> ## Documentation Index
> Fetch the complete documentation index at: https://trygradient.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Deliverables

> Output formats candidates produce during assessments

# Deliverables

A **deliverable** is the artifact a candidate produces during an assessment. It's the primary output that gets scored for quality, structure, and content accuracy.

## Supported formats

| Format           | Extension | Typical use case                                               |
| ---------------- | --------- | -------------------------------------------------------------- |
| **Presentation** | `.pptx`   | Strategy decks, product reviews, executive briefings           |
| **Document**     | `.docx`   | Briefs, proposals, reports, analysis write-ups                 |
| **Email**        | n/a       | Inbox triage, stakeholder communication, prioritization        |
| **AI Workflow**  | n/a       | A reusable prompt or skill the candidate inherits and improves |

The deliverable type is set when building an assessment and determines which editing tools are available to the candidate in the workspace.

## How candidates build deliverables

Candidates work with the AI assistant to create their deliverable. The AI can:

* **Create and edit slides/sections** - Add content, restructure, format
* **Apply templates and formatting** - Professional layouts and styling
* **Insert data from connected sources** - Pull in tables, charts, and quotes
* **Iterate based on feedback** - Refine content through conversation

The workspace shows the deliverable in real-time on the right side of the screen, while the AI chat panel occupies the left side. Candidates can also manually edit the deliverable directly.

## Scoring deliverables

The deliverable is scored by two categories in the [rubric](/concepts/scoring):

* **Correctness** (deterministic): Are the required elements present, the facts right, and the judgment calls sound?
* **Deliverable Quality** (AI-judged): How good is the output (polish, clarity, insight), measured as how far it improves on a first-pass AI draft? A lightly reworded AI draft scores low; a meaningful, quality-raising departure scores high.

The scoring engine examines both the final deliverable and the process of creating it (visible through session events).