---
id: reference.glossary
title: Glossary
summary: Look up Teammately capabilities, product destinations, objects, and execution terms.
kind: reference
product_area: reference
status: stable
updated: 2026-11-06
canonical: /docs/reference/glossary
---

# Glossary

## Definition

Use current product names when navigating or explaining work. Preserve issued identities and existing serialized field names in technical contracts.

## Fields, states, or lifecycle rules

### Capabilities

- **Lemon:** Teammately’s AI agent for alignment, working with experts and engineers across project assistance, elicitation, synthesis, and development.
- **Lemon Elicit:** Prepares expert Contributions and proposes standards for expert review.
- **Lemon Code:** Supports interactive Harness development through context-aware conversation and reviewable file changes.
- **Coverage Engineering:** Defines the input situations a Benchmark should represent.
- **Expert Contributions:** Scoped specialist work that captures judgment, reasoning, and decisions; Lemon Elicit prepares inquiry and develops findings for review.
- **Lemon Weave:** Constructs Cases, controlled alternatives, Worlds for configured environments, and targeted training material.
- **Trialground:** Runs saved Harness experiments in disposable sandbox computers and evaluates outputs and recorded trajectories against benchmark rubrics.
- **Lemon Goal:** Pursues a benchmark objective by investigating failures, developing competing Harness candidates, and testing them through Trialground.

### Project foundations

- **Project Agent Brief:** Stable project understanding edited in Project Context.
- **Reference Materials:** Connected sources and their indexed content.
- **Lemon instructions:** Current Markdown guidance for configured activities, with inheritance and automatic history.
- **Dimension / ontology value:** A reusable behavior axis and one of its declared values.
- **Topic / Topic Group:** A subject and an editable grouping that does not merge its members.
- **Scenario:** A reusable mechanism for constructing input situations; a Scenario can guide many concrete Cases.
- **Case:** One concrete task or situation, represented by canonical input and declared supporting materials.
- **World:** Immutable environment source snapshot, separate from an isolated runtime instance.
- **Persona:** The concept of a reusable simulated participant in Trialground, illustrated in the Assets preview marked Early reviewers only.
- **Harness:** Executable candidate logic with a mutable Draft and exact saved Versions.
- **Comparison Direction:** Guidance for meaningful response alternatives.
- **Review Screen:** A selectable expert presentation asset.
- **Regime:** The framework and settings that interpret rubric judgments as scores.

### Benchmark work

- **Coverage Design:** Canvas requirements and Inbox adoption work.
- **Coverage requirement:** An input situation with a minimum or distribution expectation, or an observation.
- **Inbox item:** An attributable input-coverage observation or proposed Case with a separate adoption decision.
- **Dataset Snapshot:** Frozen selected Cases and the recorded evaluation context.
- **Benchmark Version:** The immutable population and criterion boundary consumed by execution.
- **Contribution:** A scoped specialist request with accepted work and attributable decisions.
- **Run:** A recorded candidate execution cohort containing planned Trials.
- **Trial:** One execution of the candidate across the planned Cases.
- **Used for score:** The reference Run selected before outcomes were known.
- **Results:** The inspection/comparison surface for managed or imported Runs.
- **Lemon Goal Session:** Recorded objective, conversation, candidate work, and results for improving a saved Harness.
- **External Agents:** A copied scoped task for a coding agent in the customer's repository.
- **Weights training:** Dataset preparation, packaging, provider training and separately authorized evaluation.
- **Fieldloop log / Thread:** Captured application record and an inquiry grounded in selected sources.

## Decision checkpoint

A proposal is not an approval. Generated material is not adoption. A completed operation is not automatically a complete evaluation. Current settings are not the settings used by every historical Run.
## Related task pages

{% related-card-grid title="Related task pages" %}
- [Product Map](/docs/getting-oriented/product-map)
- [Object Model](/docs/object-model)
- [Benchmark Evaluations](/docs/benchmark-evaluations)
- [Coverage Design](/docs/coverage-design)
- [Fieldloop](/docs/fieldloop)
- [Weights Training](/docs/weights-training)
{% /related-card-grid %}
