Skip to main content

Reading the Report Card

Every completed test produces a Report Card — a single-page graded summary that tells you exactly how your API performed and what to fix.


The overall grade

Load Curl weighs 8 metrics and produces a letter grade from A to F.

GradeMeaning
AExcellent — all metrics green, ready to scale
BGood — minor latency issues, low production risk
CFair — needs tuning before handling high traffic
DPoor — multiple metrics failing SLA targets
FCritical — server degradation detected, do not ship

The grade is designed to be a single, shareable signal. You can paste a public report link into a pull request or send it to a manager without needing to explain raw numbers.


The 8 graded metrics

1. P95 Response Time

The 95th-percentile latency across all requests during the test.

  • Green (A): < 300 ms
  • Yellow (B/C): 300 ms – 1 s
  • Red (D/F): > 1 s

P95 is a better signal than average latency because it captures what your slowest users actually experience. An average of 50 ms is meaningless if 5 % of requests are taking 3 seconds.

2. Error Rate

The percentage of requests that returned a non-2xx HTTP status code.

  • Green (A): < 0.1 %
  • Yellow (B/C): 0.1 % – 1 %
  • Red (D/F): > 1 %

3. Requests per Second (RPS)

Peak throughput your API delivered during the test. Higher is better, but this metric is graded relative to your concurrency level — a 50-worker test scoring 400 RPS is proportionally the same as a 1,000-worker test at 8,000 RPS.

4. Throughput

Total data transferred per second. A sudden drop in throughput mid-test often signals a connection pool exhaustion or a rate limiter kicking in.

5. Server CPU Peak

The highest CPU utilisation observed on your server during the test. Requires the optional server agent.

  • Green (A): < 70 %
  • Yellow (B/C): 70 % – 90 %
  • Red (D/F): > 90 %

6. Availability

The percentage of time your server returned any response (including error codes). A value below 99 % usually means the server was completely unreachable for some workers.

7. Latency Stability

Measures how much latency varied over the test duration. An API that starts at 200 ms and ends at 1,500 ms has poor stability — a sign of a memory leak or connection saturation over time.

8. P99 Response Time

The 99th-percentile latency. Used as a tiebreaker to distinguish a B+ from an A–. If your P95 is healthy but your P99 is very high, a small percentage of users are experiencing severe delays.


Recommendations panel

Below the grade, the Report Card shows specific, actionable recommendations for every failing dimension. For example:

P95 latency exceeds 300 ms target. Consider adding a CDN layer or optimising the database query on /v2/users — 78 % of slow requests originate from a single unindexed join.

Load Curl traces slow requests back to specific endpoints in multi-scenario tests, so you can pinpoint which call in a chain is the bottleneck.


Sharing and exporting

ActionHow
Share a public linkClick Share → toggle Public link on → copy URL
Export to PDFClick ExportDownload PDF (Pro plan and above)
Embed in a PRPaste the public link as a comment; GitHub will render a preview card
Compare to previous runClick Compare → select any prior run from the dropdown

Next step

Now that you understand your results, learn how to set threshold alerts so you get notified automatically when a key metric degrades.