This glossary defines the main terms used across RoleThread Lite.
RoleThread Lite
The local dataset app for creating, organizing, validating, editing, merging, analyzing, and exporting narrative training datasets.
Dataset
A collection of training entries, usually stored as a JSONL file.
JSONL
JSON Lines. A file format where each line is one JSON object. Many training datasets use this format.
ChatML
A message-based dataset shape built around roles such as system, user, and assistant.
ShareGPT
A conversation dataset format used by some tools. RoleThread can import and export ShareGPT-style data.
Entry
One training example in a dataset. In RoleThread, an entry usually contains a system prompt and one or more user/assistant exchanges.
Exchange
A user message paired with an assistant response.
System Prompt
The instruction or setup message at the beginning of an entry.
System Prompt Template
A reusable saved system prompt that can be loaded into Create Entry or Full Edit, then customized before saving.
Tag
Metadata used to organize entries. Tags help with filtering, search, export, review, and Insights.
Category
A group for related tags.
Built-In Tag
A RoleThread-provided tag that is locked to preserve consistent meaning across datasets.
Custom Tag
A user-created tag that can be edited, renamed, or deleted.
Archived/Imported Tag
An unknown or inactive tag preserved from imported data or older metadata. It is kept for review instead of being silently discarded.
Alias
A remembered relationship between an older tag name and its current meaning.
Sidecar
A .registry.json companion file that stores portable metadata beside a dataset.
Working Copy
A protected local copy of an untrusted dataset that RoleThread can safely edit without changing the original source file.
Trusted Dataset
A dataset RoleThread recognizes as carrying stable RoleThread identity metadata, such as dataset UUID and entry UUID information.
Untrusted Dataset
A dataset that does not clearly carry RoleThread identity metadata. It may still be valid data, but RoleThread treats it carefully before writing to it.
Clean Export
An export mode that removes RoleThread metadata from training records.
Group Chat Mode
An entry editing mode that lets you assign characters to individual turns while keeping exported roles as standard system, user, and assistant.
Character
A named person or persona stored in the Character registry for previews, Group Chat mode, and metadata.
Character Mapping
Metadata that connects a character to a specific message turn in an entry.
Dataset UUID
A stable identifier for a RoleThread-native dataset.
Entry UUID
A stable identifier for a specific entry.
Validation
The process of checking entries for structural issues, cleanup concerns, and quality-related diagnostics.
Repair
A safe automatic or manual fix for a validation issue.
Insights
The dataset quality and analysis page that reports scores, distributions, recommendations, and focused review links.
Cloud Sync
Optional batch backup mirroring to a configured cloud sync folder. It is not live collaboration.
Local Backup
A backup stored on the local machine before protected operations.
DB Backup
A backup of the local database that stores tags, characters, prompt templates, settings, and related metadata.
Portable Metadata
Metadata that can travel with a dataset, usually through a sidecar, without being inserted into clean training records.