# Endgame Wizard Context Setting Documentation

## Contexts and its Components

_Context_ refers to structured input as it is used by a GPT model to interpret and answer the questions posed to it. Contexts are made up of indexes, which in turn are made up of the following components:

- **Category**: The name of the index.
- **Tags**: Labels that further specify documents under a given index.
- **Documents**: The raw input in the form of text.

Both categories and tags are used by the model to identify the appropriate source document to resolve a given query.

Contexts must be *built* (roughly, compiled) before they can be used by the model.

## Example

- Primary colors (**index**)
- red.md (**document**)
- _{vermillion, crimson, ruby... (**tags**)}
- green.md (**document**)
- _{tags: emerald, mint, jade... (**tags**)}
- blue.md (**document**)
- _{tags: azure, teal, sky... (**tags**)}
- Minor chords (**index**)
- c-minor.md (**document**)
- _{Cm, Cmin, C-... (**tags**)}
- ...

## Considerations

- Modifications to the context are not incremental---i.e., building a new context overwrites the previous context completely, including entities that have not been changed. Given that tokens are spent when building a context, contexts should be fully worked out before building them to minimize token spending.
- Tags should not be common across multiple documents---i.e., each document must be labeled with unique tags.
- Tags are (not?) case-sensitive.
- Similarly, documents must have unique names.