STRAND C · RUNG II · ENHANCE

Change the evidence, not the topic

You have run your own task against AI and you know the number — that is the rung below, and you should do it first. This page is what to do next, where the machine earned full marks: not swap the prompt, but change what you collect. It is a validity move, not a surveillance one, and it is smaller than it sounds.

A STARTING POINT — AND IT NEEDS YOUR JUDGMENT

This page is a draft, and I’m still iterating on it. It wasn’t written by an expert in your field — it is here to start a pathway you then shape to fit you: your subject, your level, your students, your department. Your expertise runs far deeper than anything here, and none of this is finished until your judgment has been on it.

That is the whole point. Judgment is the most important part of this — yours, and the judgment you are building in your students. These pages exist to help you develop and exercise it, never to replace it. Take what is useful, change what is not, and tell me where I have gotten it wrong — that is how these get better.

THE MOVE, IN ONE LINE

Where the shortcut works, stop chasing the topic and widen the evidence — add one thing the shortcut can't produce, and grade partly on that.

YOU'LL HAVE

Five-to-seven low-effort evidence options across subjects, a rule for choosing which one, and an honest read on where this helps and where it doesn't.

WHERE THIS SITS

One rung of Strand C. Below it, find where the shortcut works. Above it, a task built to fight back. Alongside it, the assessment-design page.

THE TRAP — SWAPPING THE PROMPT IS AN ARMS RACE YOU LOSE

The reflex, when a task turns out to be AI-doable, is to change the topic. New essay prompt every semester. A fresh set of questions. It feels like a fix, and it buys you exactly one term — because you have changed what the task is about while leaving what you collect untouched, and what you collect is a finished artefact that a machine can also produce. Next semester the shortcut catches up, and you write another prompt. That is a race, and the other runner does not get tired.

Why the validity frame changes the move. The assessment-design page names the two ways a task goes wrong. This is one of them, live. When the machine can do the task, the thinking you meant to assess has quietly slipped out of what you actually collect — the construct is now underrepresented by a finished product that no longer requires it. Swapping the prompt doesn't touch that; the new artefact is just as machine-producible as the old one. Changing the evidence widens the construct back out to include the thinking. You are not building a trap for the dishonest. You are collecting the thing you always meant to grade and had stopped seeing.

EVIDENCE YOU CAN COLLECT INSTEAD — OR ALONGSIDE

None of these replaces the assignment. Each adds a second, smaller piece of evidence that the shortcut can't generate on its own — and each is something good teachers were already half-doing. Pick one that fits the task; you are not meant to bolt on all of them.

A DRAFT YOU WATCHED TAKE SHAPE

The essay written in a shared doc whose revision history you can open, or a rough draft turned in a week before the clean one. You are collecting the becoming, not just the arrival. A history paper that has no messy middle is telling you something.

A FIVE-MINUTE CONVERSATION

You ask three questions about the reasoning behind the piece. “Why this source and not that one?” A student who owns the thinking answers easily; the gap where they can't is the evidence. Works in every subject and needs no technology.

A STEP MADE IN FRONT OF YOU

One move of the work done live — a single derivation at the board, one paragraph drafted in class, one translation rendered aloud, one titration read off. Not the whole task under supervision; one step, as a sample of the skill.

AN ORAL DEFENSE

Two minutes standing next to the finished work, explaining one choice in it and answering one follow-up. Old as the viva, and still the cheapest way to find out whether the understanding is behind the product or only in front of it.

AN IN-CLASS CHECKPOINT

A short written moment on the same construct, done in the room — the thesis sketched in ten minutes, one problem of the type solved cold. It doesn't have to carry the weight; it has to corroborate the take-home piece.

A PROCESS ARTEFACT

Annotated sources with the student's own margin notes, a lab notebook, a revision log, the search terms they actually used. The by-products of real work — collect them as evidence and the work has to have happened to produce them.

A REASONING TRACE, NOT JUST THE ANSWER

Ask for the wrong turn they abandoned, the assumption they had to drop, the counter-argument they considered. A machine hands you a clean final answer; the honest record of a mind changing its own direction is harder to fake and closer to what you wanted anyway.

HOW TO CHOOSE — MATCH THE EVIDENCE TO THE CONSTRUCT

There is one rule, and it is not “pick the hardest one to cheat.” It is: match the evidence to the thing you are actually assessing. Ask what the task is really for, and collect the evidence that shows it.

If you are assessing reasoning — the conversation, the oral defense, the reasoning trace. You want to hear the why, so collect the why.
If you are assessing process — the watched draft, the revision log, the annotated sources. You care how it was built, so collect the building.
If you are assessing understanding you can perform — the step made in front of you, the in-class checkpoint. You want to see the skill fire once, unaided.

Choosing to catch people instead of to measure the construct is the failure mode here. It changes what you collect for the worse: you start gathering suspicion rather than evidence of learning, and the rubric quietly stops being about the skill. If the new evidence doesn't tie to a thing on your rubric, you have added surveillance, not validity — take it back off.

BEFORE YOU OVERSELL THIS TO YOURSELF

This course has a standing rule: where the workshop version of a claim outruns the evidence, we say so. Three honest limits on the move above.

Authentic is not cheatproof. The largest study of contract cheating to date found that authentic assessment did not reliably protect against it — students in authentic-task courses still outsourced work at meaningful rates (Bretag et al., 2019). Changing the evidence you collect raises the cost of outsourcing more than relabelling the topic does, but nothing here is a silver bullet, and a page that sold it as one would be lying to you. This is a validity improvement first; any integrity benefit is a welcome side effect, not the guarantee.

Ask it of everyone, or it reads as an accusation. The instant “collect more evidence” becomes a thing you do only to the students you already suspect, it stops being assessment and becomes an interrogation — and the students who most deserve the benefit of the doubt are usually the ones who get pulled aside. Build the extra evidence into the task for the whole class. A conversation everyone has is a checkpoint; a conversation three kids have is a tribunal.

Watch the equity and the arithmetic. A five-minute conversation is five minutes; times 120 students it is ten hours, and it is not free. Some of these land unevenly — an oral defense taxes an anxious or a multilingual student on something other than the construct, which is the exact construct-irrelevant variance the assessment page warns against. So pick spots. Add the evidence to the two or three assignments that carry real weight, not to everything. One well-chosen checkpoint beats a semester of exhausting ones you abandon by October.

WHERE THIS SITS ON THE LADDER

Grade the AI against your rubric — the rung below. Do this first; it tells you where the shortcut works, which is the only place you need to change the evidence.
Write an assessment worth trusting — the validity frame this page stands on, and the construct-irrelevance warnings behind the caveats.
Building a Contested task — the next rung up, where the tool itself is configured to withhold, argue, or seed errors rather than just be collected around.
Four positions, set per assignment — the language for saying, on the task itself, what AI is and isn't for here.

These are starting points, not a mandate. The seven evidence options above are drafts — the right one for your subject is a thing you know and I am guessing at. Where I have named the wrong evidence for what you actually teach, you are right and I am not. Bring me the specific case and we will find the piece that fits it.

ALSO IN THIS COURSE

The fully worked version of this move. The student turns in the paper, then defends it out loud for five minutes from six standing questions drawn from their own text. Not surveillance — a change in where the grade lives. Every student speaking — the Defense

Where the shortcut works, don't play whack-a-mole with the topic. Change what evidence you collect — add the one piece the shortcut can't produce, ask it of everyone, and grade it against the thing you meant to teach. That is the whole move.

Bring me one assignment where AI does the work well, and we'll redesign the evidence together — on something real, in one sitting.

Jason Beyer · Educational Technology · Regis Jesuit High School · A.M.D.G.

HelpDesk@regisjesuit.com · 303.269.8080 · LAST VERIFIED — AUGUST 2026