TL;DR: Competency-based transcripts are gaining mainstream acceptance, with 822 colleges and universities now receiving them, including Harvard and MIT. At the same time, AI makes it possible to generate a compelling competency narrative quickly and in volume. The more complete and well-packaged the write-up, the harder it becomes to judge its authenticity. The trust foundation for proving competence is changing: what matters is no longer just the final product, but timestamps, process, and accumulated density. A learning record built over ten years, one that preserves failures and revisions, is evidence that is nearly impossible to reconstruct after the fact.

This year, any high school student can use AI to produce a complete learning portfolio in an afternoon: a personal statement, learning reflections, portfolio write-ups, sincere in tone, coherent in structure.

That is already the reality. So a question now falls on every reviewer: when every submission can look this good, which one do you trust?

In “When Assignments Can Be Generated, Why Real Tasks Matter More,” I wrote about the collapse of trust around schoolwork. This piece goes one level further. After assignments come résumés. Learning portfolios, application materials, and project collections, these documents that serve as “proof of competence” are going through the same trust crisis, with higher stakes, because they are the tickets to college admission and employment.

Beyond Transcripts: What Are Universities Already Accepting?

Start with the movement side. No one understands the limits of traditional transcripts better than admissions offices: grade inflation has made scores harder to compare, and a row of numbers says nothing about what a student can actually do. Over the past several years, a competency-based transcript movement has emerged in the United States, using portfolios and evidence of competence to show what students can genuinely perform.

The degree of institutionalization may surprise you. The Mastery Transcript Consortium (MTC, now part of ETS) reports on its official list that as of July 2026, 822 colleges and universities have received applicants who submitted competency transcripts or learning records. Harvard, Stanford, and MIT are all on it. This direction will be familiar to readers in Taiwan: the learning portfolio under Taiwan’s 108 Curriculum Reform is the local expression of the same global shift.

The direction is right. The timing is the problem. Just as competency-based evidence systems are taking off, AI has made it possible to generate the “evidence” itself.

Why a More Polished Portfolio Raises More Suspicion

This is the trust asymmetry. The cost of producing a competency narrative has collapsed to nearly zero. The cost of verifying one has not dropped at all.

A reviewer has no tool to tell whether a reflection was written by a seventeen-year-old at midnight or generated by a model in three seconds. The only rational response is to discount everything that looks too polished. The paradox follows: the more perfectly you refine your portfolio, the less it functions as proof of competence. Polish has gone from an asset to a red flag.

I wrote about something similar in “AI Can Hit the Average, but It Cannot Fake You.” Applied to proof of competence, there is a harder version of the same point: AI can generate a final product. It cannot generate the process behind one.

Timestamps and Process Density: Two Things AI Cannot Fabricate

What can a reviewer still trust? My answer comes from somewhere unglamorous: a rough, scribbled record from 2015 paired with a results report from 2020 is harder to fake than any polished portfolio produced in 2026.

The first thing is the timestamp. AI can generate any content today, but it cannot place that content ten years ago. A record that spans multiple years, stored on a third-party platform, carries a date on every entry that cannot be reconstructed after the fact. To fake it, you would have had to start laying the groundwork when the child was six years old. At that point, it is not fabrication anymore. It is education.

The second thing is process density. A genuine learning record does not look presentable. It has typos. It has projects abandoned halfway through. It has contradictions between what the child thought at different points, a teacher’s feedback, and the child’s awkward response to it. These imperfect details are precisely where credibility lives. Generating a polished final product is straightforward. Generating ten years of failures, detours, and revisions that cross-reference each other and align with real-world events is an entirely different order of difficulty.

Ten Years of Seesaw Records

Our family has actually walked this path.

The students at BTS kept their learning records on Seesaw: everyday work, videos of oral presentations, teachers’ feedback at the time, stored continuously for ten years. When those records were made, nobody thought “this will be evidence someday.” It was simply part of the educational process. When I was organizing a decade’s worth of files to write “Flip to Climb,” I saw it more clearly: those 2,300-plus files are themselves a database of learning. And because these records were kept all along, in 2026 I can still call them up, organize them, and see in them the through-line this decade accumulated.

When our child went through the special-admissions process, the preparation looked nothing like what most people imagine. The work before the application was curation, not creation: going back through the records to select, then translating unconventional learning into a language the institution could read. What the review committee saw in those materials was not “here is what he says he can do.” It was “here is when he started, where he got stuck, and how things changed.” The timestamps and process density were already there. There was nothing to argue for. You opened the record and it spoke for itself.

Where to Start, If You Haven’t Already

Three things for parents who still have time.

First: start recording today, on a platform with timestamps, and do not reconstruct the past. Anything reconstructed after the fact looks the same to a reviewer as something AI-generated.

Second: record what is real, including the failures and detours. Process density comes from imperfection. Save the polishing for the final stretch before an application.

Third: do not educate for the sake of building evidence. That record was credible precisely because it was a byproduct of education. Reverse the order and it becomes hollow.

The most valuable evidence in the age of AI is evidence that cannot be generated after the fact. The page of scribbles your child makes today will carry more weight ten years from now than any polished personal statement written on deadline. Keep it, and it becomes an archive that belongs to him for life.