DebugBundle

Reducing Ingested Events

How to lower raw ingested event volume without losing useful DebugBundle signal.

Raw ingested events are the accepted events that count against your account's shared monthly event allowance. The billing breakdown shows which projects are contributing the most volume so you can tune the highest-impact capture paths first.

Use this page when the billing allowance page shows high raw event volume, especially when the volume is coming from known noise rather than product failures.

Start With The Highest-Volume Project

Open the raw ingested events breakdown from the billing page and sort by project volume. Tune one project at a time so you can see whether each change lowers accepted events without hiding real failures.

For each high-volume project, check:

  • whether request telemetry is set broader than needed
  • whether verbose logs are being captured in production
  • whether browser resource errors are coming from third-party scripts
  • whether bot or scanner traffic is producing repeated low-value browser events
  • whether expected retries, health checks, or background jobs are producing redundant events

Use The Broadest Safe Product Control

Prefer product-level controls before SDK-local code changes because they are visible, auditable, and consistent across API, CLI, MCP, and dashboard workflows.

GoalRecommended controlEffect on event volume
Lower general verbositySwitch the project capture policy from investigative to balanced or minimalReduces which event categories are accepted
Stop known noise entirelyAdd a project capture rule with dropDiscards matching events before persistence when enforceable
Keep a representative sampleAdd a project capture rule with sampleKeeps only a deterministic fraction of matching events
Keep context but stop incident churnAdd a project capture rule with demotePrevents matching events from opening or reopening incidents, but kept events can still count as accepted volume
Stop broad request telemetrySet request capture to failures_only or narrower filtersReduces accepted request events
Stop standalone frontend noiseKeep breadcrumbs exception-only instead of standalonePreserves context on exceptions without sending every breadcrumb independently

For exact preset and rule behavior, see Capture Policy.

Common Reductions

Request Events

Request capture is often the largest raw event source. In production, most projects should avoid capture_request_events: "all" unless they are actively investigating a live issue.

Use failures_only for normal operation, then add path-scoped client-error incident rules for critical routes that should never return selected 4xx responses.

Logs

Production log capture should usually start at warning or error. If info logs are useful during an incident, use a temporary investigative window instead of leaving them enabled indefinitely.

Browser Resource Errors

Third-party script failures, blocked analytics calls, stale extension errors, and bot-only browser noise can create high volume with little debugging value. Confirm the pattern from an incident or bundle, then use a narrow capture rule keyed by host, browser event kind, first_party, client_kind, or bot_family.

Remote Probes

Remote probe activations are useful for short investigations. Keep activation TTLs short, deactivate probes after the incident, and avoid standalone probe capture for labels that fire on every request unless the investigation needs that level of detail.

When To Use SDK-Local Filters

Use SDK-local filters for application-owned capture choices that the product cannot infer safely, such as tenant-specific suppression, final redaction, or browser network filtering before an event leaves the page.

Use project capture rules first for operational noise that should be visible to the team. SDK-local drops are harder for another operator or agent to discover later.

Verify The Change

After changing capture policy or rules:

  1. Wait for SDKs to fetch the updated configuration.
  2. Reproduce or observe the noisy path.
  3. Confirm the billing breakdown stops increasing at the previous rate.
  4. Confirm important incidents still open and have enough context.

For incident-specific triage, see Managing Noise. For ingestion semantics and quota behavior, see API Ingestion.

On this page