#038Guide3 min

Understand analysis coverage

Separate examined systems, missing checks, partial evidence and points that require human proof.

Hackaiz team · updated 21/11/2026

Illustration: a team reads a security report and marks priorities.
Illustration: a team reads a security report and marks priorities. Illustrative photo, not a screenshot.

01Coverage describes the reach of the evidence

Coverage answers a practical question: what did this analysis actually examine? It is not the number of issues detected. Two reports can each contain one finding while covering different targets or different checks. Read the scope alongside the result so that a narrow examination is not mistaken for a review of the entire application.

A connected repository gives access to source code; it is not automatically a connection to the running database. Likewise, a website result is not proof that private server code was inspected. Start by listing the systems involved in the decision you need to make, then compare that list with the targets and evidence in the report. Treat a missing system as a gap, not as an implicit success.

02Read the control states as distinct conclusions

The technical pre-audit alignment report links control points to stored analysis evidence. Expand the relevant standard to see the states beneath it. ‘Issue detected: see evidence’ calls for examination of the finding. ‘Examined, nothing found’ means the applicable examination produced no issue; it does not turn every related security question into a proven fact.

‘Partly examined’ and ‘Not examined by this analysis’ preserve important uncertainty. Other entries say ‘Your answers or documents are expected’ or ‘Must be proven by a person’. If you supplied an answer, ‘Declared by you: in place (not proven)’ remains a declaration rather than independent evidence. These labels prevent an automated technical result from becoming an invented compliance guarantee.

03Check selection, access and execution separately

There are three different boundaries to review: what you selected, what the connections allowed, and what finished running. A selected check can still lack access. A connected system can still have unfinished work. An analysis can finish its selected work while leaving a deliberately excluded topic unanswered. Use the report states rather than assuming these boundaries are identical.

Unticked checks are neither executed nor billed. Work that does run costs credits, and a credit interruption can leave part of the scope unfinished. For a website or API, domain ownership proof is a prerequisite: use DNS TXT _hackaiz, the HTTPS verification file, or ask the project’s AI to create that file. Ownership proof authorizes examination; it does not itself establish good security.

  1. 1Identify each target whose behavior matters to your question.
  2. 2Compare the available connections and selected checks with that target list.
  3. 3Review completion and control states for each target separately.
  4. 4Record missing access, unfinished work and required human evidence as different follow-up tasks.

04Use coverage to make a bounded decision

Suppose your team is preparing a release that changes database permissions. Evidence from package checks may still be useful, but it does not answer whether the new permissions isolate customer records. Request the relevant authorized examination and human review instead of describing the existing report as broad enough simply because it has finished.

Where available, ‘Download (JSON)’ exports the technical alignment report with control states and linked evidence. Use that record to explain both the observations and their limits. It is explicitly not certification or legal advice, and no compliance percentage is computed. When presenting the report, name the target, the evidence obtained and the unanswered questions rather than inventing a single percentage for application security.

Frequently asked questions

Was this article helpful?

Related articles