Week 1. The Firm as a Labor-Allocation System: Frameworks, Facts, and Measurement

Learning Objectives¶

By the end of this week, students should be able to:

1. distinguish workers’ skills from effort, assignment, managerial inputs, and coordination;

2. explain why wages, output, evaluations, and welfare are different objects;

3. translate an organizational decision into a testable labor mechanism;

4. identify the records and variation a credible study would require;

5. state what a study within one firm does and does not establish.

Opening Orientation¶

Central question: What becomes visible when we study workers, jobs, managers, and teams inside firms rather than treating the firm as a single production unit?

Consider two workers with the same occupation, education, and contracted hours. They may face different customers, supervisors, schedules, incentives, access to colleagues, or opportunities for promotion. Even identical worker attributes need not imply identical output or identical welfare. Conversely, similar observed output can mask different effort costs, task difficulty, or support from colleagues. The empirical question is not simply who performs better. It is which part of the organization generates the difference, and what would change under another organizational rule.

This course studies the firm as a labor-allocation system. The firm does more than purchase a quantity of labor. It selects people, assigns work, creates information, evaluates performance, delegates authority, develops skills, and divides surplus. Workers respond through effort, cooperation, applications, learning, bargaining, and exit. The relevant choices therefore occur both between firms and within them.

Three readings orient the course. Lazear and Shaw provide the personnel-economics foundations; Hoffman and Stanton connect recent research on people and practices to productivity; Benson and Shaw organize research around what managers actually do Lazear & Shaw, 2007Hoffman & Stanton, 2025Benson & Shaw, 2025. These are complementary reading maps, not three versions of the same compulsory survey. Week 1 establishes the framework used to read the empirical papers that follow.

Core Points¶

Bridge¶

Labor I and Labor II provide the language of labor supply, human capital, search, wages, and firm behavior. Here we open the firm and study the decisions behind those aggregate objects. The special topics on behavior, gender, technology, and labor design supply relevant mechanisms; this course organizes them around a continuing employment relationship rather than a particular shock or population.

The recurring sequence is organizational decision → worker response → observable outcome → credible comparison → firm and worker incidence. Week 1 introduces that sequence. Later weeks deepen individual mechanisms rather than restarting the course from a new model each time.

Field Core¶

1. Define the organization before asking whether it performs well¶

Start with actors and decisions. A firm owner may value profits; a frontline supervisor may be rewarded for team targets; an employee may value earnings, predictable hours, learning, or promotion. These objectives can coincide, but need not. The first task is to identify who chooses the rule, who implements it, and who responds.

The observational unit matters just as much. A worker-month panel is useful for wages and retention. A worker-task-day panel may reveal output and task assignment. A manager-team panel can reveal span of control and team outcomes. A plant panel may measure production and management practices but cannot automatically explain individual careers. Legal employers, establishments, teams, managers, and occupations are not interchangeable identifiers.

Baker, Gibbs, and Holmström’s study of personnel records shows what becomes visible inside a firm: job levels, transitions, promotion patterns, and pay variation within as well as across levels Baker et al., 1994. Use it as an example of disciplined descriptive work. A hierarchy and transition matrix are informative economic objects before any causal coefficient is estimated; they are not, on their own, evidence that the hierarchy is efficient or discriminatory.

In class, translate a raw title such as “senior analyst” into a data question. Is it a pay grade, a supervisory role, a task family, or a renamed job? A title change can be a promotion, a reclassification, or an administrative artifact. Students should specify what evidence would distinguish these possibilities before creating a promotion indicator.

2. A compact model: people, jobs, effort, and organizational rules¶

The following is a teaching framework, not a claim to reproduce any cited paper’s exact model. Let the organization choose rules

Here $f$ indexes the firm; $\mathcal S_f$ is the workforce selected, $a_f$ the assignment rule, $b_f$ the pay rule, $\mathcal M_f$ management and authority, $G_f$ collaboration, and $\mathcal H_f$ training. For worker $i$ at time $t$, performance in task or job $k(i,t)$ can be represented as

Here $y_{it}$ is output, $f_k(\cdot)$ the production function for job $k$, $\theta_i$ persistent skill, $h_{it}$ accumulated knowledge, and $e_{it}$ effort. The index $m(i,t)$ identifies the current manager; $M_{m(i,t),t}$ measures manager inputs, $G_{it}$ the relevant team environment, and $\varepsilon_{it}$ an output shock. The notation is intentionally permissive: a manager can affect assignment or information, not just add a fixed quantity to output.

The firm chooses rules anticipating workers’ responses. One accounting representation is

The objective is expected profit for the accounting period: $R_f$ is revenue, $w_{it}$ compensation, $C_f$ the cost of organizational practices, and $K_f^{\mathrm{turnover}}$ turnover costs. Choices remain subject to production, staffing, contractual, and legal constraints. This is not an assumption that observed firms implement the globally optimal rule. Limited information, managerial self-interest, coordination costs, and implementation failures can make observed choices depart from that benchmark. Nor is firm profit the course’s welfare objective.

An employee instead evaluates a bundle:

Here $U_{it}$ is worker value, $u$ utility from pay, $c_i$ effort cost, $A_i$ the value of job amenities, and $\delta\in[0,1)$ a discount factor. The continuation value $V_{i,t+1}$ includes learning, promotion prospects, and future outside options. A pay increase, a task transfer, and a better supervisor may therefore improve welfare through different channels. Mas and Pallais provide an empirical example of valuing work arrangements through choices over job offers rather than inferring amenity value from a single wage comparison Mas & Pallais, 2017.

What the framework buys us. A changed rule can affect the workforce, what workers do, how they do it, what they learn, and whether they remain. The model tells us which records are needed to separate these responses. It does not tell us that the responses are identified just because all the variables appear in a regression.

3. Assignment: the best worker is not necessarily the worker for this job¶

Take two workers and two jobs with the following hypothetical outputs. These numbers are a constructed teaching example, not empirical estimates.

Worker A has an absolute advantage in both jobs. Assigning A to Job 1 and B to Job 2 yields 15. Assigning B to Job 1 and A to Job 2 yields 21. A ranking of individual ability is insufficient: assignment depends on comparative advantage and on the opportunity cost of occupying a particular role.

In a larger organization the benchmark is

The binary indicator $a_{ik}$ assigns worker $i$ to job $k$; $v_{ik}$ is that match’s output value and $n_k$ job capacity. The constraints permit at most one job per worker and respect each job’s capacity. The values $v_{ik}$ need not be observed, and current assignments reveal only a subset of potential worker–job outcomes. This is why an assignment question is also a measurement and identification question. Predicting performance in the current job does not reveal performance in an untried role.

Benson, Li, and Shue compare current sales performance, promotion decisions, and subsequent managerial performance Benson et al., 2019. The study is useful here because it separates the employee’s present job from the job into which that person is promoted. Do not interpret the result as proof that any promotion based on performance is irrational: promotion may simultaneously allocate talent and provide incentives. Week 6 will examine that tradeoff in depth.

Haegele’s study provides a newer organizational friction: managers’ incentives to retain strong workers can inhibit internal applications and talent allocation Haegele, 2026. Week 1 uses this to motivate the distinction between team objectives and organization-wide objectives; it does not attempt the full identification argument reserved for the internal-mobility week.

4. Incentives, selection, and measurement are different explanations¶

Suppose average recorded output rises after a change in pay. At least four explanations deserve attention: incumbent workers change effort; the firm attracts or retains different workers; tasks or customers are reassigned; or the recording system changes. Learning and technology can add further channels. A raw before–after mean does not distinguish them.

Lazear’s Safelite study links a move toward piece rates to productivity and workforce selection Lazear, 2000. Shearer’s tree-planting field experiment changes compensation for workers observed under both schemes Shearer, 2004. Read them together as contrasting routes to the effort-versus-selection question. The experiment narrows selection concerns for its experimental comparison; it does not make its environment representative of all jobs. The observational reform captures a broader organizational response but needs additional discipline to separate mechanisms.

A useful accounting exercise decomposes a change in average output. Let $F_d(x)$ be the distribution of observed worker types under regime $d$, and $\mu_d(x)$ their mean outcome. On a common, well-defined support,

Here $d\in\{0,1\}$ denotes the regime, $x$ observed worker type, and $\bar y_d=\int\mu_d(x)\,dF_d(x)$ the regime’s mean output. The first term holds the baseline composition fixed; the second changes composition at regime-1 outcomes. This is an accounting identity, not an identified decomposition of effort and selection. The within-type change can still include changing task difficulty, shocks, and unmeasured selection. Reversing the reference distribution can change the allocation between components.

Practical implication: keep entrant, stayer, mover, and leaver records distinct. Do not interpret a balanced panel of survivors as the full policy effect. Do not code a missing performance review after exit as zero productivity. Also do not omit exit just because output is missing: retention is itself an outcome.

5. The measurement problem: what is the number actually measuring?¶

Output per paid hour, completed tasks, defect rates, supervisor ratings, earnings, and customer satisfaction are not alternative labels for a single latent quantity. Each records a different part of production or reward. A time-use log may measure activity rather than value; a sales total may reflect the territory assigned; a rating may contain information and incentives; a wage may include rents, insurance, compensating differentials, and institutional pay rules.

Use a simple measurement distinction:

The performance measure $R_{it}$ and the pay rule $b_f$ are separate objects. Here $q$ converts output, quality, and difficulty into a recorded assessment; $\nu_{it}$ is measurement error, and $w_{it}$ is compensation. An organizational intervention can change either measurement or pay. The equation does not license treating $\nu_{it}$ as classical noise; evaluator bias, strategic reporting, and selective observation may make it nonclassical.

Gibbons and Holmström–Milgrom motivate why rewarding a measurable task can redirect effort away from valuable but poorly measured work Gibbons, 1998Holmström & Milgrom, 1991. Week 1 draws the measurement implication: collect at least one outcome that is not simply the target the organization rewards. Week 4 will develop the incentive model.

For each paper, students should prepare a measurement card: unit, timing, denominator, producer of the record, incentive to manipulate it, missingness rule, and the claim it can support. For example, performance among retained workers and retention among all initially assigned workers have different denominators. Reporting both does not automatically identify performance for those who exited.

6. Managers, teams, and systems: what is hidden in a worker residual?¶

Lazear, Shaw, and Stanton use worker–supervisor productivity data to study the contribution of bosses Lazear et al., 2015. Hoffman and Tadelis connect manager assessments to employee attrition and manager rewards Hoffman & Tadelis, 2021. These papers justify observing the supervisor and worker relationship rather than assigning all unexplained variation to worker ability. Their findings do not imply that every manager quality score is causal or that a supervisor has the same effect in all production environments.

A descriptive panel specification might be

Here $Y_{it}$ is a measured outcome, $\alpha_i$ a worker effect, $\gamma_{m(i,t)}$ a manager effect, and $\delta_{k(i,t),t}$ a job-by-time effect. The index $u(i,t)$ identifies the unit receiving the rule change, $D_{u(i,t),t}$ its exposure indicator, $\beta$ an exposure coefficient, and $\epsilon_{it}$ a residual. Worker effects account for fixed worker differences; job-by-time effects address some task/time conditions. A causal interpretation still needs an argument about manager assignment and reform timing. Manager changes after declining performance can generate misleading patterns. Sufficient worker–manager mobility and connected comparisons are needed even to separate additive effects; fixed effects cannot be recovered independently when identifiers are perfectly nested. A one-firm sample also cannot identify between-firm effects.

Mas and Moretti use detailed workplace output and coworker exposure to study peer effects Mas & Moretti, 2009. Use the paper to ask what coworkers can observe and how interactions occur. A team average is not a sufficient description of peer exposure when visibility, task dependence, and interaction timing matter.

Some organizational practices may be complements. The steel-line evidence in Ichniowski, Shaw, and Prennushi studies systems of human-resource practices Ichniowski et al., 1997. The management field experiment in Bloom and coauthors studies a package of practices Bloom et al., 2013. Neither setting licenses attributing a bundled effect to every component independently. This is the empirical counterpart to asking whether a training program needs supervisor support or an incentive scheme needs reliable measurement.

7. What detailed organizational data add to inequality research¶

For log wages $\ell_i$, the law of total variance gives

Here $f(i)$ is worker $i$’s employer and $\ell_i$ the log wage. The identity separates between-employer mean differences from within-employer dispersion, using a clearly stated worker-weighted population. It does not separate productivity, rents, sorting, discrimination, or compensation for amenities. Changing the weighting to equal firm weights changes the statistic.

Song and coauthors’ published study of earnings inequality distinguishes within-firm and between-firm changes and emphasizes worker composition Song et al., 2019. The teaching lesson is not that internal labor markets are irrelevant whenever between-firm differences are important. Internal recruitment, training, promotion, and retention can shape who ends up at which firm. Conversely, within-firm dispersion need not come entirely from internal policy.

The new course complements, rather than replaces, matched employer–employee research: broader administrative data describe population patterns; rich personnel data can reveal a decision that generates some of those patterns. Neither dataset automatically supplies the counterfactual missing from the other.

8. Five research designs students should recognize¶

Organize empirical work by the variation, not by the prestige of the dataset.

A descriptive personnel-panel design can establish career patterns and reject simplistic descriptions of how a firm works. Baker–Gibbs–Holmström is the reference point. The limitation is causal attribution: promotion and pay decisions are chosen.

A within-firm policy change can connect a rule to outcomes, as in the performance-pay literature. Students must ask why the timing changed, what else changed, and whether the workforce or task mix changed. A regression with many controls does not answer these questions automatically.

A worker–manager or worker–team transition design exploits changes in relationships. It needs a story for why reassignment is informative rather than a response to unobserved performance shocks. Plot transitions, preserve event timing, and inspect whether both old and new managers have comparable workers.

A randomized organizational intervention can identify the effect of an offer or rule in its implemented environment. Bloom and coauthors’ management experiment is an example; the hybrid-work study supplies the Week 1 lab Bloom et al., 2013Bloom et al., 2024. The assignment unit, spillovers, take-up, and follow-up process remain part of the design.

A choice or survey-linked design can measure preferences or beliefs that administrative records do not contain. Mas–Pallais and Hoffman–Tadelis illustrate different uses of such information Mas & Pallais, 2017Hoffman & Tadelis, 2021. Survey answers are measurements, not automatically revealed preferences; choices with stakes still depend on the available menu and the participating population.

9. A personnel-data workflow that students can actually start¶

Begin with an organizational decision and request the smallest data structure that could distinguish its mechanisms. An initial specification should include a stable worker ID; employment spells; the timing of job, team, and manager assignments; pay components; outcome timestamps and definitions; and records of policy eligibility and actual exposure. Add rejected applications, training opportunities, or outside offers only when the question needs them. “All HR data” is not a design.

Preserve assignment histories rather than joining outcomes to the employee’s latest manager. Treat firms, establishments, and teams separately. Reconcile calendar time, payroll periods, and event time. Document the first date each field became reliable. A new dashboard can create apparent performance growth simply by recording more events.

Public and restricted resources offer different starting points. The WMS provides questionnaires and public data pathways, but identified or panel releases have additional conditions and a download-maintenance notice was visible when this pack was prepared. Census MOPS provides public documentation and aggregate tables; these are not an unrestricted worker-level panel. IAB’s LPP links employer and employee perspectives through controlled-access products. O*NET provides occupation-level descriptions, not employee performance records World Management Survey, n.d.U.S. Census Bureau, n.d.Institute for Employment Research, Research Data Centre, n.d.National Center for O*NET Development, n.d.. The resource map records these distinctions so students do not design a project around nonexistent access.

For research partnerships, agree on the research question, data custody, confidentiality, publication review limited to disclosure issues, and the treatment of null findings before analysis. Do not release employee identities, raw communications, proprietary records, or inferred sensitive attributes in a teaching repository. This is a research-workflow recommendation, not a substitute for institutional review or a data agreement.

10. Why a one-firm study can be broadly informative—and where it stops¶

The aim is not to prove that a particular firm represents the economy. It is to explain why the setting reveals a mechanism and which features govern whether that mechanism travels. A workplace with individually recorded output, stable teams, and standardized tasks may be informative about monitoring or peer pressure; its magnitudes need not transfer to creative project work with diffuse output.

Require a short portability statement for every empirical paper: production technology; workers and jobs observed; outside options; organizational rule; compliance and implementation; and outcomes missing from the study. A multi-firm sample widens coverage but does not erase selection into participation or observation. A single firm with many workers can provide rich variation without supplying many independent organizations.

The same distinction matters for frontier work. Generative AI at Work studies a particular technology introduction and its workplace outcomes; it does not estimate the economy-wide employment effect of AI Brynjolfsson et al., 2025. Talent hoarding is evidence about a specific internal allocation friction, not proof that all internal labor markets are blocked Haegele, 2026. Week 1 should teach students to formulate these limits as part of the scientific contribution rather than as an apologetic closing paragraph.

11. Research opportunities: ask which organizational margin is missing¶

The following are proposed research directions, not claims that the literature is empty.

From screening to careers. Hiring data often stop at entry, while personnel data begin with hires. Linking the funnel to later assignments, training, and exits could show when screening selects productive workers and when organizational support changes the meaning of early signals. The starting reference is Discretion in Hiring; the new question must specify the outcome missing from that design Hoffman et al., 2018.

From manager ratings to managerial actions. A manager effect can summarize teaching, task allocation, or retention. Combining outcomes with time-stamped decisions can separate those mechanisms. The design must address why managers undertake the measured action, not only whether the action predicts performance Lazear et al., 2015Hoffman & Tadelis, 2021.

From one employee’s career to colleagues’ opportunities. A promotion or delayed exit can change the positions available to others. Bianchi and coauthors provide an application of career spillovers Bianchi et al., 2023. A new study could measure the vacancy, internal application, and training margins behind such adjustment rather than assume they are all displacement.

From tool adoption to organizational adaptation. Ask whether the intervention changes delegation, training, communication, or task division, and whether those changes persist. The unit of inquiry is the organizational response, not a generic claim that AI raises productivity Brynjolfsson et al., 2025.

From firm gains to joint incidence. A practice can change turnover costs, employees’ schedules, workloads, and future opportunities. Collect outcomes that make conflicting effects visible. Neither a profit increase nor a quit reduction alone is a welfare calculation.

Each proposed project should finish this sentence: “The existing design observes ___; my design adds ___, allowing us to distinguish ___ from ___.”

Research Lab¶

Primary paper and purpose¶

The primary lab anchor is Bloom, Han, and Liang (2024), Hybrid Working from Home Improves Retention without Damaging Performance. This week uses the experiment to audit organizational units, outcomes, and measurement; Week 9 develops the full work-design discussion. The article identifies a public data-and-code repository at DOI 10.7910/DVN/6X4ZZL Bloom et al., 2024Bloom et al., n.d.. The lab records the inspected release, terms, file hashes, and definitions so that each reported result can be traced to its inputs.

Reproduce¶

The lab offers two explicitly separate paths. Mode A re-estimates one retention contrast and one performance-review contrast from inspected official-release files. Its Python adapter follows the documented comparisons; it does not execute the full original Stata/R analysis. Document the experimental sample, assignment, treatment offer, and outcome window before interpreting either result. Preserve the official anonymization and file splitting rather than attempting to reconstruct employee identities or join keys.

Mode B, the guaranteed offline path, reconstructs the published rounded retention contrast: control quits of 7.2% and treatment quits of 4.8% imply a −2.4 percentage-point difference and a one-third relative reduction. Label the result “published-statistic reconstruction; not microdata replication.” These rates are published inputs, not newly estimated quantities. This path produces no estimated standard error, p-value, confidence interval, or group counts.

Diagnose¶

Prepare an outcome ledger distinguishing quit rates, satisfaction, performance reviews, promotions, and task output. Identify the relevant sample and missingness for each. Explain why an intention-to-treat contrast differs from the effect of actual home-working days. Explain why no detected difference is not a proof of exact equality. The paper’s review-score equivalence tests use bounds of ±0.5 grade points on a five-point scale. Evidence that a difference lies within those bounds is a claim about that margin, not an exactly zero effect. It also does not recover unobserved reviews for workers who left Bloom et al., 2024.

Use the research question to interrogate portability: this treatment is a particular hybrid schedule in a particular organization. It is not a generic treatment of “remote work.” Do not translate a treatment subgroup’s significance and another subgroup’s nonsignificance into a significant treatment-effect difference without a direct test.

Transfer¶

Use a small, explicitly synthetic personnel panel to study a hypothetical organizational policy. Hold average output changes fixed while varying worker selection and incumbent productivity. Compare repeated-cross-section output, stayer output, entry, exit, and a task-quality measure. Students should see how identical headline productivity gains can imply different mechanisms. The synthetic data are not calibrated estimates of the primary paper.

The challenge reading is The Value of Bosses: redesign the data ledger for a manager-transition study and explain why the same outcome table does not supply the same identification argument Lazear et al., 2015.

Submit the outcome ledger, a result with its reproduction mode stated, the synthetic scenario comparison, and a transfer memo. The runnable handout below specifies the files and checks.

Methods Box¶

What belongs here, rather than in a methods course? The practical connection between organizational records and the comparison. Students should know the assignment unit, outcome denominator, measurement process, transition support, and dependent observations. They should not be told that a method name makes a design credible.

If a policy is randomized at the firm or team level, individual records do not become independent treatment assignments. If workers move between managers, dependence and identifying comparisons must reflect that structure. If the measured practice and the outcome are both chosen, controls and clustered standard errors do not repair the missing causal argument. Detailed inference derivations can be assigned from the Empirical Methods course; this box makes their organizational relevance explicit.

Reading Ladder And References¶

Use this chapter for one extended lecture or two shorter meetings. The required framing and one rotating discussion paper form the common workload; the remaining references support deeper inquiry and later weeks.

Required framing: Lazear–Shaw; the introduction, productivity, and external-validity portions of Hoffman–Stanton; Baker–Gibbs–Holmström as the original personnel-data example. Read the primary lab paper’s design, outcome definitions, and data-availability sections Lazear & Shaw, 2007Hoffman & Stanton, 2025Baker et al., 1994Bloom et al., 2024.

Required rotating discussion: assign one of The Value of Bosses or Talent Hoarding in Organizations to half the class each; compare their decision-makers, data, and counterfactuals Lazear et al., 2015Haegele, 2026. This preserves a manageable required load while making current research visible.

Model and measurement pathway: Gibbons, 1998Holmström & Milgrom, 1991Lazear, 2000Shearer, 2004Bloom & Van Reenen, 2007Ichniowski et al., 1997.

Managers, careers, and interaction pathway: Benson & Shaw, 2025Hoffman & Tadelis, 2021Benson et al., 2019Mas & Moretti, 2009Bianchi et al., 2023.

Scale, entry, work design, and frontier pathway: Song et al., 2019Hoffman et al., 2018Bloom et al., 2013Mas & Pallais, 2017Brynjolfsson et al., 2025.

Exercises And Discussion Prompts¶

1. Solve the two-worker assignment example and check Figure Figure 1. What extra observations would be needed to learn comparative advantage from observed assignments rather than from the supplied hypothetical matrix?

2. A firm’s output per worker rises after a bonus reform. Write three observationally compatible mechanisms and one additional outcome or comparison that would help distinguish each.

3. Construct a measurement card for sales, a supervisor rating, and a quit indicator. For each, state the denominator, the observation process, and one claim it cannot support.

4. A manager changes just after a team misses its target. Explain why an event-study recovery in performance may not identify managerial quality. Propose a better source of variation and the records it needs.

5. Write a portability paragraph for one required empirical paper without claiming that its firm is representative.

6. Draft a one-page project: decision-maker; rule; worker response; firm and worker outcomes; data; comparison; principal threat; and one result that would falsify the proposed mechanism.

Reproducibility And Code Lab Note¶

Download the teaching repository (ZIP), or clone the repository. Keep its folder structure and open books/special-topic8-organizational/labs/01-the-firm-as-a-labor-allocation-system/; the lab README supplies the run instructions. Weeks 1–3 use the shared plotting module supplied with the repository.

The Week 1 lab handout guides Reproduce, Diagnose, and Transfer. Its folder is labs/01-the-firm-as-a-labor-allocation-system/. Run the guaranteed offline path from that folder:

ENV_NAME=research bash smoke.sh

It reconstructs published rounded statistics, builds a separate synthetic personnel panel, and checks the arithmetic, observation states, comparison support, and reproducibility of the generated outputs. It requires no live download or credentials. Read each output’s provenance label: published statistics, official-release estimates, and constructed teaching data support different claims. Keep raw employee records and restricted data outside the public repository.

The assignment figure is generated independently with:

conda run -n research python assets/scripts/build-week1-assignment.py

Slide Companion Note¶

The Week 1 slide deck (PDF) follows the assignment example, measurement distinctions, paper comparisons, and research lab. It highlights the economic logic and essential equations; the chapter supplies the fuller reading pathways. The Beamer source lives alongside the PDF at slides/week1/01-the-firm-as-a-labor-allocation-system.tex.

Bridge Forward¶

Week 2 opens the recruiting process: whom the firm sees, how it screens, and which candidates enter. The framework established here will keep those questions connected to post-hire assignment, learning, performance, and retention rather than treating recruitment as an isolated prediction exercise.