After the Exam: Credentials in a World Where Everyone Has an Answer Machine

Imagine an examination hall on a Tuesday morning. The blue books are stacked on the lectern, the clock is above the door, and a proctor walks the aisle with the calm of someone who has done this for twenty years. Halfway down the room, a candidate has a phone under the desk, its screen dimmed to a glow the size of a coin. The proctor sees it. The proctor also has a laptop open on the lectern, because the marking scheme was drafted with a chatbot and the proctor wants to check that the answers agree. Who is cheating? The candidate who consults a machine to answer, or the teacher who consults the same machine to mark?

This hall is imagined. The problem it describes is not. Universities in several countries are asking the same question, and I think the most useful answers so far come from institutions that have stopped pretending the old examination still measures what it was built to measure.

The Memory Test Was Always a Race Against a Clock

For most of the history of schooling, an examination measured what a student could hold in her head and reproduce under time pressure. That was a reasonable proxy when the knowledge a student could reach was mostly the knowledge already in the student's head. I hold this as judgment: the memory test was a fair measure of a scarce thing, and the scarcity has ended.

The natural response to cheating was a detector. In August 2026, Inside Higher Ed reported that Yale, Vanderbilt, Johns Hopkins and Indiana had prohibited or discouraged AI detectors as sole evidence of cheating. Northwestern, Georgetown and NYU had disabled Turnitin's AI detection, and "at least a dozen" institutions had done something similar. Among faculty, 73% say they have personally dealt with AI-related academic integrity issues. Inside Higher Ed

Jennifer Frederick, executive director of Yale's Poorvu Center, put the logic plainly: the aim is to avoid "the inevitable cat and mouse game created by AI detection tools." The institutions have not given up on cheating. They have switched off the instrument that turned the problem into an arms race.

The memory test is dying, in my judgment, not because students became dishonest but because it stopped measuring what it was meant to measure. A correct answer no longer tells you whether a person can produce one.

Oral Exams Are a Patch, Not a Principle

The obvious reply is the one universities are reaching for now: sit the student in a room and ask questions. Oral exams are coming back. In a piece for The Conversation Africa published on 17 August 2026, Nicole Brownlie, a lecturer in education at the University of Southern Queensland, describes universities turning to them to stop students using AI to cheat, and she is unsparing about the trouble. The Conversation Africa

She cites a Norwegian study in which initial student presentations ranged from five to 16 minutes, follow-up discussions from seven to 23 minutes, and "some students received fewer than ten questions while others received nearly 50." The examination a candidate sat depended on the conversation they happened to have. The oral exam, sold as a cure for the unreliable essay, carries its own unreliability, and the flaw is harder to see because it lives in a voice rather than on a page.

The same piece reports a German study that found more anxiety in oral exams than in written ones, and warns that students speaking a non-native language may be disadvantaged. For learners who sit assessments in a second language, that is not a footnote. A test of composure is not a test of mastery.

Brownlie's sharpest line is that using oral exams as an "AI lie detector" turns assessment into a test of confidence. I think she is right. Imagine two students. The first knows the material, freezes when the examiner asks her to explain one step, and leaves the room believing she has failed. The second knows less, has rehearsed the answers with a machine for three nights, and speaks without a tremor. (Imagined composite.) To the examiner they can look like the same person, and the examiner has no means of telling which is which.

I believe the institutions that hold their standards through this decade will assess the process of making an answer, not only the answer itself. Oral defence will be one instrument among several. It cannot be the only one, and it cannot be the lie detector.

Proof of Work, and Why the Degree Still Opens the Door

If the memory test is failing, what replaces it? Proof of work is a credential assembled from evidence of doing, gathered over time and checkable by someone other than the learner: a public portfolio, an oral defence of a project, a live piece of work with real users, a reproducible process log that shows the wrong turns as well as the right ones, and an apprenticeship record signed by the person who supervised the work.

Who recognises such a credential? Here the data is less comforting. A 2026 peer-reviewed paper by Xiang Wenjing, Hana Marešová and Shamshad Khan, published in Frontiers in Computer Science, concludes that "employer recognition is highly variable, with greater recognition in technical fields than soft or cross-disciplinary skill areas." Frontiers in Computer Science

That is the first surprise of proof of work. The paper studies micro-credentials, not portfolios, so its finding is a warning I extend to portfolios by inference. A portfolio of code is easy to recognise. A portfolio of a community health survey, or of a negotiated settlement between two village committees, is harder for an employer to read, and so it is easier to ignore.

The second surprise is older. A 2024 report by Harvard Business School and the Burning Glass Institute, "Skills-Based Hiring: The Long Road from Pronouncements to Practice," found, as HR Dive reported, that about 45% of companies that announced changes had made them "in name only," with "no meaningful difference in hiring behavior." Roles dropping degree requirements, HR Dive reported, grew nearly fourfold between 2014 and 2023, yet fewer than one in 700 new hires benefited from the change.

This is the image I keep returning to: the paper has changed, and the room has not. A proof-of-work credential that nobody reads is only a more elaborate certificate.

Kenya's Schools Have Already Moved

The most useful facts in this essay do not come from a university lecture hall. They come from a Kenyan school calendar. Arena News reported on 10 May 2026 on the Kenya National Examinations Council's Senior School roadmap. Grade 10 school-based assessment counts for 15% of the final mark, "every project, practical, and written test now carries weight," and the learner is registered under the Assessment Number acquired in Grade 3. Projects and practicals run from May to September 2026, and written tests in Term III end on 15 October 2026. The same article says 30% of the final mark comes from internal school assessments. Arena News

Kenyans.co.ke reported on 11 May 2026 that KNEC had told principals to register as Senior School assessment centres, and that registration was "mandatory for schools preparing to admit and assess Grade 10 learners." Kenyans.co.ke

A Kenyan child is already carrying a thread I would call proof of work, though so far it is only an identifier. The Assessment Number issued in Grade 3 follows the child forward through the new system. What evidence goes into that record, how long it is kept, and who may read it: those questions are still open, and they are the whole argument of this essay.

The Standard reported that Kenya created 822,100 new jobs in 2025, according to the Kenya National Bureau of Statistics, and roughly nine in ten were informal. The share of formal employment fell from 18.5% in 2010 to 15.5% in 2024. In the article's words, "while more young people are earning degrees, fewer are finding opportunities in stable, salaried jobs," and "academic qualifications alone no longer guarantee economic security." The Standard

Across the continent, the World Data Lab and Mastercard Foundation's Africa Youth Employment Outlook 2026, published in February, estimates that about 57% of young Africans, some 304 million people, were working as of 2025. About 90% of employed young Africans work in informal jobs, and only 9% have completed tertiary education, roughly one young person in eleven. In Algeria, Egypt and Morocco, "nearly 30% of university graduates remain unemployed or inactive." World Data Lab

A Kenyan Secondary School in 2028

The design in this section is a Pawanax proposal. It is judgment, not a description of a school that already runs it.

Imagine a Grade 12 learner in Nakuru sitting at a school desk. She has spent a term building a sensor that warns a village when its water pump's battery is failing. Her panel asks her, "Why this battery?" She has an answer. She also has a process log showing the three batteries she rejected and the reason each one failed her tests. The panel asks to see the log, and she scrolls to the second week, where her first model was wrong. (Imagined.) That is the viva I want in 2028.

The design has six parts.

First, one portfolio per learner, keyed to the Assessment Number from Grade 3. Projects, practicals, written tests and process logs sit in one record across Grades 10 to 12. The learner decides what becomes public. Private evidence stays with the school.

Second, reproducible process logs. Every project is kept in a version history: notes, failed attempts, corrected data. Marks reward the reasoning shown in the wrong turns as much as the finished product, and a panel must be able to rerun the work.

Third, one termly oral defence of a single project before a panel of three: a teacher, an examiner from another school or the county, and a parent or employer. The script is fixed in advance. Each candidate gets the same opening prompt, the same follow-up prompts, a published rubric and a defined time. That follows Brownlie's advice for a short, consistent conversation with clear criteria, and it is how the Norwegian spread, from fewer than ten questions to nearly fifty, stops becoming the grade.

Fourth, an apprenticeship record. A learner who does supervised work at a local firm, a clinic, a cooperative or a county youth programme receives a signed record of what she did, with the supervisor's name and a plain description of the tasks. The credibility comes from the signer, not from the school.

Fifth, accommodations built into the design. A learner may defend in the language she thinks in. A learner who freezes may take a written defence with a second examiner present. The German study found more anxiety in oral exams than in written ones, and the Norwegian figures show how widely oral conversations vary in length and number of questions. Neither anxiety nor the luck of the draw measures mastery, so no learner should be graded on either.

Sixth, a credential that says what the learner can do, with links to the evidence. Not "Grade B in Biology," but a plain statement: built a water-level sensor, defended its design choices, reran the test on a second dataset, and here is the log.

I believe the cost is smaller than people fear. The 60 person-hours figure below comes from an imagined school, not a costing. Imagine a school with 120 Grade 10 learners and a ten-minute defence each: one round is 20 hours of examining time, or 60 person-hours across a panel of three, before preparation. Marc Watkins of the University of Mississippi said, "I can't even fathom the cost this will have for a single university", but that worry concerns university lecture halls. I believe the harder cost in a secondary school is moderation: getting two panels to grade the same work the same way. I would publish the moderation results every term so that anyone can check them.

What the Child Carries

Imagine a card with a string of digits on it, issued in Grade 3 and carried for nine years. It is a thread, not a grade. In 2028 it could carry evidence along its length: a poem read aloud at assembly, a pump that keeps running, an apprenticeship signed by a mechanic, a log in which a seventeen-year-old admits that her first battery was wrong. None of that can be memorised the night before. All of it can be checked.

The memory test asked what you could remember. The proof-of-work test asks what you can show, and how you found your way back from the mistake.

Go back to the examination hall. The phone is still glowing under the desk, and the proctor's laptop is still open at the lectern. The question in that room was never whether a machine helped with the answer. Machines will always help. The question is whether the child can show the work, and explain the wrong turn in her own voice.

Ask the child to show the work. Then listen to how she explains the mistake.

Editorial note: This essay expresses the founder’s ideas, developed with AI-assisted research and writing (parallel Claude Haiku 5.5 agents) and checked against the linked sources on 30 October 2026. Convictions, proposals and forecasts are identified in the text; scenes framed as imagined are illustrative.

Cover photograph / illustration: Alison wood (Wikimedia Commons user). CC BY 3.0. Original source. Illustrative image: a student writing an examination by hand in a classroom. Photograph: Alison wood, via Wikimedia Commons. CC BY 3.0. This image does not document the events described.