True Test of Knowledge (TTK) is a framework for designing assessments that produce reliable evidence of student understanding in the age of generative AI. It has three pillars.
Pillar 1: In-person administration
An assessment is only as trustworthy as the conditions under which it's administered. TTK starts from the premise that summative assessments — the ones that determine grades, placement, or mastery — must happen in a controlled environment.
In practice, this means:
- Students physically present
- Devices controlled (either not used, or on a locked-down exam mode)
- Time-bounded
- Proctored
This is conservative by 2015 standards but increasingly the only way to produce trustworthy individual assessment data in 2025.
Pillar 2: Questions that require student-specific context
TTK assessments ask questions that a generic AI cannot answer well, because the answer depends on information the student gained from being in your class specifically.
Weak prompt (AI can answer):
"Explain the causes of the French Revolution."
Strong TTK prompt (AI can't answer without being there):
"Using the primary source we analyzed on March 12 — the 1789 pamphlet from a third-estate lawyer — explain how one of the causes of the French Revolution we discussed on Wednesday is reflected in the author's framing."
A student who attended class and did the reading can answer this. A student who only has the general topic can't.
The specificity doesn't have to be dramatic
You don't need to restructure every prompt. Even small additions — "building on Tuesday's discussion" or "using the lab data from your group" — shift the question away from generic recall.
Pillar 3: Process visibility
TTK assessments make the student's reasoning process visible, not just the final answer. In multiple choice, this is less relevant (the process is instant). In written response, it's essential.
Techniques:
- Show your work. Math and science require explicit intermediate steps, not just the final answer.
- Explain your reasoning. After a multiple-choice question, an optional explanation prompt — "why did you pick this answer?" — surfaces whether the student understands or just guessed.
- Compare and contrast. Asking students to explain the difference between two concepts is harder to AI-assist than defining either one.
- Personal reflection with specifics. "What did you find confusing about this unit?" — fine but weak. "Identify one misconception you had at the start of this unit and explain what changed your thinking." — much stronger.
Putting it together: a TTK-compliant test
A single test can hit all three pillars:
Multiple choice section (auto-graded): 10 items, class-specific scenarios where possible. E.g., instead of generic word problems, reference a scenario from a prior class activity.
Short answer section (rubric-graded): 3 items, each requiring specific reference to course material from a defined time window. E.g., "In the last two weeks we've discussed X. Apply one of those concepts to..."
Written response section (rubric-graded): 1 longer prompt asking the student to synthesize, compare, or reflect. Explicit requirement to reference specific class discussions or readings.
Administration: In-class, 50 minutes, no devices, proctored.
Grading: Multiple choice auto-graded on scan. Short answer and written response graded in a web interface (PaperScorer makes this fast — responses appear side-by-side with your rubric).
What TTK doesn't solve
TTK works for summative assessment. It doesn't solve:
- Evaluating work that's genuinely supposed to happen outside class (research projects, extended writing)
- Assessing collaborative work
- Capturing skills that don't surface in a timed test (creativity, sustained investigation, oral communication)
For those, different approaches apply: process portfolios, oral defenses, peer review. TTK isn't the whole assessment toolkit — it's the reliable-evidence-under-AI-conditions piece of it.
Key takeaway
Three pillars: in-person administration, course-specific prompts, visible process. Design assessments around these and you get trustworthy evidence of individual student knowledge — even when AI is everywhere.
