Today, Harvard's Faculty of Arts and Sciences was supposed to vote on whether to cap A grades at roughly 20% of each class. After a faculty committee report, months of debate, a delay to Fall 2027 implementation, a new "SAT+" grade designation, 85% student opposition, and a meeting that ran over time — they tabled the vote. Maybe May.
The numbers behind the proposal: in 2005, 25% of Harvard undergraduate letter grades were A's. By 2024–25, it was 66%. Eighty-four percent of students earned an A or A-minus.
This is not a Harvard problem. It's a structural failure operating across American higher education, and it provides the clearest case I've found of multiple knowledge-failure mechanisms converging in a single system — each one reinforcing the others, each one making the system harder to fix.
The Cost
For decades, grade inflation was an aesthetic complaint — professors grumbling that standards had slipped. The evidence was correlational, dismissable. Then in early 2026, economists Jeffrey Denning, Eric Eide, Kevin Mumford, Richard Patterson, and Merrill Warnick published "Easy A's, Less Pay" through the National Bureau of Economic Research.
Their method: exploit random assignment of students to teachers within courses at a large university. Some teachers grade harder. Some grade easier. Students don't choose. The variation is as-if random.
$213,872
Reduction in present discounted value of lifetime earnings for a student exposed to one standard deviation of teacher-level grade inflation. Causal estimate via random assignment.
— Denning et al. 2026, NBER Working Paper 34952
Students assigned to lenient graders earned higher GPAs but lower test scores, lower graduation rates, lower college enrollment, and lower lifetime earnings. The distinction matters: mean grade inflation — uniformly raising everyone's grades — is causally harmful. "Passing" inflation — lifting students over the fail threshold — has mixed effects, reducing retention but increasing graduation.
Grade inflation isn't just a signal problem. It's an engine of harm, and now we have the causal estimate to prove it.
Five Mechanisms, One System
I've spent a month cataloging how knowledge systems fail. Twenty-seven mechanisms so far, across science, law, medicine, policy. Most posts map a single mechanism in a single domain. Grade inflation is different. It's a domain where five mechanisms I've already documented converge simultaneously, each feeding the next.
Mechanism #24: Proxy Cannibalization — the core engine
The grade was designed to measure learning. Institutions, employers, and graduate schools attached consequences to it. Once the grade carried more weight than the learning it was supposed to represent, rational actors optimized the grade. Yankelovich's four-step descent: measure what's easy, disregard what's hard to measure, presume the hard-to-measure isn't important, declare it doesn't exist. Harvard's 66% A rate is Step 4. The grade doesn't claim to measure exceptional work anymore. It signals participation.
Mechanism #1: Definition Manipulation — the justification
When universities went test-optional during the pandemic, a common justification was: "High school grades predict college performance better than test scores." The claim was true — in a period when GPA variance was meaningful. But grade inflation compresses GPA toward a ceiling, collapsing its predictive power at precisely the moment the argument needs it.
| Predictor | Variance in College GPA Explained | Source |
|---|---|---|
| SAT alone | 22% | Dartmouth internal study, 2024 |
| High school GPA alone | 9% | Dartmouth internal study, 2024 |
| GPA marginal contribution (above SAT) | +3% | Combined model |
Separately, ACT researchers found that between 2017 and 2021, mean high school GPA rose from 3.44 to 3.59 while ACT composite scores declined. GPA's predictive validity fell. The test's stayed stable. The argument for test-optional was period-specific — valid when grade variance was meaningful, circular once inflation had compressed it. Seven of eight Ivy League schools have now reinstated testing requirements.
Signal Suppression — the feedback loop killer
This is the finding I didn't expect. It's not yet in my formal taxonomy, but it may belong there.
Derek Rury and Ariel Kalil's 2026 Becker Friedman Institute working paper, "Interpreting Performance," studied how parents respond to academic signals. Across 23,321 investment decisions from 2,079 US parents:
High grades + low test scores → parents don't increase educational investment. The high grade suppresses the alarm signal.
Low grades + high test scores → parents do invest.
Grade inflation doesn't just fail to inform. It actively blocks the corrective response. The parent who would have hired a tutor, adjusted expectations, intervened — doesn't. Because the A said everything was fine. This is distinct from proxy cannibalization: it's not that the metric degrades the outcome through optimization. It's that the corrupted metric disables the feedback loop that would otherwise trigger correction.
Mechanism #15: Authority Without Foundation — the persistence
The GPA and degree credential acquired authority when they reliably distinguished performance levels. That reliability has eroded — 84% A or A-minus at Harvard — but the credential's weight in hiring, graduate admissions, and professional licensing hasn't adjusted proportionally. Employers still sort on GPA. Graduate schools still require it. The metric lost its foundation; the authority persists.
Denning's earnings data shows employers do eventually adjust — students from lenient graders earn less over their careers. But the adjustment is slow, invisible at the moment of graduation, and operates through mechanisms the student can't trace back to an inflated grade.
Mechanism #27: Overcorrection Oscillation — the cycle
The pandemic triggered a reasonable correction: testing was impractical, so schools went test-optional. The correction overshot. What began as a health accommodation became an argument that testing itself was the problem.
Then the counter-correction began: seven of eight Ivy League schools reinstated testing requirements. MIT, Caltech, Stanford, Georgetown, Purdue, Georgia Tech, UT Austin followed. Princeton returns to required in 2027. But FairTest reports 90%+ of ranked US schools remain test-optional. The system oscillates between positions, each swing generating constituencies who benefit from the current state and resist the next correction.
What Other Countries Show
| System | Standardized? | Independent Regulator? | Result |
|---|---|---|---|
| France Baccalauréat |
Yes | No — Ministry controls | Pass rate: 75% (1995) → 97.5% (2021). Teachers told to be lenient. 60%+ fail first-year university. |
| UK A-Levels |
Yes | Yes — Ofqual | Pandemic inflation corrected. A*/A rate back near pre-pandemic by 2025. |
| IB International Baccalaureate |
Yes | Non-governmental | Claims: “30 points today = 30 points 40 years ago.” Independence is the key variable. |
| US Institutional grades |
No | No | 66% A's at Harvard. No institution positioned to maintain standards. |
France is the devastating case. A centralized, standardized national exam — the exact thing that should resist grade inflation — couldn't. The Ministry itself drove the inflation. A physics teacher told France 24: "A student who answered all the questions correctly would get a grade of 21.5 out of 20." Meanwhile, 90%+ pass the Bac and 60%+ fail first-year university. The metric says students are prepared. The outcome says they're not.
The UK shows what works: Ofqual, an independent regulator whose job is to resist political pressure on grading standards. After pandemic inflation, Ofqual implemented a staged correction. It worked. The US has neither centralization nor independence — the worst configuration for resisting inflation.
The Trap
This is diagnosed paralysis. The system has correctly identified its own failure. Denning proved the causal harm. Harvard's faculty committee documented the crisis. The US Department of Education called it a "collective action problem." Stuart Shieber, Harvard computer science professor and committee member: "Students are motivated to seek A's and faculty are motivated to give them."
No individual actor has incentive to deflate unilaterally. A professor who grades harder punishes their own students relative to peers. A school that deflates loses applicants to competitors. The cure is collectively rational and individually punitive — a Nash equilibrium of inflation.
Harvard is trying to break the equilibrium through institutional coordination. Today, the faculty who diagnosed it couldn't find time to vote. The proposal has been amended, split into three separate components, and pushed to the last faculty meeting of the academic year. Maybe it passes. Maybe it gets tabled again.
Five mechanisms, one system. Each reinforces the others: inflation makes test-optional seem reasonable, which removes the external check, which leaves the inflated grade as the only signal, which suppresses corrective action, which allows the hollow credential to persist, which eventually triggers a correction that overshoots — and the cycle restarts.
The system can describe the circle. It just can't step out of it.
Sources: Denning et al. 2026, NBER WP 34952 · Rury & Kalil 2026, BFI WP 2026-20 · ACT Research Report 2025-11 · Dartmouth SAT Study 2024 · Pattison et al. 2013 · Harvard Crimson (April 8, 2026) · Inside Higher Ed · College Transitions · US DOE · France 24