# Project insights

Interpret hosted traffic, errors, operational signals, and per-flag totals accurately.

Project Insights is a read-only operational surface built from privacy-safe hosted aggregates. Use it
to find configuration and runtime work, not as product experimentation analytics.

Insights covers requests that reach FeatureGate's hosted runtime API. Local Node.js snapshot
evaluations remain inside the application and are not counted as exposures.

## Choose a range and environment

Select a 7-, 14-, or 30-day range. Add an environment filter when production traffic, staging tests,
and development activity would otherwise obscure each other.

Start broad when looking for project-wide change. Narrow to one environment when diagnosing an
incident, rollout, or rate-limit pattern.

The selected range changes aggregation context. It does not recover raw requests or evaluation values.

## Interpret the summaries

Insights can summarize:

- Hosted runtime requests.
- Request errors and status classes.
- Rate-limit events.
- Volume by environment.
- Direct hosted evaluation totals by flag.
- Operational signals that link to a flag or environment.

Use rate-limit and error trends before blaming a targeting rule for an application change. A drop in
requests may be deployment traffic, credential failure, or migration to local snapshot evaluation.

## Use operational signals

Action-first signals can lead to kill-switch state, fallback review, stale lifecycle work, or an
environment that needs configuration.

Follow the link to the owning resource and validate current state. A signal is a prompt for review,
not an automatic diagnosis or configuration change.

Projects with no flag, key, or first hosted request should use project activation and environment
readiness. Insights is not the setup wizard for an inactive project.

## Understand per-flag totals

Direct per-flag totals count existing requested flags in hosted evaluation calls. They do not count:

- Unique users or accounts.
- Local snapshot getter calls.
- Every application code-path execution.
- Which value each subject received.
- Conversion, retention, or experiment outcome.

A high total can represent repeated requests from a small population. A low total can coexist with
substantial local SDK use.

## Correlate with application telemetry

Use application metrics, logs, traces, and business analytics to answer questions about user impact.
Correlate through time windows, deployments, and privacy-safe request IDs.

Do not copy full context or resolved flag values into FeatureGate notes. Avoid logging them in the
application merely to compensate for aggregate platform telemetry.

## Privacy boundaries

FeatureGate does not store raw context, user identifiers, targeting values, or resolved values in
these aggregates. Insights is intentionally not a raw request explorer.

This protects callers but limits retrospective analysis. Design application observability before a
high-risk rollout rather than expecting FeatureGate to reconstruct exposure later.

## Common interpretations

- **Errors increased after key rotation:** verify the deployed key and scopes.
- **Rate limits increased:** batch evaluations and remove per-flag request loops.
- **Production has no hosted traffic:** confirm the integration uses HTTP rather than local snapshots.
- **One flag has no direct totals:** confirm it is requested through hosted evaluation and exists.
- **A rollout outcome is unclear:** use application-owned exposure and business telemetry.

Related pages: [Environment readiness](/docs/operate/environment-readiness),
[Run a controlled rollout](/docs/operate/controlled-rollout), and
[Errors and fallbacks](/docs/evaluate/errors-and-fallbacks).