I believe that one of the most important problems created by trying to boil teaching down to a checklist of non-negotiables is the lethal mutations that are likely to follow. When we say that every teacher, in every lesson, must do X, Y, and Z, we strip away the principle-based component of teaching and turn it into a playlist: Do A, then B, then C, in this order, every day, and students will learn. Obviously, that’s not how learning works. Take retrieval practice, for example. If we say every teacher must begin every class with retrieval practice, or every teacher must end every lesson with an exit ticket, or every lesson must include explicit modeling, we inevitably start trying to fit square pegs into round holes. Practices that can be powerful when used purposefully, appropriately, and in alignment with the principles behind them become mutated versions of themselves. Eventually, the practice can become so divorced from the conditions and principles that made it effective in the first place that we are still technically “doing” the strategy, but we are no longer doing the thing the evidence actually supports. The goal shouldn’t be to create a universal playlist of practices for teachers to execute. It should be to understand the principles that make those practices effective, and then give teachers the knowledge and judgment to determine when, why, and how to use them.

Occasionally, explicit instruction gets oversimplified into a process where we simply tell students what they need to know and what they need to do. Sure, that is one component of explicit instruction, but it is far from the whole picture. Explicit instruction also means showing students not just what to do, but how to do it. That requires explicitly modeling the steps within a process while thinking aloud, making the expert thinking that is normally invisible visible to students. I did this today with my students as we moved from our Life in the Sahara and the Sahel case study into transfer. Students had learned about specific environmental challenges and human adaptations within the Sahara and Sahel. We had then generalized those examples into deeper, transferable patterns. Today, I modeled how to take those patterns and apply them to a new desert environment that was deliberately different from the context in which the knowledge was originally learned. I walked students through the process step by step, thinking aloud as I went: What am I noticing? What environmental challenge is present? Which underlying pattern does this connect to? What evidence supports that connection? How can I use what I already know to reason about this unfamiliar context? But I’ve added a new wrinkle to my think-alouds: ten minutes of reflection after the modeling. Instead of immediately asking students to imitate the process independently, I give them time to explicitly think about what they just heard and saw. What did I do? What steps did I follow? What questions was I asking myself? What disciplinary thinking did I use to move from what I knew to what I needed to figure out? And, most importantly, what was the thinking behind the steps? The goal of modeling and thinking aloud is not for students to watch an expert solve a problem. It is to help them begin to build a mental model of the process itself. That distinction is essential for transfer. If we want students to use knowledge in new and unfamiliar contexts, we cannot just teach them the answer or even demonstrate the procedure. We need to make the thinking processes that connect knowledge to action visible and then give students deliberate opportunities to notice, articulate, and eventually reproduce those processes themselves. Tell them. Show them. Think aloud. Give them time to process the thinking. Then make them do the thinking themselves. That is a much richer conception of what explicit instruction can be.

AI will mark essays better than humans in the very near future and thankfully, we can avoid this farce every year. Well trained/tested LLMs can already mark every script several times, using multiple agents to identify outlier/flawed judgements and send disagreements for human review. Early evidence suggests that, on clearly defined tasks, some AI systems already agree with examiner consensus more closely than examiners agree with one another. arxiv.org/abs/2606.24973