Give the contrary case a hearing
If a prompt contains the same complete facts, clear labels, and unchanged authority, reordering two blocks need not change what answer is justified. It may leave the response unchanged too. That is a fair case for treating some ordering choices as formatting. Whether reliability changes is a further question.
Change only the order
Imagine two complete blocks: a proposal for a twenty-inch print and a dated decision explicitly rejecting it in favor of sixteen inches. Swap their positions while retaining the wording, labels, message roles, request, and authority. The approved size has not changed.
The comparison asks whether a model recovers that decision equally reliably in both arrangements. A single difference between two responses would not establish a dependable order effect; repeated comparisons would be needed.
Dropping the decision tests omission. Calling the proposal “Current instructions” changes its assigned role. Separating a correction from its referent may make the connection harder to follow. These are useful comparisons, but they should not be presented as evidence about order alone.
Pause here: If wording, roles, evidence, and authority stay fixed, what observation would persuade you that order mattered here?
Inspect the arrangement, test its effects
The encyclopedia's historical prompt-assembly draft makes sequence explicit. Its later admission doctrine requires ownership, provenance, time, uncertainty, and authority to survive rendering. The regulator specification describes rebuilding authorized context for each invocation.
Together these sources describe an inspectable design. They do not demonstrate that a particular sequence works best. Good labels and attribution can support that inspection while the performance question remains open.
A useful record would preserve both assembled inputs, the model and settings used, the expected decision, and the responses. That makes a later disagreement inspectable without assuming that the preferred arrangement was responsible for a better result.
My answer: Order is one part of presenting context. It may matter to reliable interpretation, and the effect should be tested under controlled comparisons. Moving a passage cannot create authority, alter the recorded decision, or guarantee compliance.