Glossary

This page defines the words the documentation uses. Each entry is short and links to the page that explains the idea in full. Terms are in alphabetical order.

Term	Meaning	Learn more
Alert rule	A rule that watches one metric of a project, such as error rate, latency, volume or cost, and notifies channels when the metric crosses a line. A rule is a threshold rule or an anomaly rule.	Alert rules and channels
API key	A project API key. It belongs to one project and starts with spl_live_. Your application sends traces with it. It is shown once, when you create it.	API keys and tokens, Tenancy and access
Budget	A spend limit on a scope (the project, a gateway key, an end user or a model) over a UTC day or month. It notifies channels when the spend passes the amount, and a budget set to block also makes the gateway refuse further calls in that scope.	Budgets and the weekly digest
Dataset	A named list of test cases. Each item has an input, an optional expected_output and some metadata. You replay a dataset through a prompt in an experiment.	Datasets
Detector	A fixed rule that reads the spans of a time window and reports a finding when a pattern appears. The Doctor runs fifteen of them. No language model decides whether something is a problem.	The Doctor, How the Doctor works, Detector catalogue
Environment	A label on a trace that says where the code ran, such as production or staging. You choose the value. It is not an object you create first.	The data model
Evaluator	Something that scores traces for you. It is a rule (heuristic), an LLM judge or a webhook to your own endpoint. It runs online on a sample of traces and inside experiments.	Evaluators
Experiment	A comparison that joins a dataset, a target (a prompt version or typed messages) and some evaluators. Each run sends every dataset item through the target, scores the answers and records the cost.	Experiments
Gateway route	The part of the gateway that decides which provider credentials serve a call, how failures are retried and when to fall back to another target. Each gateway key uses one route.	Routes, retries and fallbacks
Insight	A finding of the Doctor: a named problem with the evidence for it, the layer where it lives, a suggested fix and a way to check the fix. An insight is open, acknowledged, muted or resolved.	The Doctor, How the Doctor works
Kind	The type of a span: llm, tool, retrieval, chain, http or other. The default is other.	The data model
Organization	The account that owns projects and has members. Each member has one role: owner, admin, member or viewer. Also called an org.	The data model, Tenancy and access, Roles and permissions
Personal access token (PAT)	A credential that belongs to a person and starts with spl_pat_. A script uses it to call the dashboard API as that person. It has a read or write scope.	API keys and tokens
Project	One application, or one slice of one, whose traces you read together. A project is the unit of data isolation. It also holds settings such as the retention period.	The data model, Tenancy and access
Prompt	A named template stored in Spanlight as immutable numbered versions. Labels such as production point at the version your application should use.	Prompts
Release	A label on a trace that names the version of your application that produced it. It lets you compare two versions.	Releases and users
Scope	What a credential may do. A project API key has scopes such as ingest:write, traces:read, scores:write and prompts:read. A personal access token has a read or write scope.	API keys and tokens, Roles and permissions
Score	A judgement about a trace, or about one span in it, with a name, a value and a source. You can filter and chart scores.	Scores and review queues
Session	Every trace that shares one session_id, such as a conversation. It is a grouping in the dashboard and is unrelated to the sign-in sessions of the people who use Spanlight.	The data model, Traces and sessions
Span	One unit of work with a start and an end, such as a model call, a tool call or a retrieval step. A span points to its parent, which is how the trace page draws the tree.	The data model, How ingestion works
Trace	The set of spans that belong to one request or one job. A trace exists because spans name it, and its totals are worked out from its stored spans.	The data model, Traces and sessions, The Overview dashboard
User	In observability data, the user_id you send on a trace to name the end user it was produced for. Spanlight stores it as you send it. It is not a Spanlight account.	The data model, Releases and users

Related pages

The data modelHow organizations, projects, traces and spans fit together.

Tenancy and accessRoles, credentials and how projects stay separate.

Limits referenceEvery size, count, time and rate limit.

Roles and permissionsWho can do what, and which key scopes exist.