Automatic Insights

Insights are plain-language alerts generated automatically from your traffic data — no dashboards to configure, no thresholds to set manually.

They appear in the dashboard's Insights panel and are recalculated on every flush cycle (default: every 60 seconds). Filter chips let you show only a specific type.

Insight types

ANOMALY

Detects statistically abnormal spikes in P99 latency using Z-score analysis.

"GET /search P99 latency is 8× above the 7-day average (current: 1240ms, normal: 148ms). Possible cause: dependency timeout or unusual traffic pattern."

Triggers when: P99 latency exceeds the 7-day moving average by more than 3 standard deviations.

---

DRIFT

Detects progressive, long-term latency degradation that would be invisible on a real-time dashboard — a slow leak rather than a spike.

"GET /products has been progressively degrading for 18 days: +3.2ms/day. 30-day projection: +96ms."

How it works: Ordinary least squares (OLS) linear regression over daily P90 buckets for the last 30 days, computed per endpoint.

Triggers when:

• At least 7 days of data are available for the endpoint
• The regression slope is ≥ 5ms per day

The insight includes the slope, the number of days observed, and a 30-day projection.

---

DEAD

Identifies endpoints that have received no traffic for an extended period — safe candidates for deprecation.

"DELETE /legacy/import has received no requests in 23 days. Consider deprecating this endpoint."

Triggers when: An endpoint has zero requests for the past 21 days.

---

PERF / OK

Compares performance metrics before and after a deployment, when a release tag is configured.

PERF — "POST /orders P99 latency increased by 40% after deploying v1.3.0 (14h32 yesterday). Before: 230ms — After: 322ms."

OK — "Deploy v1.4.0 (3 hours ago) introduced no detectable regression. Latency stable, error rate unchanged."

Triggers when: A release value is set in the config and at least 30 minutes of post-deploy data is available.

See Release Tracking for setup.

---

UNTRACKED

Identifies routes declared in your router that have never received a single request. Useful for spotting dead code, misconfigured routing, or undocumented endpoints.

"POST /admin/export is declared in the router but has never received any traffic."

Triggers when: A route appears in the Express / FastAPI router but has zero entries in the metrics database.

Node.js only

UNTRACKED insights require that the SDK can inspect the Express router at startup. They are not generated for routes registered after the first request.

---

How insights are prioritized

Insights are sorted by severity in the dashboard:

Priority	Types
🔴 Critical	ANOMALY with Z-score > 5, PERF regression > 50%
🟠 Warning	DRIFT (progressive degradation), PERF regression 20–50%, ANOMALY Z-score 3–5
⚫ Info	DEAD endpoints, UNTRACKED routes
🟢 Positive	OK (clean deploy confirmation)

Enabling release insights

Set the release option to activate PERF and OK insights:

Node.js

app.use(apiforge({
  mode: 'local',
  release: process.env.npm_package_version,
}))

Python

import os
app.add_middleware(
    ApiForgeMiddleware,
    mode="local",
    release=os.environ.get("RELEASE"),
)

Each time the value changes (i.e. on a new deploy), APIForge creates a comparison baseline automatically.