Watching never counts.

There is a moment in every automated check where something gets treated as evidence. A script finished. A recording replayed. A checklist got ticked. Somebody decides that counts, and from then on a green result means the process works. That decision is where most testing quietly goes wrong.

Finishing is not the same as working

A script that drives your quote screen can finish perfectly while the business rule it was meant to protect has stopped applying. It clicked every control it was told to click. Nothing threw an error. The run is green.

But nobody checked that the 25% quote actually entered Finance Review. Nobody checked that sending it before approval was actually refused. The script proved that clicking finished, which is a much smaller claim than the one you wanted. This is not a flaw in any particular tool. It is what happens when the thing being asserted is the sequence of actions rather than the outcome.

Why a demonstration is not proof either

When you show Milita a process, she is learning. She watches you create the quote, submit it, and hit the wall at 25%. She records what she saw and what she thinks it means. None of that counts as proof, and she will not let it.

The reason is simple: you helped. You knew which screen to open. You had the right permissions already. Your data was sitting there from last time. A process that works when an expert walks it through is not the same as a process that works. So the demonstration produces a draft. Nothing more.

What earns the word

A Proof happens when Milita performs the whole process herself, in a fresh cloud session, as the people the rule actually requires, with fresh test data and no help from anyone. Then she checks the results that matter:

• The 20% quote reached Ready to Send
• The 25% quote entered Finance Review
• Sending was refused before approval
• The same quote went through after Finance approved it

If every required result holds, that is a Proof. If one does not, it is a Change, and she tells you what is different and when it last worked. The distinction sounds pedantic until the day something breaks. Then it is the only thing that matters, because a green result you cannot interrogate is worth roughly nothing.

The uncomfortable version

Most teams do not have a way to answer "does this still work?" without a person going and looking. That is not negligence. Salesforce does not tell you when a process quietly stops working, and a test suite that asserts on clicks will not either. The fix is not more tests. It is deciding what counts as evidence, and refusing to accept anything weaker.