Process, Product and Perspective

Test Community Network, September 2026

There is a phrase that comes up constantly in assessment conversations right now: process over product. It sounds right. It feels like progress. And it is not enough.

The argument goes like this. AI can produce polished outputs, so we should stop grading the final product and start grading how the student or candidate got there. Show your working. Evidence the journey. Capture the process.

I agree with the direction. But I think the framing is weak, and I want to explain why.

Process captures what someone did. The steps they took, the sequence, the actions. Product captures what they produced. The output, the deliverable, the final answer. Both matter. But neither captures why they made the choices they made, what they considered and rejected, what they would do differently if conditions changed, and whether they actually understand the reasoning underneath their own work.

That is a third layer. And without it, the evidence is incomplete.

"Process tells you what someone did. Product tells you what they made. Neither tells you why. And in assessment, the why is where competence actually lives."
— Tim Burnett, Founder, Test Community Network

The hundred-year trade-off

The reason this layer has been missing is not that nobody thought of it. It is that capturing and interpreting reasoning evidence was too expensive to do at scale.

Multiple-choice questions are the clearest example. An MCQ separates the cost of creating the assessment from the cost of evaluating the response. You invest once in writing the item. Every candidate who sits it costs almost nothing to score. That economic model made large-scale assessment viable. But it came with a trade-off that we quietly accepted: you get an answer, but you never get the reasoning.

"MCQ assumes that because someone knows the answer, they understand how they know it. That has never been true. We just could not afford to ask the follow-up question."
— Tim Burnett, Founder, Test Community Network

This is not a criticism of the people who built these systems. Given the constraints they worked within, MCQ was a rational solution. The problem is that those constraints no longer exist, and we are still designing as though they do.

AI changes the economics. Not by replacing human judgement in marking, but by dramatically reducing the cost of processing reasoning evidence. AI can interpret, sort, flag and summarise. A human expert still makes the judgement call, but the effort required to reach that judgement is a fraction of what it used to be.

"For the first time, you can ask a candidate not just what they would do but why, and have that reasoning captured, interpreted and made available for human review at a cost that actually works at scale."
— Tim Burnett, Founder, Test Community Network

This is not a theoretical possibility. The barrier that propped up the weakest assessment format for a century has fallen. The question is whether we are willing to walk through the gap.

Process, product, perspective

So what does a three-layer evidence model look like?

Process captures the actions. What steps did they take? What sequence? What tools did they use and when? This is the layer that processfolios, design logs, version histories and checkpoint evidence already target. It is valuable. But on its own it tells you someone went through the motions, not that they understood what they were doing.

Product captures the output. What did they produce? Does it meet the standard? Is it accurate, complete, fit for purpose? This is what most traditional assessment measures. It is necessary. But AI can now produce outputs that meet the standard without any human understanding behind them.

Perspective captures the reasoning. Why did they make those choices? What did they weigh up? What would they change? What risks did they consider? What did they challenge or reject? This is where competence actually becomes visible, and it is the layer that has been missing.

"Two candidates can follow the same process and produce the same product. One understood what they were doing and could adapt if conditions changed. The other followed a recipe. You cannot tell them apart without the third layer."
— Tim Burnett, Founder, Test Community Network

The three layers work together. Process without product is activity without outcome. Product without process could be someone else's work. Either without perspective could be performance without understanding.

Why this changes every level of the grid

In the first article in this series, I introduced the AI Reliance and Assessment Resilience Grid, which maps five levels of AI involvement from fully unaided to fully delegated. Perspective evidence matters at every level, but for different reasons.

At Level 1, where the person works without AI, perspective evidence is what distinguishes genuine understanding from rehearsed performance. Someone can memorise the right answer. They cannot fake the reasoning behind it if they are asked to explain, defend and adapt it.

At Level 2, where the person works with AI, perspective is what separates thoughtful collaboration from passive acceptance. This is the critical level, because it is where most professional work is heading.

"At Level 2, the product might look identical whether the expert contributed or not. The process might look similar. It is only when you ask why they accepted this output and challenged that one, why they changed this section and left that one, that the professional value becomes visible."
— Tim Burnett, Founder, Test Community Network

At Level 3, perspective evidence reveals whether someone can generalise from catching one error to recognising a different kind of error next time. At Level 4, it separates expert judgement under pressure from a lucky guess. At Level 5, where the person is not performing the task at all, the perspective is the competence. There is nothing else to assess.

AI is not a calculator

There is a temptation to treat AI as the next calculator: a more powerful tool, but still a tool. I think that is wrong, and the distinction matters for how we design assessment.

"A calculator does exactly what you tell it. It does not suggest an alternative approach. It does not produce a plausible but wrong answer with complete confidence. It does not reshape how you think about the problem. AI does all of those things. It is not a tool you use. It is closer to a colleague you work with."
— Tim Burnett, Founder, Test Community Network

Working with a colleague requires different competencies from operating a tool. Knowing when to trust, when to challenge, when to override, when to ask a different question, when to step back and do it yourself. Those are not the old constructs delivered through a new interface. They are genuinely new constructs.

I think the constructs are shifting. This is not like having a calculator. AI is a brain with tools, and that is fundamentally different from anything assessment has had to deal with before.

That means the evidence has to shift too. If the relationship with AI is collaborative rather than instrumental, then the assessment has to capture the quality of that collaboration. And you cannot capture the quality of a collaboration without perspective evidence: what the person contributed, what they challenged, what professional value they added that the AI could not.

Research from AQA on AI-generated test items found something instructive here. When item authors only reviewed AI output, the result was what I would call surface plausibility: items that looked right and read well but lacked the depth that comes from genuine thought. The fix was simple in principle. Authors responded to questions themselves before seeing the AI-generated answer. That forced the thinking to happen first.

"It is not enough to have an expert in the loop. You need value in the loop. You need effort in the loop. The expert has to be doing the cognitive work that justifies their involvement, not just signing off someone else's thinking."
— Tim Burnett, Founder, Test Community Network

The same principle applies to any assessment where AI is involved. The question is not just whether the expert was present, but whether they did the thinking. And without perspective evidence, you cannot answer that question.

What this means in practice

None of this requires a revolution in assessment design. It requires adding a layer that should have been there all along but was too expensive to capture.

A written assessment adds a reflective rationale: why did you structure it this way, what did you consider, what would you change? A practical assessment adds a professional conversation: talk me through what you did and why. A simulation adds a debrief. An AI-assisted task adds a record of what the person challenged, changed and contributed.

The formats already exist. The cost of processing the evidence they generate has been the barrier. That barrier is falling.

"We have accepted a hundred-year compromise. We tested the what because the why was too expensive. AI has just made the why affordable. The question now is whether we are willing to change."
— Tim Burnett, Founder, Test Community Network

Tim Burnett is the founder of the Test Community Network.

Article 1 - Hands Off, Eyes Open, Brain Engaged. This is Article 2 of a four-part series. Article 3 will explore what recent research into AI-enabled cheating reveals about assessment format and design.