is scored on the primary target only (`target_everest` here, `primary_target` on whichever dataset you were served). The auxiliary targets are a research tool only.

## What the values mean

The binning is computed per exped (cross-section), so a `1.0` means "in the top 5% of that day's cross-section". In the tournament dataset the levels follow the bell-shaped split above. Another dataset can declare a different distribution across the same five levels, so where the exact shape matters, read `target_encoding` from `get_dataset_schema` rather than assuming this one.

## Labels overlap across adjacent expeds

Because the target is forward-looking over 20 trading sessions, the labels of adjacent expeds overlap. This is why the evaluation protocol requires a gap between your last fit exped and your first hold-out exped: without a gap, your hold-out label is partly determined by information that was available in your fit set. See the [research guide](/research-guide/evaluation-protocol.md) for the correct protocol.

## Targets outside the train split

In `validation` and `live`, every target column is present in the file but every value is `NaN` (blanked). The server scores your predictions against a held-out answer key. You cannot self-score on these splits because the targets are absent.

A `NaN` target in the `train` split means the forward return was uncomputable for that row. Drop those rows when training on that target.


---

# 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 dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.everesteer.ai/data/targets.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

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.