The email arrived at 4:47 on a Thursday afternoon, which Sarah Chen would later remember as the exact moment when abstraction became her problem. The subject line read "URGENT: Sandel lecture fallout" and the body contained a single sentence from the university's Vice Provost for Enrollment: "We need to talk about the system."
She had not attended Michael Sandel's lecture the previous evening, though she had seen the promotional materials plastered across campus—that leonine face, the promise of "provocative questions about merit and justice in higher education." Sarah had been, instead, in a windowless conference room on the third floor of the administrative building, listening to the admissions software vendor explain why their decade-old scoring algorithm could not, in fact, produce an audit trail for any decision made before 2019. The vendor representative, a man in his fifties with the weary manner of someone who had watched too many clients discover the technical debt they had inherited, had used the word "opaque" seven times in forty minutes.
Now, apparently, opacity had become a crisis.
There is a particular variety of institutional panic that afflicts universities, one that differs from the blunt urgency of corporate catastrophe. In corporations, disaster announces itself in quarterly earnings, in market share surrendered, in revenue bleeding away like arterial blood. Universities, by contrast, operate on longer horizures of harm—reputation damage that accretes over years, public relations erosions that shift the landscape of who applies and who declines admission, lawsuits that metastasize slowly through the appeals process. The panic, when it comes, arrives not as a heart attack but as the sudden recognition that one has been walking around with a serious illness for years without noticing.
Sandel's lecture had evidently diagnosed such an illness.
Sarah spent the weekend reading. Not the vendor documentation this time, but Sandel's book—The Tyranny of Merit—which she had purchased in hardcover and read with the kind of attention she usually reserved for technical specifications. The argument was compelling in its simplicity: that meritocracy, far from being a cure for inequality, had become a new form of hubris, a way for the successful to believe they deserved their success and for the unsuccessful to internalize their failure. The rhetoric of merit, Sandel argued, allowed the privileged to forget the role of fortune in their achievements—the accident of birth, the lottery of genes, the privilege of circumstance.
What struck Sarah most forcefully was not the philosophical argument itself, which had a certain austere elegance, but rather how precisely it mapped onto the system she had been tasked with replacing. The old admissions algorithm was a black box, literally—a proprietary scoring mechanism that ingested dozens of variables and produced a single number between 0 and 100, which the admissions committee could then use to sort applicants into rough tiers. No one could explain why one applicant received an 87 and another an 86. The system was, in this sense, a perfect metaphor for the kind of opaque meritocracy Sandel critiqued: it rendered judgment, it ranked and sorted, but it could not be interrogated or questioned. It simply was.
The stakeholder interviews began on Monday. Sarah had requested meetings with representatives from four constituencies: faculty, admissions staff, legal counsel, and the alumni relations office. She had expected disagreement—universities were, after all, ecosystems of competing interests—but she had not anticipated the depth of the contradictions.
The faculty representative, a philosophy professor named Dr. Whitmore, insisted that any new system must preserve "holistic evaluation." By this he meant, Sarah gradually understood, the preservation of faculty discretion to champion unusual candidates—the student who had published poetry in obscure journals, the applicant who had spent a gap year studying Akkadian. These were not accomplishments that registered well in quantified metrics, Dr. Whitmore explained, but they spoke to a certain quality of mind, a seriousness of intellectual purpose.
The legal counsel, by contrast, wanted everything quantified and documented. Every decision needed a paper trail, every score needed to be decomposable into its constituent parts, every exception needed to be justified in writing and approved by a committee. This was not, the counsel explained with the careful diction of someone who had spent decades translating between law and institutional practice, because the university was necessarily doing anything wrong, but because they needed to be able to demonstrate, in court if necessary, that they were doing everything right.
The admissions staff wanted something user-friendly, something that would not add hours to their already Sisyphean workload. The director of undergraduate admissions, a woman who had been reading applications for thirty years and who possessed the kind of exhausted competence that comes from managing impossible volume, pointed out that they reviewed forty thousand applications in four months. Any system that required more than two minutes per file would create a bottleneck that would delay decision letters and anger applicants.
And then there was the alumni relations representative.
This was where Sarah first encountered what she would later call, in her private notes, "the legacy problem." The representative, a polished man in his early forties who had the unnerving habit of referring to donors by their first names, explained that certain applicants required "special consideration." These were the children of major donors, the offspring of trustees, the relatives of distinguished alumni. It was important, he said, that the new system allow for these considerations to be taken into account.
Sarah asked how this differed from the old system.
The representative smiled in a way that suggested he was accustomed to this question. The old system, he explained, had certain informal mechanisms. A note in the file, a flag in the database, a phone call from the development office to admissions. These mechanisms were, he acknowledged, not particularly transparent, but they had the advantage of being discreet.
Sarah understood immediately what he was saying: the opacity of the old system had been a feature, not a bug. It had allowed the university to practice legacy preferences while maintaining plausible deniability about the extent of those preferences. A transparent system would make such preferences visible, and visibility would create pressure to eliminate them or, at minimum, to justify them more rigorously.
There is a certain clarity that comes from recognizing that one's technical problem is actually a political problem in disguise. Sarah had been hired as a Product Manager, which suggested a domain of clean requirements and technical solutions. But what she was really being asked to do was to make visible a set of decisions that the university had, for decades, preferred to keep obscured. The question was not whether the new system could be made transparent and auditable—that was technically straightforward. The question was whether the university actually wanted transparency and auditability, or merely the appearance of such.
The first principle of the new system, Sarah decided, would be decomposability. Every score would be built from discrete, explicable components. If an applicant received a 7 out of 10 on academic preparation, there would be a clear formula showing how that 7 was calculated: high school GPA weighted at 40%, rigor of curriculum at 30%, standardized test scores at 20%, teacher recommendations at 10%. Nothing would be hidden in proprietary algorithms or inscrutable machine learning models.
This decision immediately created problems.
The data science team wanted to use more sophisticated models. Machine learning, they argued, could identify patterns that simple weighted averages missed—subtle correlations between seemingly unrelated factors, nonlinear relationships that traditional statistics could not capture. One data scientist, a recent PhD who had published papers on fairness in algorithmic decision-making, suggested that they could train a model on historical admissions decisions and then use techniques from interpretable AI to explain individual predictions.
Sarah vetoed this approach, though she understood its appeal. The problem was not technical but epistemological. A machine learning model, no matter how sophisticated, would learn from historical decisions, and those historical decisions had been made using the opaque system they were trying to replace. Training a new algorithm on old data would risk encoding old biases into new infrastructure. It would be, she thought, like asking an AI to learn how to be fair by studying the judgments of a system that could not explain its own fairness.
Better, she decided, to build something simpler and more legible, even if it meant sacrificing some predictive accuracy.
Feature selection became a political minefield. Every proposed variable had constituencies. The admissions staff wanted to include "demonstrated interest"—had the applicant visited campus, attended information sessions, corresponded with admissions officers? The athletics department wanted special consideration for recruited athletes. The diversity and inclusion office wanted to ensure that first-generation status and socioeconomic background were weighted appropriately. The faculty wanted evidence of intellectual curiosity, though they could not agree on what that evidence might look like.
Sarah spent weeks in spreadsheets, modeling different weighting schemes and testing them against synthetic applicant profiles. She created scenarios: the brilliant but economically disadvantaged student from a failing school district, the wealthy applicant with perfect scores and mediocre recommendations, the athlete with strong academics and genuine intellectual interests, the legacy applicant with good-but-not-great credentials.
The synthetic testing revealed something unsettling. No matter how carefully she calibrated the weights, there was always a way to game the system. If academic preparation was weighted too heavily, it advantaged students from wealthy school districts with inflated grading curves. If extracurriculars were emphasized, it favored students whose families could afford music lessons, travel sports, and summer enrichment programs. If the system tried to correct for these advantages by adding weights for first-generation status or family income, it created perverse incentives for applicants to emphasize hardship narratives, turning genuine struggle into a strategic advantage.
There was, she began to realize, no configuration of weights and factors that would produce outcomes everyone agreed were just. This was not a technical failing but a reflection of deeper disagreements about what justice in admissions meant. Was it about rewarding individual achievement? Correcting for structural inequality? Preserving intellectual community? Building a diverse class? These goals were not necessarily compatible, and any system that tried to optimize for all of them simultaneously would end up satisfying none.
The discovery of the "legacy exceptions" database came in April, five months into the project. Sarah had been conducting data lineage audits—tracing how historical admissions data flowed through various university systems—when she found a table in the enrollment database that was not documented in any of the official schemas. The table was called "dev_office_priority" and it contained a list of applicant IDs along with codes like "L1" and "L2" and "TR."
She asked the database administrator what the codes meant.
The administrator, who had been managing the university's data systems since before Sarah was born, sighed in the manner of someone who had been waiting years for this question. L1 meant "legacy tier one"—child of a donor who had given more than five million dollars. L2 meant "legacy tier two"—child of a donor who had given between one and five million. TR meant "trustee relation."
Sarah asked what these codes did.
They didn't do anything, technically. They just sat there in the database. But the admissions staff had access to the table, and they could see the codes when they reviewed applications. The old system had never formally incorporated these flags into the scoring algorithm. They were just... information. Background context. A nudge.
This was the genius of the old system, Sarah realized. It had never explicitly given bonus points to legacy applicants. It had simply made sure that the right people knew which applicants to look at more carefully, to view more charitably, to find reasons to admit despite borderline scores. The opacity of the algorithm had provided cover for human discretion, and human discretion had provided cover for preferential treatment.
The question now was what to do with this information. Sarah could eliminate the legacy database entirely, make it impossible for admissions staff to see development office flags. But this would enrage the alumni relations office and, by extension, the donors who funded scholarships and endowed professorships and built new dormitories. Or she could make the legacy preferences explicit—add them as a formal factor in the new system, with transparent weights showing exactly how much advantage the child of a major donor received.
Neither option seemed viable. The first would be political suicide. The second would be moral suicide, a public admission that the university was selling admission for money.
Sarah proposed a third approach: an appeals workflow. The new system would generate initial scores based purely on academic and extracurricular factors, with no legacy consideration. But any applicant who was denied admission could file an appeal, and these appeals could consider additional factors—including, though this would not be stated explicitly, development office priorities. The appeals would require written justification, approval from a faculty committee, and permanent documentation in the applicant's file.
This was a compromise in the most precise sense: a solution that satisfied no one completely but that everyone could live with. The transparency advocates could point to the clean initial scoring. The development office could point to the appeals process. The faculty could maintain their role as arbiters of exceptions.
The new system went live in September, just as the admissions cycle for the following year's class began. For the first six weeks, everything appeared to function smoothly. The admissions staff reported that the interface was intuitive, that scores were being generated quickly, that the decomposable components made it easier to explain decisions to colleagues. The legal counsel expressed satisfaction with the audit trails. The faculty committee on admissions voted to commend Sarah's work.
Then the metrics began arriving.
The diversity and inclusion office had insisted on disparate impact monitoring—tracking whether the new system produced demographically different outcomes compared to the old system. The first report, covering the early decision round, showed a 7% decline in admission offers to Black and Latino applicants. The raw scores, it turned out, reflected exactly what one would expect from a system that weighted academic metrics heavily: they favored applicants from well-resourced schools and affluent neighborhoods.
The old system, with all its opacity, had somehow produced more diverse outcomes. This was not because the algorithm was deliberately biased toward diversity—no one knew what the algorithm was biased toward—but because the holistic, human-mediated process had allowed admissions officers to recognize potential in applicants whose quantified metrics were merely good rather than excellent. The transparency that Sarah had built into the new system made such intuitive judgments harder to justify, because they required explicit deviation from the calculated scores.
Sarah found herself in the paradoxical position of having built a more fair system that produced less equitable outcomes. Or perhaps—and this was what kept her awake at night—she had built a system that was more accurately unfair, that made visible the structural inequalities that the old system had obscured.
The alumni response arrived via a series of increasingly angry emails forwarded from the development office. Three families whose children had been denied admission despite legacy status were threatening to withdraw major gifts. One alumnus, whose donation had funded an entire academic building, wrote a letter to the university president suggesting that the new admissions system represented a betrayal of the university's commitment to its "extended family."
Sarah reviewed the three cases. In each instance, the applicant had been denied after initial scoring, filed an appeal, and had that appeal rejected by the faculty committee. The rejections were properly documented, with written justifications explaining that the applicants' academic credentials were substantially below the threshold for admission and that no compelling additional factors warranted exception.
What the documentation did not say, but what Sarah understood from reading between the lines, was that the faculty committee was using the transparency of the new system as a tool to resist development office pressure. Under the old system, legacy admits had been nearly invisible—they appeared in the statistics but could not be individually identified or questioned. Under the new system, every exception required explicit justification and left a paper trail. The faculty, it seemed, had discovered that transparency could be a weapon against the very discretion it was meant to regulate.
By November, Sarah began noticing something troubling in the data. There was a small but persistent pattern of applicants who received initial scores suggesting denial but who were being admitted anyway through channels that bypassed the formal appeals process. When she investigated, she found that admissions staff were creating manual overrides in the system—a feature Sarah had included for technical emergencies, like fixing data entry errors—and using them to admit certain applicants without filing formal appeals.
She requested a meeting with the director of admissions, who arrived looking even more exhausted than usual. The overrides, the director explained, were necessary. There were cases that didn't fit neatly into any category—the international student whose credentials were impossible to evaluate against domestic standards, the applicant with severe learning disabilities whose grades didn't reflect their actual intellectual capacity, the student from a disrupted family situation whose record showed gaps that needed contextual explanation.
Sarah understood what was actually being said: the backchannels had re-emerged. The university had not, after all, wanted transparency. It had wanted the appearance of transparency, a system that could produce audit trails and defend against lawsuits while still preserving the informal mechanisms through which exceptions were made and power was exercised.
The realization should have felt like failure, but it didn't, quite. What Sarah felt instead was a kind of clarity about the limits of technical solutions to political problems. She had built a good system—genuinely transparent, rigorously auditable, methodologically sound. But a good system could not, by itself, resolve underlying conflicts about what admissions should accomplish or whose interests it should serve. It could only make those conflicts more visible, which was its own kind of progress.
On a cold afternoon in December, Sarah sat in her office reviewing the midpoint report she would present to the provost. The new system had processed fifteen thousand applications. It had generated audit trails for every decision. It had survived legal review. It had reduced processing time by 30%. By any technical measure, it was successful.
Whether it was just—that was a different question, one that Sarah no longer believed software could answer. Justice in admissions, like justice anywhere, required not just good procedures but good judgment, not just transparent systems but virtuous people making difficult choices under conditions of uncertainty and competing values. The algorithm could support such judgment, structure it, document it, but it could not replace it.
There was something oddly liberating about accepting this limitation. Sarah had begun the project believing, in the way that Product Managers sometimes do, that the right technical design could solve any problem. She ended it understanding that some problems were not meant to be solved but managed, negotiated, lived with in an ongoing state of provisional settlement.
She thought about Sandel's lecture, the one she hadn't attended, the one that had started all this. His critique of meritocracy had been, she now understood, less about the concept itself than about the hubris of believing we could ever build a truly meritocratic system, one that reliably identified and rewarded genuine desert while factoring out the influence of luck and circumstance. The old system had been opaque, which was bad. The new system was transparent, which was better. But neither could escape the fundamental problem: that merit itself was a contested concept, shaped by privilege, inflected by bias, inseparable from the accidents of birth and fortune that Sandel had identified.
Sarah saved the report, closed her laptop, and looked out the window at the campus below—students crossing the quad in winter coats, their breath visible in the cold air, each one a complex assemblage of achievements and circumstances, each one both deserving and lucky, each one admitted or denied through a process that was now, at least, more honest about its own limitations.
The email notification chimed: another angry alumnus, another threatened withdrawal, another crisis demanding attention. Sarah opened it, read it, and began composing her response. There would be no final resolution, no moment when all stakeholders agreed that the system was perfect. There would only be this: the daily work of trying to be fair in a world that was not, of building tools that acknowledged their own insufficiency, of knowing that transparency was not the same as justice but believing it was, nonetheless, a necessary starting point.
Outside, snow had begun to fall.