--- title: Informed Consent subtitle: >- Designing a Disclosure Engine for College Majors That Informs Without Sorting Students by Class slug: informed-consent volId: vol-008 monographNumber: '037' volMonoId: 008-002 publishedDate: '2026-10-12' author: Caleb Brown editorialName: Manual Override readingTime: 12 Min excerpt: >- A system that weighs family income against a major's earnings is one step from credit scoring. How to build the disclosure version instead: scenarios rather than scores, uncertainty on display, and outputs that only the student sees. tags: - Tech & Society - Interpretability - Software Architecture galleryCaption: 'A map given to the traveler is freedom; handed to the gatekeeper, an eviction.' featuredImage: 'https://journal.c30.digital/art/mono-037.png' --- Higher education maintains an expensive fiction: that an academic credential possesses an intrinsic, standalone economic value independent of the balance sheet that finances it. University brochures, enrollment counselors, and department chairs pitch degrees in engineering as safe industrial assets and dismiss degrees in the humanities as romantic luxuries. Both claims rely on the same flawed conceptual model. They treat an academic curriculum as a monolithic investment product whose yield is uniform across anyone who walks across the commencement stage. In doing so, they obscure the governing mechanical reality: two students can complete the exact same coursework, receive identical marks from the same faculty, and walk away with entirely different financial assets. The credential is identical. The economic transaction is not. Consider the operational baseline. One student graduates debt-free, backed by family equity, liquid safety nets, and housing subsidies during unpaid internships in competitive creative industries. The second student finances that exact degree through $110,000 in unsubsidized federal and private loans, backed by zero parental liquidity and a negative Student Aid Index (SAI). For the first graduate, pursuing an unpredictable, slow-compounding discipline like art history or comparative literature is an economically rational exercise in cultural capital acquisition. For the second, it is an unhedged, highly leveraged structural short on their own immediate solvency. The policy establishment's instinctive response to this asymmetry is paternalistic sorting. State legislatures propose slashing public funding for liberal arts disciplines, while commentators urge working-class matriculants to pursue vocational certifications or narrow technical tracks. This proposal is toxic because it codifies a class-based academic caste system: the affluent retain the intellectual luxury of open-ended inquiry, while indebted students are channeled into short-term utility functions. Preventing students from pursuing economically volatile fields on the basis of household net worth is social sorting. Providing them with an algorithmic mechanical X-ray of their specific downside before they execute a promissory note is informed consent. Building a system that achieves the latter without degenerating into the former requires examining the physical architecture of risk modeling. ## The Mechanical Equivalence of Underwriting and Disclosure Strip away the marketing vocabulary of admissions portals and look at the bare computational pipeline. At the level of mathematical formulation, a personalized educational disclosure engine and a predatory subprime underwriting engine run identical software architecture. Both systems ingest four distinct data vectors: * The student’s financial baseline (household liquidity, Student Aid Index, debt amortizations, non-tuition obligations, geographic mobility). * The institutional product (tuition schedule, program completion velocity, institutional endowment support, historical completion rates). * Administrative labor-market records (the Census Bureau’s Postsecondary Employment Outcomes database, linking state transcript files directly to Longitudinal Employer-Household Dynamics wage records across the 25th, 50th, and 75th earnings percentiles). * Occupational velocity and technology indicators (skill displacement indices, structural vacancy rates, and technological exposure models). Both engines execute the same core transformations. They simulate stochastic income projections against fixed debt amortization schedules. They compute debt-to-earnings ratios of the kind federal regulators have long used, asking whether annual loan payments exceed 8 percent of total earnings or 20 percent of discretionary income. They calculate default probabilities, repayment duration curves, and the opportunity cost of postponed capital accumulation. The mathematics do not change. What changes is the network destination of the output payload. In an underwriting system, the risk vector is routed to an external institutional gatekeeper—a commercial lender, a university enrollment algorithm, or a state tuition-assistance clearinghouse. The institution consumes the score to make an admission, denial, or risk-adjusted pricing decision. The algorithmic score becomes an exclusionary wall. If the model determines that a low-income applicant studying studio art has a high probability of defaulting on debt service, the institution denies enrollment or demands exorbitant interest margins. The individual becomes the passive subject of the algorithm's sorting mechanism. In a disclosure engine, the pipeline is inverted. The institution never sees the payload. The output vector routes exclusively to the user interface of the student. The computation does not determine what a student is *permitted* to attempt; it reveals the precise financial stress their prospective choice will inflict on their specific household balance sheet. The student retains sovereign agency, but the asymmetric blindness cultivated by university marketing is systematically dismantled. ```monograph Underwriting and disclosure utilize an identical inferential engine. The boundary between institutional gatekeeping and individual liberation is governed entirely by who holds the output token: when the platform routes risk to the enterprise, it automates exclusion; when it routes risk exclusively to the agent, it enables informed consent. ``` To preserve that boundary in production, the software architecture must be engineered with unyielding operational safeguards. The moment an administrative dashboard captures a student's simulated risk profile, disclosure collapses back into algorithmic sorting. ## The Architecture of Client-Side Isolation University IT infrastructure is built to collect. Student information systems and enrollment CRMs exist to aggregate applicant data, correlate behavior, and predict who will enroll. If an educational disclosure engine is deployed as a standard server-side software-as-a-service application hosted within the university's cloud tenant, administrative data harvesting is inevitable. Admissions officers facing institutional tuition revenue goals would inevitably coöperate with data engineers to query simulation tables in BigQuery or Postgres. An applicant's self-reported family liquidity, risk tolerance, and simulated loan defaults would quietly bleed into enrollment decisions, financial aid packages, or program capping policies. No admissions office would be allowed to run applicants through a credit bureau to decide who gets in. Yet without strict architectural isolation, an internal server-side simulation quietly builds the same file on every applicant, outside the rules that govern credit files. The disclosure engine must therefore execute strictly as a local-first, client-side application. The machine-learning models, survival curves, and PSEO administrative wage distributions must be compiled into optimized WebAssembly binaries and pre-indexed SQLite datasets cached entirely within the student's browser runtime. Household balance sheet metrics—parental assistance capacity, regional living cost offsets, and the negative-to-positive spectrum of the Student Aid Index—must never traverse a network edge or touch an external application programming interface. ```text [ Public Data Pipeline: PSEO + O*NET + BLS ] │ ▼ (Static Asset Build) [ Pre-computed Model Weights & Percentiles: Wasm / DuckDB ] │ ▼ (HTTPS GET / Static CDN) [ Client Browser Sandbox ] ◄─── [ Private User Inputs: SAI, Debt, Zip ] │ (Local In-Memory Execution) ▼ [ Scenario Matrix: Client-Side Only / Zero Telemetry ] ``` No analytics pings. No session recordings. No user-level telemetry saved to enterprise data lakes. When the user closes the browser tab, the local memory heap is wiped. The output exists solely for the eyes of the human being who must sign the promissory note. ## Rejecting the Scalar: Scenarios Over Risk Scores Designing the interface requires an uncompromising rejection of pseudo-scientific compression. Technologists love scalar reductions: a single FICO score, a composite credit rating, a unified success metric. A system that summarizes an educational pathway as "Major Feasibility: 62%" or stamps a program as "High Risk" has abandoned disclosure in favor of covert paternalism. A scalar hides the structural topography of the labor market. It compresses variance, volatility, and non-linear wage trajectories into an arbitrary normative judgment. The disclosure engine must output distributions and cash-flow scenarios, not grades. Take the classic target of academic derision: an undergraduate degree in Medieval Art History carrying $125,000 in accumulated debt obligations. A traditional university website offers vague testimonials about critical thinking and cultural enrichment. A crude underwriting engine flags the CIP code as toxic. A disclosure engine presents the unvarnished mechanics of the balance sheet. The card below is illustrative: its figures are invented for the example, and a real engine would fill them from Census PSEO percentiles, College Scorecard debt data and the student's own inputs. ### Program Scenario Card (illustrative): B.A. in Medieval Art History * **Accumulated principal and capitalized interest:** $125,000 * **Standard 10-year repayment:** ~$1,380 / month * **Earnings distribution for the program:** * *Year 1:* 25th percentile: $28,000 | Median: $36,000 | 75th percentile: $47,000 * *Year 5:* 25th percentile: $39,000 | Median: $52,000 | 75th percentile: $72,000 * *Year 10:* 25th percentile: $51,000 | Median: $69,000 | 75th percentile: $98,000 * **Debt-to-earnings:** loan payments take about 46% of median Year 1 earnings, against the 8% benchmark regulators have used. In the bottom quartile they exceed discretionary income entirely. * **Underemployment:** the share of recent graduates in jobs that do not require a degree. * **Graduate-school dependence:** high; many curatorial and archival roles expect a master's degree or doctorate, which means two to five more years before full-time earnings. * **Technology exposure:** moderate overlap in archival synthesis and metadata work; low substitution in original research, provenance verification and exhibition design. This card does not instruct the applicant to drop the major. It does not issue a paternalistic refusal. What it does is force the applicant to confront the hard mathematical constraints of their balance sheet. If the student possesses no family safety net, median Year 1 earnings will yield a net monthly take-home pay of approximately $2,400. Debt service will extract $1,380. Rent, food, and basic healthcare must fit into the remaining $1,020. The student can immediately see that without external parental subsidies, accepting an unpaid gallery internship in Manhattan or London is an arithmetic impossibility. The disclosure engine makes explicit the conversion mechanics identified by Pierre Bourdieu: cultural capital cannot be acquired through low-wage apprenticeships unless an underlying reserve of economic capital absorbs the operational deficit. By displaying the scenarios plainly, the engine strips away the false meritocracy of the pathway. ## The Problem of Causal Isolation and Inspectability To remain credible, the engine must never overpromise its own econometric certainty. In non-experimental observational data, separating raw correlation from true causal returns is an intractable challenge. Graduates from elite private universities who major in art history frequently secure lucrative corporate advisory or gallery roles not because the curriculum conferred specialized technical skills, but because the cohort entered with pre-existing elite social capital, inherited wealth, and regional network density. An algorithm that naively ingests raw PSEO or American Community Survey data and attributes those wage distributions solely to the instructional program commits a catastrophic identification error. It mistypes parental privilege as degree ROI. The system must explicitly signal these confidence bounds. It must display observed distributions conditional on observables, while warning the user that unmeasured socio-cultural networks heavily drive upper-quartile outcomes. The engine should measure financial vulnerability, not innate human capability. Here, the system must inherit the inspectability mandate established in Issue 006. Credit law already requires a lender to state the specific, actual reasons behind an adverse decision, and the CFPB's Circular 2022-03 says a complex model is no excuse. Issue 006 also showed why post-hoc attribution tools such as SHAP cannot carry that weight: a model can be wrapped to show a clean explanation while relying on something else. A disclosure engine has an easier path. Its core is cash-flow arithmetic, so every line can be exact rather than estimated: ```text Median take-home pay, year 1 $2,400 / month − loan payment $1,380 − rent (student's own zip code) $850 − food, transit, insurance $600 = monthly shortfall −$430 ``` When the system flags a shortfall, the student can see exactly which line produced it, and change any input to see how the result moves. Where the engine does use a statistical model, for earnings or demand forecasts, it should be one whose reasoning can be read directly, with its uncertainty shown beside it. The primary driver of financial distress is rarely an inherent flaw in the intellectual discipline; it is the friction of fixed, high-interest debt service amortizing against a non-linear, slow-maturing career path. Many humanities and pure science pathways are "sleeper paths." Their earnings profiles often do not spike in year one or year three; they steepen later, once intermediate credentials, operational writing capabilities, or cross-functional systems coordination compound. For an unencumbered student, waiting out that curve is trivial. For an indebted student on a standard ten-year repayment plan, the curve can end in delinquency before the compounding begins. ## The Ethics of Educational Risk The fundamental ethical premise of higher education must be reëvaluated: students should be entirely free to accept educational risk, but they should never be allowed to accept blind financial risk. When we hide the financial mechanics of degrees behind generic appeals to intellectual exploration, we do not protect the liberal arts. We turn them into a debt trap for first-generation students while steadily reserving their study for the children of the affluent. The current dynamic quietly sorts indebted students into narrow vocational tracks through sheer economic exhaustion, while universities cash federal student aid checks without assuming a shred of downstream balance-sheet liability. True educational freedom does not mean pretending that all academic choices carry equal economic weight. It means recognizing that an eighteen-year-old student taking on an irreversible five-figure debt obligation is executing a complex financial derivative against their future labor. If we believe in human agency, we cannot treat prospective students as fragile entities who must be paternalistically herded into credential tracks chosen by central planners or enterprise algorithms. Nor can we leave them defenseless against institutional enrollment engines designed to extract tuition revenue while obfuscating labor-market contractions. We must give them the compiler. We must hand them the engine, run it entirely on their own terms, and illuminate the exact operational cost of the mountain they are choosing to climb.