Campus hiring, measured.

A résumé is a device for reading a work history, and the average final-year student has none to read. Campus hiring is selection science under its hardest constraints: thousands of applicants, a season compressed into weeks, and files so alike that the usual first sort has nothing to grip. In experienced hiring the file carries the signal and the interview verifies it; on campus that sequence inverts. No panel can interview a placement season's whole applicant pool, and no screening team can credibly rank files that all list the same three internships, so the seat in the interview room has to be earned, and the only currency that scales is measurement.

The scale is what makes the problem scientific rather than clerical. At campus volume every defect compounds by thousands: an unvalidated screen misclassifies at scale, a leaked form invalidates a season's scores, an unmonitored funnel drifts with nobody watching, and the failure is public, because students compare notes across colleges within hours. The same volume is also the opportunity. A funnel this dense generates measurement data no experienced-hire process will ever see, and a program that treats the season as a measurement project keeps that data as calibration.

A funnel built for these constraints is an instrument in its own right: stages chosen the way test items are chosen, for what each adds to the decision; measurement placed first, because measurement is the only stage that scales; the season's measurements kept. That is a different discipline from running the experienced-hire process faster, and it starts with a clear view of what a fresher's file actually contains.

Why every file in the pile reads the same

The résumé-first playbook presumes variance. It works, to the extent it works at all, because experienced candidates differ in employers, tenures, titles, and accomplishments, and a reader can rank those differences. For graduating cohorts the fields collapse: the same degree from the same year, grades compressed toward the top of the scale, coursework shared with everyone else in the department, and internships drawn from the same short list of firms that recruit at the same colleges. The résumé was designed to summarize a career, and these candidates have not had one yet.

Some of that uniformity is structural rather than a failure of candidate effort. Universities standardize curricula precisely so that graduates are comparable; grading policies compress distributions at the top because grades serve certification, not selection; and the internship market is itself downstream of an earlier round of campus selection, since branded internships go disproportionately to students at the colleges those brands already visit. A screener who reads internships as merit is double-counting a filter that already ran. The file is not lying. Almost everything in it was simply produced by systems optimizing for something other than distinguishing these candidates for this role.

The one quantitative field left is grades, and grades are a weaker anchor than they look. They relate to later job performance modestly, and by less than employers tend to assume (Roth, BeVier, Switzer, & Schippmann, 1996). The companion review of résumé screening accuracy carries that evidence in full, including what screeners actually do when they read thin files at speed.

None of that makes the academic record empty. Kuncel, Hezlett, and Ones (2004), synthesizing decades of studies across settings, found that general academic and ability measures predict meaningfully across both academic and work outcomes; a strong transcript is a genuine signal about a general capacity, not just about exam technique. It is one signal, though, from one construct. A selection decision built on it alone inherits both its compression in the file and its silence about everything else a role demands.

What happens in practice, when a screening team must rank thousands of near-identical files anyway, is substitution. The reader falls back on the variance that is available instead of the variance that predicts: campus tier, brand-name internships, formatting fluency, the accident of a familiar surname on a reference. Figure 1 sets the two inventories side by side, the file's fields, which barely vary across the pool, and a battery's constructs, which vary because they are measured at the person, not the cohort. The left column is why campus screening defaults to pedigree; the right column is what replaces it.

Measure what the file cannot say

If the file cannot sort the pool, the question becomes which constructs can, for candidates who have never held the job or any job. The best current estimates come from Sackett, Zhang, Berry, and Lievens (2022), who re-estimated the meta-analytic validity of selection methods using less aggressive corrections for range restriction (the adjustment for only ever observing the people who were hired) than earlier syntheses. Their value for general mental ability against job performance is .31, and for candidates without work histories a reasoning measure is the nearest available substitute for the track record the file cannot supply. The full ranking of methods, and the logic for reading it, belongs to the series' flagship review of what predicts job performance.

Reasoning is the floor, not the battery. Most campus programs hire into named tracks, software, data, finance operations, sales engineering, and a track implies a subject: a role-ready graduate is one whose subject knowledge has been shown directly, since a course title on a transcript shows only enrollment. Subject batteries do that sorting directly, and for technical tracks the construct extends to working code produced under realistic constraints, the design problem the companion article on coding assessments in hiring takes up. Where the role is client-facing or the working language of the organization is not the campus's teaching language, working English is a construct of its own, and an interview accent is not a measurement of it.

Personality belongs in the battery for a different reason. Conscientiousness carries a validity of .19 in the same 2022 re-analysis, a modest coefficient that earns its place because the measure is inexpensive at volume and uncorrelated enough with reasoning to say something the ability score has not already said. It should inform a shortlist, not gate one. Its larger dividend arrives after the offer, when a facet-level profile becomes the first development conversation a new graduate has, a thread the implications section returns to.

Two design rules hold the battery together. One instrument per construct: a test that tries to read reasoning, knowledge, and temperament through a single score cannot tell you which one moved. And constructs before brands: the program chooses what to measure from the evidence, then finds instruments whose claims survive scrutiny, the measurement-first tradition that runs back through Schmidt and Hunter (1998). What each instrument measures, and the evidence behind it, is the sort of documentation a buyer should expect to find published.

A battery also changes where the funnel can afford to look. A résumé-first program concentrates on the campuses it can physically visit and already trusts, because reading files from everywhere is impossible; a measurement-first program can accept applications from a far longer tail of colleges at nearly zero marginal cost and let scores argue for candidates no recruiter would have met. The widening is not charity. Pools the prestige filter never swept are where an instrument finds the candidates competitors misprice, and a program that measures the whole pool is the only kind that will find them.

Measure early, interview late

Sequencing follows from cost structure and validity together. The battery costs nearly nothing per additional candidate; the interview is the most expensive hour in the process and, run with structure, also the most valid stage in the funnel, at .42 (Sackett et al., 2022). The design that respects both facts puts the cheap, valid instrument first, across everyone, and the expensive, most valid instrument last, aimed only at candidates the measurement has already sorted. Campus hiring goes wrong in exactly one sequencing direction: spending interviewer hours on unsorted volume, which buys noise at the highest price a high-volume hiring process can pay.

Figure 2 traces a season as a computed illustration, not an observed program: 12,000 applications; 1,800 shortlisted by battery, a 15% pass rate; 400 interviewed; 120 offers. Read it for the proportions, not the totals. One file in a hundred ends in an offer, so each stage exists to make the next stage affordable, and the battery does in days what no screening team could do defensibly in a season. That is the arithmetic the thin-file evidence forces: when the file cannot carry the first cut, an instrument must, because something always does, and the unexamined default is pedigree.

The middle step, 1,800 to 400 in the illustration, is governed by one rule: combine what the battery measured by a formula fixed before the season, with weights written into the plan, not by committee impressions of each score report. Decided in advance, the combination is a policy the program can defend to a candidate, a campus, or a court. Decided candidate by candidate, it is the résumé problem again, wearing scores.

The late funnel earns its validity only through structure: questions derived from the job, asked identically of every candidate, scored on anchored scales, the elements catalogued in Campion, Palmer, and Campion's (1997) review of interview structure. At campus volume the discipline matters twice over, because dozens of interviewers work in parallel and every uninstructed panel is a private rubric. The companion playbook on running structured interviews turns those elements into an operating protocol, panel by panel.

For role-ready tracks the finalist round can carry one more measurement: a work sample, at .33 in the 2022 estimates, a half-day exercise that is unaffordable for the illustration's 12,000 candidates and cheap for its 120 finalists. The calendar does the rest. Offers on campus compete with other offers measured in days, so the sequence has to land inside the season's window, which is what the week markers in Figure 2 are for: a funnel that measures early can afford to decide quickly at the end, when speed matters most.

At volume, integrity is infrastructure

A defect that costs one candidate in an experienced-hire process costs a cohort here. A compromised form does not fail the candidate who leaked it; it fails every score on that form, in a compressed window with no time to re-run, in front of an audience that talks. Campus placement tests are a public event in a way corporate assessments never are: the morning sitting is discussed in group chats before the afternoon sitting begins, across colleges and cities. Integrity at this scale is not a reaction to suspicious candidates. It is infrastructure, designed before the season, priced into the funnel, and applied by stage rather than by hunch.

Delivery at scale is professionally charted territory. The International Test Commission's guidelines on computer-based and internet-delivered testing (2006) set out the obligations of online delivery, from technical equivalence to security to the treatment of test takers, and a campus program that follows them starts defensible instead of arguing its way there later. The leak economics are equally well understood: a fixed form shown to thousands of candidates is, in effect, a published form. Adaptive delivery counters this by assembling a different test for each candidate from a calibrated pool, an argument the briefing on adaptive testing in hiring makes in depth, item security included.

The remaining design choice is how much scrutiny each stage carries, and the answer should track what each stage decides. Figure 3 maps the season's three sittings against four integrity measures. The practice window is deliberately open: no identity binding, no monitoring, because its purpose is familiarization and its score decides nothing. The screening battery binds identity to the sitting and runs monitored delivery at a proportionate tier, because it decides the shortlist. The finalist round adds the strongest control available, a verified second sitting: a supervised or highly monitored retest that confirms the remote score before an offer rests on it. That two-stage design, screen cheap and verify scarce, is the standard repair for remote screening, and the companion article on unproctored testing and score verification sets out when it is mandatory rather than optional.

Tiering is also what keeps the candidate experience proportionate. A whole applicant pool should not sit a maximally surveilled screen for a first-pass shortlist, both because trust erodes at scale exactly as fast as leaks spread and because the strictest tier is expensive enough to deserve rationing. Reserve it for the sittings where a wrong score changes an offer, announce every control before the sitting begins, and let the practice window teach candidates the interface with nothing on the line.

Volume forces one further discipline a single-cohort process never faces. A season's battery cannot run in one sitting, so scores must stay comparable across days, campuses, and forms, a psychometric obligation rather than a scheduling detail. Device reality belongs in the same plan: a funnel that assumes broadband and a private room will fail candidates by geography, and those failures arrive in the data looking like low ability unless completion is tracked separately from performance, which is the first entry in the season's telemetry and the subject of the next section.

The season is a measurement project of its own

While the funnel selects candidates, it also produces data about itself, and most programs discard exactly the trail they will need in the spring. Completion rates by campus and by device show where the delivery failed people rather than the other way around. Item performance shows which questions did their job and which leaked, drifted, or confused. Adverse-impact ratios, monitored while the season runs rather than computed after it closes, show whether a cutoff is filtering a protected group at a rate the program cannot defend, the discipline the briefing on the four-fifths rule details. And yield by campus shows where offers convert, which is the number the next season's travel budget should be built on.

Kept together, those numbers make the season self-correcting, and Figure 4 closes the loop. This season's score distributions set next season's cutoffs on evidence instead of instinct. This season's item statistics decide which questions retire and where the pool needs new calibration. This season's completion and yield numbers redraw the campus map. Cutoffs are the clearest case: a first-season cutoff is a guess dressed in a number, while a third-season cutoff is a policy with a track record, set where the score distributions and the downstream interview outcomes say the trade-off actually sits. A graduate recruitment assessment program differs from a season of tests in exactly this respect: the program remembers, and each pass through the loop makes the instrument better than the volume alone ever could.

Campus hiring is won before the season opens

By the time the first campus visit is scheduled, a campus hiring program's ceiling is already set: the instruments are chosen or they are not, the sequence is designed or it will improvise, and the telemetry will be kept or lost. The implications of the evidence are therefore commitments made in the off-season, and four of them carry most of the weight.

• Pick instruments before campuses. Choose constructs from the evidence, then instruments with published validity claims, mapped to the role tracks the program hires into. A campus assessment program assembled mid-season is a screen without a validation argument, and at this volume that is a liability with a cohort's worth of exposure.
• Publish the process to candidates. What is measured, when results arrive, and what each stage decides. Students share everything across colleges within hours; a process that is clear and consistent compounds through the same channels that punish one that is arbitrary. The practice window is part of this: it teaches the interface before anything is at stake.
• Structure the late funnel as tightly as the early one. The battery's discipline is wasted if the interview stage reverts to conversation. Same questions, anchored ratings, and scores recorded before any debrief, for every panel on every campus, because parallel panels are where a season's consistency goes to die.
• Keep the telemetry, and keep the profiles. The season's data calibrates the next season, and the shortlist's profiles keep paying after the hire: the briefing on assessment data after hiring follows a graduate's profile into onboarding, development, and early-attrition watchfulness.

A program built to these commitments is auditable by construction, and when the organization wants that made formal, the capstone guide to building a standards-based assessment program supplies the catalogue. The deeper return is cumulative. The first season is the hardest a campus hiring program will ever run; every season after it inherits calibrated cutoffs, a seasoned item pool, and a map of where offers convert. Volume, which begins as the constraint that breaks the résumé playbook, ends as the asset no smaller process can match.