What I conclude, and the evidence it takes.

The most useful sentence a reviewer can say is that the evidence does not support a conclusion here. It is also the least commercially convenient one, which is why it is worth publishing the standard in advance rather than discovering it mid-engagement.

• From documents alone I will conclude what documents can establish: what was agreed, what was promised, what is internally inconsistent, what changed and when. Never that a system behaves as described.
• That a system behaves as described requires primary evidence. Raw output, telemetry, exports, the test running while I watch it. A report that a test passed is a claim about evidence. It is not sufficient evidence on its own.
• Intent, sequence and omission require people. Documents record what survived the meeting.
• "Unable to conclude" is a real outcome. Where I cannot get there, the opinion says so, names what is missing and what it would take to close it. I do not fill the gap with judgment and call it a finding.
• A system's own report of its success is not evidence of success. That holds for Your supplier's test harness, and it holds for every automation I use in my own work.