> ## Documentation Index
> Fetch the complete documentation index at: https://langwatch.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# What counts as a billable event

> The complete list of events LangWatch Cloud bills for, what is never billed, and how deduplication works

LangWatch Cloud bills on **ingested events per billing period**. This page is the complete list of what produces one, so you can predict a bill from your own workload rather than from a sample.

<Note>
  Self-hosted deployments are not metered and send no usage data to LangWatch. This page applies to LangWatch Cloud. See [Editions and licensing](/docs/self-hosting/licensing).
</Note>

## The complete list

These are the only events that increment your count. Everything else the platform does is free.

| What you did                                                                                | Billable event                       | Counted                   |
| ------------------------------------------------------------------------------------------- | ------------------------------------ | ------------------------- |
| Your app sent a span (an LLM call, tool call, retrieval step, or any other step in a trace) | `lw.obs.trace.span_received`         | One per span              |
| An evaluation produced a result                                                             | `lw.evaluation.reported`             | One per evaluation result |
| An experiment run started                                                                   | `lw.experiment_run.started`          | One per run               |
| An experiment recorded a target's output                                                    | `lw.experiment_run.target_result`    | One per target result     |
| An experiment recorded an evaluator's score                                                 | `lw.experiment_run.evaluator_result` | One per evaluator result  |
| A scenario simulation started                                                               | `lw.simulation_run.started`          | One per run               |
| A simulation recorded a conversation snapshot                                               | `lw.simulation_run.message_snapshot` | One per snapshot          |

Spans dominate the count for most customers. A single trace of an agent turn typically contains several spans, one per step.

### Reading an experiment's cost

An experiment is not one event. A run over 100 dataset rows with 2 evaluators produces roughly `1 started + 100 target results + 200 evaluator results`, so around 301 events. This is worth knowing before a large sweep.

### Reading a simulation's cost

A scenario run produces one `started` event plus one snapshot per conversation turn, so a 10-turn simulation is about 11 events.

## What is never billed

* Reading your data: dashboards, the trace explorer, analytics queries, and search.
* Anything you create: prompts, evaluators, scenarios, agents, datasets, dashboards, custom graphs, automations, projects, or teams.
* Users. Seats are priced separately from events, not metered as events.
* Retries and replays of an event you already sent (see deduplication below).
* Platform-internal work: projections, rollups, queue processing, and background jobs.

## Deduplication

Events carry an idempotency key, and the meter counts **distinct keys** per billing period. A retry from your SDK, a redelivery inside our queues, or a replay of a projection all collapse to the single unit you were already charged for. You are billed for what you sent, not for how many times it moved through the system.

## Checking your own usage

Your current billing-period count is on **Settings → Usage** in the platform. It reflects the same deduplicated count the invoice is built from.

## Retention

The event count and how long events are kept are separate levers. Each plan includes a default retention window; extending it is available on Enterprise plans or with a self-hosted license. See [Pricing](/docs/pricing#custom-data-retention).