---
id: quickstart.product
title: Product Quickstart
summary: Begin with an expert question, coverage need, Harness, or training objective and connect the result to your next development decision.
kind: quickstart
product_area: introduction
status: stable
updated: 2026-11-06
canonical: /docs/quickstart
---

# Product Quickstart

Begin with one useful assignment. You might need to understand an expert’s judgment, test an existing Harness, represent a missing situation, or prepare material for training.

As the work develops, connect its findings to shared standards and a benchmark your team can interpret.

## Decision checkpoint

| Your starting point | Open | First useful outcome |
| --- | --- | --- |
| An expert question | Expert Contributions | A reviewed distinction or proposed standard |
| Missing situations | Coverage Facets and Coverage Design | A deliberate set of Cases |
| An existing Harness | Assets → Harnesses | A saved candidate ready for a trial |
| A candidate executed in your own environment | Benchmark Evaluations → Runs | Imported responses assessed against the selected Benchmark Version |
| A measured behavior to improve | Harness improvement | A Goal or scoped external-agent task |
| A training objective | Weights training | A dataset guided by criteria and coverage |
| Production interactions | Fieldloop | An investigated Thread and focused next action |

## Prerequisites

Choose a Project and identify the people who can settle the relevant judgment or engineering decision. Bring the context you already have: representative inputs, reference material, an implementation, or a concrete learning objective.

## Before and after

| Before | After |
| --- | --- |
| A specialist’s judgment is implicit | Reviewed criteria and conditions |
| Important situations are scattered | A deliberate benchmark Case mix |
| A candidate seems promising | Recorded trials and interpretable results |
| A recurring failure lacks teaching material | Targeted examples and separate assessment |
| Follow-up work is broad | A focused expert question or engineering objective |

## Establish shared context

1. Write the **Project Agent Brief** in Project Context and connect **Reference Materials**. Inspect the indexed reference blocks Lemon can use.
2. Configure **Project Input Schema** for the input architecture and required materials.
3. Identify reusable Dimensions, Topics, and Scenarios as the coverage needs become clear.

## Complete one connected loop

1. Import representative Cases or use **Coverage Design → Canvas** to construct the situations the Benchmark should represent.
2. Request a focused **Expert Contribution** where expected behavior needs clarification. Review the resulting policies, rubrics, and coverage findings.
3. Inspect the Case population and applicable standards, then create a Dataset Snapshot and Benchmark Version.
4. Choose the execution path in **Benchmark Evaluations → Runs**. For managed execution, publish the intended Harness Version and select the candidate, machine, and Trial count. For execution in your own environment, obtain the exact Case manifest, execute the candidate, and [import its responses or recorded failures](/docs/benchmark-evaluations/output-mapping). Confirm completion separately from evaluation progress.
5. Inspect **Results** and **Dashboard**. Follow important movement into the relevant Cases, rubric judgments, gates, and point scores.
6. Choose the next action: a Lemon Code change, a Lemon Goal experiment, an External Agents task, a coverage update, or targeted training material.

## Object and state changes

Project settings and reusable Assets support several Benchmarks. A Snapshot and Benchmark Version preserve the Cases and criteria used for evaluation. Managed Runs identify the saved Harness Version; imported Runs preserve the submitted responses and external candidate context. Both retain their recorded evaluation conditions. Contributions and development work retain their own histories.

## Success criteria

Your team can explain the intended behavior, the situations tested, the specialist standard applied, and what the candidate did. The next development decision follows from those findings.

## Common failure modes

- Reviewing a score without its Case population or applicable criteria.
- Comparing candidates with different experiment conditions.
- Submitting responses against different Case IDs from those in the selected Benchmark Version.
- Applying a proposed standard before its expert review.
- Using teaching examples as the only assessment of a trained candidate.

## Related reference pages

{% related-card-grid title="Continue your assignment" %}
- [Product Map](/docs/getting-oriented/product-map)
- [Working with Lemon](/docs/concepts/lemon)
- [Run a Benchmark Evaluation](/docs/benchmark-evaluations/run-evaluation)
- [Import External Evaluation Outputs](/docs/benchmark-evaluations/output-mapping)
- [Weights Training](/docs/weights-training)
{% /related-card-grid %}

## Related troubleshooting pages

{% related-card-grid title="Resolve an interruption" %}
- [Permissions](/docs/troubleshooting/permissions)
- [Benchmark Runs](/docs/troubleshooting/benchmark-runs)
{% /related-card-grid %}
