For code that works in the demo—but still needs evidence

AI coding tools can build useful software quickly. They can also repeat the same assumption across implementation, tests and explanation. A confident answer from the model is not independent evidence that the software behaves correctly.

Spotted Ridge Engineering checks one agreed question against observable results. That question might concern a release blocker, a regression, an integration, a data mismatch, a boundary case or a fix that needs to be verified before handover.

Start small: one issue, one expected behaviour and one decision you need to make. The standard single-issue verification is A$249.

What the check can cover

  • Expected versus observed behaviour: reproduce the failure and define what a valid result would look like.
  • Regression risk: check whether a proposed repair fixes the target issue without breaking the nearby behaviour included in scope.
  • Interfaces and data: examine an agreed API, schema, dependency or integration mismatch.
  • Boundary conditions: test relevant empty, maximum, invalid, timing or failure-path inputs.
  • Evidence quality: separate what the results establish from what remains uncertain or outside scope.

What you receive

You receive a concise written finding showing the agreed question, relevant evidence, the result, known limits and a practical next step. When a repair is included in the agreed scope, the report can show the affected section before and after the change without exposing unrelated code.

A passing result applies only to the tested conditions. It is not a blanket guarantee, penetration test, compliance certification or full-codebase security audit unless one of those services is separately scoped in writing.

AI code audit or verification?

People search for “AI code audit”, “vibe code audit” and “AI code review” to describe several different services. A broad audit surveys a larger codebase for categories of risk. This service begins with a narrower verification question and produces evidence about that question. The smaller scope makes it practical when one blocker is holding up a release or handover.

If the problem spans an entire workflow or system slice, the A$2,800 verification sprint is the better starting point. We confirm that before any paid work begins.

Tools do not change the standard

The software may have been created with Cursor, Lovable, Bolt, Replit, Claude, Codex, Copilot or a conventional development team. The verification standard is the same: define the required behaviour, challenge the relevant assumptions, preserve the evidence and state the result without stretching it beyond the scope.