> 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/splits-and-obfuscation.md).

# Splits & obfuscation

Train, validation, live: what each holds, what is blank, and how the data is obfuscated.

One obfuscated feature matrix with rank-based targets, split three ways. Every row carries a `data_type` column whose value is `train`, `validation` or `live`, so you can confirm which split a row belongs to without relying on the filename.

| Split        | Targets     | What it is for                                                                                                                            |
| ------------ | ----------- | ----------------------------------------------------------------------------------------------------------------------------------------- |
| `train`      | **Labeled** | Full history with targets. Fit your model here. Carve a hold-out from it to score yourself offline before spending a submission.          |
| `validation` | Blanked     | The practice board: same features, targets withheld. Scored server-side. Use this to tune your pipeline before the first round opens.     |
| `live`       | Blanked     | The currently open round. Same features, targets withheld. The instrument id namespace is new each round: re-download `live` every round. |

Your predictions on the two blanked splits are what the boards score, server-side, against an answer key you never receive.

{% hint style="danger" %}
Predicting on the wrong one of `validation` / `live` gives a frame whose ids do not join the open round. `download_dataset(split="live")` serves whichever round is currently open; `split="validation"` serves the practice board.
{% endhint %}

During an event, the `train` split is the labeled training set for offline tuning. The scored split (validation for the practice board, live for each sealed round) ships with target columns present but every value is NaN. The server holds the answer key separately.

### Hold-out evaluation

Take a hold-out from `train` only, not from `validation`. The validation split is the practice board: the server scores it, and you cannot self-score it because the targets are absent. Carve a labelled block of later expeds from `train`, leave a gap at least as long as the primary target horizon between your last fit exped and your first hold-out exped, so no hold-out label overlaps the fit window. In-sample fit is never rewarded.

See the [research guide](/research-guide/evaluation-protocol.md) for the full protocol.

## Obfuscation

{% stepper %}
{% step %}

### Rank

Every feature is ranked across the cross-section of its exped. The rank transformation makes features unitless and comparable across expeds.
{% endstep %}

{% step %}

### Bin

The ranks are binned into ten equal-count levels. The stored values are `0` to `9` as `int8`. A value of `-1` means the source data was unavailable for that instrument on that day: treat it as missing, never as an ordinal value below 0.
{% endstep %}

{% step %}

### Rename

The original feature name and the original value are not in the file. Raw source data is never exposed through the API or the dataset.
{% endstep %}
{% endstepper %}

### Target values

Targets are also rank-based and binned into five levels, on a bell-shaped split: 5%, 20%, 50%, 20% and 5% of each exped. The stored values are `0.0`, `0.25`, `0.5`, `0.75` and `1.0` as `float32`. A target of `NaN` means the forward return was uncomputable for that row: drop those rows when training on that target. The target columns are described in the [targets](/data/targets.md) page.


---

# 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/splits-and-obfuscation.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.