---
title: "Glossary"
description: "Definitions for the observability, telemetry and reliability terms used across the writing."
doc_version: "1"
last_updated: "2026-07-29"
canonical: "https://btrindade.com/glossary"
---

# Glossary

Definitions for the observability, telemetry and reliability terms used across the writing.

### Observability

How well the internal state of a system can be explained from the data it already emits, without shipping new code to find out. Distinct from monitoring, which watches for conditions someone thought of in advance.

### Telemetry

The data a running system emits about itself. Conventionally grouped into three kinds: metrics, logs and traces.

### Metric

A numeric measurement aggregated over time, such as requests per second or error rate. Cheap to store and good for trends and alerting, but it cannot explain any single request.

### Log

A timestamped record of one discrete event. Structured logs carry key/value fields rather than a sentence of prose, which is what makes them searchable rather than merely readable.

### Trace

The path of a single request as it crosses services, assembled from spans. Answers where the time went, which neither metrics nor logs can do on their own.

### Span

One unit of work within a trace, carrying a start time, a duration and attributes. Spans nest inside one another to record what called what.

### Instrumentation

The code that emits telemetry. Automatic instrumentation comes from libraries and covers common frameworks; manual instrumentation is what gets added to describe domain-specific work.

### OpenTelemetry (OTel)

A CNCF project that standardises how telemetry is described, collected and transmitted, so that the choice of backend no longer dictates the choice of client library.

### OTLP (OpenTelemetry Protocol)

The wire format OpenTelemetry components use to send telemetry to each other and to backends.

### Collector

A standalone OpenTelemetry process that receives telemetry, transforms it and forwards it on. Its pipeline is built from receivers, processors and exporters, which means backends can be swapped, data filtered or sampling changed without redeploying the applications.

### Cardinality

The number of distinct values an attribute can take. High-cardinality attributes such as user IDs multiply the number of stored series, which is the usual reason an observability bill grows faster than traffic.

### Sampling

Keeping a subset of traces rather than all of them. Head sampling decides at the start of a request; tail sampling decides once the trace is complete, so the slow and failed ones can be kept in preference to the ordinary ones.

### SLI (Service Level Indicator)

A measurement of how a service is behaving, expressed as a proportion of good events. For example, the share of requests served in under 300ms.

### SLO (Service Level Objective)

The target set for an SLI over a window of time, such as 99.9% of requests under 300ms across 30 days. The distance between the target and 100% is the error budget, which is what makes it a decision-making tool rather than a dashboard.

### p99

The value that 99% of measurements fall below. A p99 latency of two seconds means one request in a hundred was slower than that, which averages hide completely.

### Eval

A repeatable test for output that is not deterministic, such as a model response. Scores quality against expectations instead of asserting equality, and fills the role unit tests cannot for AI features.

## Sitemap

- [Home](https://btrindade.com)
- [Writing](https://btrindade.com/writing)
- [Glossary](https://btrindade.com/glossary)
- [Colophon](https://btrindade.com/colophon)
- [Same DevOps, new gremlins](https://btrindade.com/writing/same-devops-new-gremlins)
- [Telemetry: the backbone of modern observability](https://btrindade.com/writing/telemetry-backbone-modern-observability)