Why the two drift apart
The reasons are ordinary. The fix was merged but not deployed, or deployed to staging only. It was deployed to one of three services that share the flaw. The patch blocks the exact payload in the report and not the variant next to it. A dependency was upgraded in one manifest while a second lockfile still pins the old version. A later change reverted the configuration. In each case the person who closed the ticket acted in good faith.
This is why "the developer says it is done" and "the pull request was merged" are useful signals and poor evidence. They describe intent and effort. Verification describes the state of the system.
What counts as verification
Verification means running the same kind of test that found the problem, against the same target, and seeing that the problem is no longer detected. Each part of that sentence is there for a reason.
The same kind of test. A finding from dynamic testing is verified by dynamic testing of the endpoint. A vulnerable dependency is verified by scanning the dependency manifest again. Reading the diff and agreeing with it is code review, which is valuable and is a different thing.
The same target. A clean result from staging does not verify a production finding. Nor does a clean result from a different host that happens to run the same code.
A test that ran to completion. This is the condition that gets skipped. If the re-test timed out, was blocked by a firewall, could not log in or stopped halfway, then not seeing the finding tells you nothing.
A way to recognise the finding. The re-test has to be able to say "this is the same issue as before". Scanners do this with a fingerprint built from stable properties such as the rule, the location and the parameter. A finding that cannot be matched cannot be confirmed as gone by a machine, and needs a person to check it.
Three outcomes, not two
It is tempting to treat a re-test as pass or fail. There are three results, and the third is the one that decides whether your numbers can be trusted.
1. Still present. The re-test found the issue again. The finding goes back to open, with the new evidence attached. This holds even if the re-test was otherwise incomplete: a positive detection is a positive detection.
2. Not detected. The re-test completed against the right target and did not find the issue. The fix is confirmed and the finding can be recorded as verified.
3. Inconclusive. The re-test did not complete, ran against a different target, or had no fingerprint to compare. Nothing has been learned. The finding stays where it was, with a note saying why, until someone runs the test again or checks by hand.
The failure to avoid is treating the third case as the second. A system that records "verified" whenever a re-test does not report the issue will mark fixes as confirmed every time a scan fails to start. The dashboard then looks healthy for exactly the wrong reason. If your compliance evidence is built from verified findings, the evidence is wrong as well.
Reopening without blame
A finding that comes back is information. Teams that treat a reopen as a mark against the developer get fewer reopens and no fewer vulnerabilities, because people stop re-testing. Make reopening cheap and normal instead.
Reopening also applies to findings that were closed for other reasons. An accepted risk is accepted under certain conditions, and conditions change: the internal service becomes internet-facing, or an exploit is published. A false positive may turn out to be real once someone looks more closely. Anyone with the right role should be able to reopen a closed finding, and doing so should clear its resolution date so that your time-to-fix figures stay honest.
Keep the history
For each finding, keep who changed its status, from what to what, when, and the note they left. Keep the re-test results too, including the inconclusive ones.
The history answers questions that come up later. An auditor asks how you know a critical finding from March was fixed. A new team member asks why a finding was accepted. A finding reopens for the third time and you want to know what was tried before. Without the history each of these is an archaeology project. With it, each takes a minute.
One practical rule follows: do not delete findings that have history. Close them. A deleted finding takes its audit trail with it.
A short routine
1. When a fix is deployed, mark the finding as fixed and say where it was deployed.
2. Re-test against the same target. Wait for the deployment to finish first.
3. Read the outcome as one of three. Treat inconclusive as "not yet known" and run it again.
4. If a re-test cannot run at all, for example because the original test needed credentials that have expired, verify by hand and write down how.
5. Count verified fixes, and not closed tickets, when you report progress.
How Spoorder handles it
In Spoorder, marking a finding as fixed moves it to awaiting verification and, for findings from live-target scans, starts a re-test. You can also start one with Verify fix. A re-test that detects the issue again reopens the finding. A complete re-test of the same target that does not detect it marks the finding verified. Anything else is recorded as inconclusive and leaves the status unchanged. Each step appears in the finding's history timeline, and a verified or accepted finding can be reopened from the finding page.
Ready to scan your application?
Free health check: HTTPS, TLS and security headers for one URL, in about a minute. Free account, no credit card.
Scan your app free →