1. Your results file stays on your device
The JTL/CSV you analyse is parsed on your own device, in both editions, and is not uploaded to our servers. In the online edition the credit meter receives the number of rows and analyses, never their content.
- Saved runs are stored in your account only when you press Save: the run name, the file names, the API labels and their aggregated KPIs. Not the samples.
- We do not receive response bodies, request bodies, headers, cookies, tokens or credentials from the system you test.
- If you configure an integration with another service, what the dashboard sends goes directly from your device to that service, not through us.
Check it yourself: in the online edition, open the developer tools of your browser on the Network tab and load a file. You will see the credit check, which carries row counts, and no request carrying the file.
2. How each figure is calculated
Every setting that changes a figure can be seen and changed in the dashboard. These are the defaults and the alternatives:
| Figure | Default | You can also choose |
|---|---|---|
| Percentiles (p50, p90, p95, p99) | Linear interpolation (Type 7): the default of R and NumPy, and Excel's PERCENTILE.INC. Computed from the individual samples. | Exact estimators: Type 6 (Excel PERCENTILE.EXC), Type 8, Type 9, Type 4 and nearest rank (Type 1, always a real measured value). Approximations, labelled as such: trimmed, winsorized, weighted, HDR Histogram and t-digest. |
| Error rate | 4xx and 5xx responses plus network errors, over all samples. | Also count 3xx, or count every response that is not 2xx. |
| Grouping | Per transaction / endpoint. | Global, per thread group, per host, per time window, weighted by threads, per user session, or thread group then transaction. |
| Time units | Detected automatically (milliseconds or seconds). | Set them by hand if the numbers look a thousand times off. |
The report JMeter generates can give slightly different percentiles from the same file, because it uses a different estimator. Both figures are correct for their method, and support can tell you which estimator matches the other tool.
3. Check a figure yourself
Pick one transaction and one percentile, and compute it outside i9x from the same file:
- Open the same JTL/CSV and keep only the rows of that label, in the same time window and with the same exclusions you applied in the dashboard.
- Compute the percentile of the
elapsedcolumn with a tool that shares no code with i9x, for example:
Excel =PERCENTILE.INC(B2:B10001, 0.95) Python (NumPy) numpy.percentile(elapsed, 95) R quantile(elapsed, 0.95) # type = 7 by default
- Compare the result with the figure the dashboard shows with the Type 7 estimator. They must agree to the rounding shown.
4. Nothing is left out without telling you
- If the row limit of your plan or the memory of the browser leaves records out, the dashboard says so: Showing X of Y records. Conclusions drawn from it are conclusions about a sample.
- KPIs shown while a test is still running are incremental and can change until it ends.
- Results from the local-simulation mode are synthetic, and the application marks them as simulated.
5. If a figure looks wrong
Report it as the support policy describes. We reproduce the figure with the same data, filters and estimator and compare it with an independent calculation. When the difference comes from the application, we confirm it as a defect and fix it in a new version.
6. Checking the installer
Every build is published on GitHub Releases with an Authenticode signature. The updater of the desktop app refuses any update that is not signed with our certificate or that changed after it was signed.
The certificate is currently issued by us and not by a public certificate authority, so Windows may show the publisher as unverified. To confirm that an installer is ours, right-click it and open Properties, Digital Signatures, Details, View Certificate, Details, and compare the Thumbprint with this one:
44F849DA0AEEDF516896BE4C1A5D5A352CD4C4D0
If Windows stops the installer, the installation guide explains what each warning means and what to do.
7. Account and payment security
- Encryption in transit with TLS 1.2 or higher.
- Passwords stored only as a value derived with
scrypt: we cannot read them. - Sessions in HttpOnly, SameSite cookies, with an anti-CSRF header requirement.
- Optional two-step verification with an authenticator app, single-use recovery codes and devices you can remember for 30 days. Codes are never sent by email or text message.
- Encrypted database backups, rate limiting and account lockout against brute force, and an audit log of security events.
- Payments go through Paddle, our Merchant of Record: card details are entered in its payment window and never reach our servers.
The details are in the privacy notice.
8. What we do not claim
- i9x Performance does not hold an ISO 27001 or SOC 2 certification. This page describes how the product works; it is not an audit by a third party.
- The analysis is information produced from the file you load. It is not a certification of your system or a guarantee of its capacity (terms, section 12).
- A correct calculation of an unrepresentative test is still an unrepresentative result. How well the figures describe production depends on how the test was designed.
Questions about security or about a figure: support@i9xperformance.com.