The Silence and the Secrecy: S&P 500 Layoffs and the H-1B Spike

Amazon filed 268 H-1B visa petitions the quarter it announced layoffs. The next quarter, it filed 4,250.

That difference, roughly a 1,400% increase, is unusual but not unique. Meta went from 40 petitions to 1,034 in a single quarter. Tesla, 30 to 478. IBM, 82 to 571. Across 308 layoff events at 110 S&P 500 companies, over six years of quarterly data,¹ a pattern repeats with a consistency that stops looking like coincidence somewhere around the thirtieth company: H-1B visa petitions drop to near zero the quarter of a layoff announcement, then surge dramatically the quarter after, often in the same job categories, at companies whose quarterly financials show nothing resembling distress.

I want to show you that data in detail. But I think the numbers make more sense, and are harder to dismiss, if you understand the pattern they're measuring. The pattern is old. It is roughly 400 years old, and it starts with a congressman from Ohio who saw it clearly enough to describe it in language that still works.

The Pattern Is Old

In 1885, a former labor organizer named Martin Foran stood on the floor of the House of Representatives and drew a distinction that would, 140 years later, remain the central unresolved question of American immigration policy. He was introducing the Alien Contract Labor Law, prompted by the use of imported Belgian glassworkers as strikebreakers, and the distinction he drew was between two kinds of foreign workers: those who come "of his own volition" and those brought by capitalists "whose only object is to obtain labor at the lowest possible rate." He called the latter group "virtually the slaves of those greedy corporations" and framed his bill as an antislavery measure.²

The Foran Act passed. It prohibited companies from importing workers under pre-arranged contracts. It contained an exemption: skilled workers and professionals were excluded. That exemption is the crack through which the next century and a half of contract labor would flow, because the crack was not a flaw in the architecture; it was the architecture. Every subsequent framework for importing labor would reproduce it: a stated purpose of supplementation, a prevailing wage requirement, protections against displacement, and an exemption wide enough to drive an entire industry through.

The H-1 visa (1952). The H-1B visa (1990). The outsourcing boom (2000s). The exemption changes names. The function persists.

The first labor importation system in British North America was indentured servitude. Roughly half to two-thirds of all white immigrants to the colonies before the Revolution arrived under contracts pledging four to seven years of labor in exchange for transatlantic passage. The structural parallels to modern visa-dependent work are not subtle: the worker was bound to a specific employer, the employer controlled their legal status, and the system existed because colonial employers argued they could not find sufficient local labor at the wages they were willing to pay. The key difference was that indentured servants eventually earned their freedom. By the 1820s the system had largely died out, not because of moral progress but because improving conditions in Europe reduced the supply of people desperate enough to sign away years of their lives.³

After indentured servitude declined, employers turned to Chinese laborers, recruited through a system critics called the "coolie trade." The Central Pacific Railroad employed thousands of Chinese workers in the 1860s to build the transcontinental railroad. They were paid 30 to 50 percent less than white workers doing identical work, housed in separate camps, and assigned the most dangerous tasks, particularly handling explosives for tunnel blasting. The justification was identical to what you would hear today, almost word for word, "there simply were not enough willing domestic workers for the job." When the railroad was completed and a recession hit in the 1870s, the same workers who had been deemed essential were suddenly cast as a threat. The Chinese Exclusion Act of 1882 followed. The rhetorical arc from "we need them" to "they're taking our jobs" took less than fifteen years.

I realize that calling something a 'pattern' across 400 years of wildly different economic contexts is the historiographical equivalent of noting that every U.S. president has had a head, and concluding that heads are presidential. The counterargument (each era is different, the legal frameworks are different, the economies are different) has force. But the specific structural features keep recurring with a regularity that is either coincidental or diagnostic, and I don't think it is coincidental.

Lincoln's 1864 Act to Encourage Immigration. The Foran Act of 1885. The padrone system (1860s through 1920s), in which Italian labor brokers recruited workers abroad, advanced their passage, controlled their wages and movement, and placed them with American employers, sometimes behind barbed wire fences patrolled by armed guards.⁴ The Dillingham Commission (1907 to 1911), which produced 41 volumes of data attempting to measure immigration's effects on American labor and found that employers were using "new immigrants" to undercut "old immigrants" and native workers by accepting lower wages. The National Origins Act of 1924, which slashed immigration from 350,000 to 150,000 per year and, contrary to industry predictions of economic catastrophe, led to mechanization and wage increases for domestic workers rather than collapse.

Then the Bracero Program. This is the one that matters most for understanding what the data shows now.

The Bracero Parallel

From 1942 to 1964, the United States and Mexico operated a bilateral guest worker program that brought 4.6 million Mexican men to work on American farms under temporary contracts. The program promised prevailing wages, decent housing, no discrimination, and protections against displacement of domestic workers. Over 22 years, every single one of those protections was systematically violated.

Farm wages dropped sharply as a share of manufacturing wages during the Bracero era, a direct and measurable consequence of the labor surplus the program created. Employers ignored wage guarantees, knowing workers could not complain without risking deportation. The government withheld 10 percent of bracero wages in mandatory "savings accounts" that were supposed to be returned when workers went home; most of the money was never returned. When the program ended in 1964, growers did not raise wages to attract domestic workers. They mechanized.

The parallels to the modern H-1B need almost no narration; they narrate themselves.⁵

The vocabulary updates. The org chart stays the same. The collar changed from blue to white. The ~16 wage discount stays the same.⁶

The Methodology

What follows is the methodology, which I'm laying out in full because if the joins are wrong the conclusions are worthless, and you deserve to see the seams.⁷

The data pipeline cross-references five public federal and state datasets: quarterly financial filings from the SEC (10-Q and 10-K reports, pulled via EDGAR), H-1B Labor Condition Application filings from the Department of Labor, H-1B employer approval data from USCIS, WARN Act layoff notices scraped from 35 state databases, and monthly macro layoff figures from FRED. All of this information is public. None of it is leaked, stolen, or obtained through any process more dramatic than downloading Excel files from government websites and writing a lot of entity resolution code. The entire pipeline, from raw data pulls through analysis, is open source on my repo.

The core analytical unit is the five-quarter window. For each layoff event at an S&P 500 company, we pull the two quarters before the layoff (Q-2, Q-1), the layoff quarter itself (Q0), and the two quarters after (Q+1, Q+2). Into each window we load: the company's revenue, operating margins, R&D spending, CapEx, free cash flow, debt-to-equity ratio, and interest coverage; the number of H-1B petitions filed; the number flagged as AI/ML roles; the average wage on those petitions; the SOC codes being hired for; and the wage distribution across salary brackets.

The dataset covers 308 layoff-quarter events across 110 S&P 500 companies, with 607,288 H-1B LCA records and 29,600 WARN Act notices serving as the underlying granular data. These are large enough numbers to feel authoritative and small enough, if the entity resolution is wrong in systematic ways I have not caught, to be completely misleading. I believe the joins are correct. I also believed my first three versions were correct, and they were not, which is why I am showing you the methodology rather than asking you to trust me.

Now. What does it show.

The Spike

The single most consistent pattern in the data is the one I opened with: H-1B petitions drop to their lowest point during the quarter a company announces layoffs, then surge dramatically in Q+1 and Q+2.

The numbers you saw at the top are typical. They are the central tendency. Meta's 40-to-1,034 spike came immediately after layoffs that Meta publicly framed around "efficiency" and "flattening the organization." Tesla's 30-to-478 spike came one quarter after Elon Musk announced he was cutting 10% of the global workforce. IDEX Corporation, 41 to 463. KLA Corporation, 1 to 80. The effect is visible from orbit. You do not need to squint at a regression coefficient or argue about statistical significance over beers at a conference. These are not rounding errors.⁸

The pattern also holds for companies with repeat layoffs. Intel had layoffs in 15 of the 24 quarters in our dataset, with H-1B filings trending upward across the entire period. The word "restructuring" does not survive contact with 15 consecutive quarters of the same pattern. A permanent labor substitution program that announces itself in quarterly installments looks exactly like this. Amazon had layoffs in 11 quarters with the same pattern: increasing H-1B filings. Wells Fargo hit 13 layoff quarters though without the H-1B ramp: pure cost-cutting with no transformation narrative and no visa-sponsored replacement.⁹

Two claims live inside this pattern, and they need separating.

The narrow claim: at the company level, across a large sample of S&P 500 firms, layoff events and subsequent H-1B filing surges are temporally correlated. This is what the numbers say, and the numbers are clear about it.

The wider claim, the one a reader might reach for, is that the visa hires replaced the laid-off workers. Same roles, same desks, lower wages. The data cannot confirm this through individual worker matching. The WARN Act filings do not include job titles. The LCA data does include job titles and SOC codes, but without both sides of the ledger in the same row, you cannot prove one-to-one substitution. The data draws the outline. It does not fill it in.

But occasionally someone fills it in for you.

In 2014, Disney laid off approximately 250 IT workers at its Orlando parks and replaced them with H-1B contractors supplied by Cognizant and HCL. Leo Perrero, one of the displaced workers, was required to train his replacement as a condition of receiving his severance. So was Dena Moore, a database administrator who had worked at Disney for over a decade. Both later testified before Congress. Their case is the clearest public instance of the pattern the data describes, and the reason it became public is that they refused to stay quiet about it, which is not something a system built on silence and severance agreements is designed to accommodate.¹⁰

Cognizant, incidentally, is one of the two firms that supplied Disney's replacements, which makes what happened next feel less like a sequel and more like a second season of the same show. In October 2024, a federal jury in the Central District of California found that Cognizant had engaged in a "pattern or practice" of intentional discrimination against non-Indian and non-South Asian employees across a class of roughly 2,300 former workers. The mechanism was elegant in its bureaucratic simplicity: Cognizant maintained internal "Visa Readiness" and "Visa Utilization" policies that built a pipeline of H-1B holders before actual positions existed, then staffed those workers on projects while leaving non-Indian employees "benched" (given no work) until they could be terminated for having no work. The jury authorized punitive damages; amounts are pending.¹¹ The verdict is technically broader than H-1B replacement (it covers national origin and race discrimination under Title VII), but the visa pipeline was the delivery system, and the delivery system is what the jury examined for three weeks before concluding that what it delivered was intentional.

The data so far shows timing. The next three sections ask whether it also shows the same jobs, at the same desks, for less money.

The Seasonality Objection

The strongest counterargument has to do with the calendar.

The H-1B program runs on an annual cycle that creates a natural seasonal spike in filings. Employers register in March, the lottery (for cap-subject new hires) runs in late March, selected petitions are filed April through June, and employment begins October 1. If you are looking at quarterly LCA filings, you will see more applications in calendar Q1 and Q2 (January through June, when the lottery runs) of any given year regardless of whether the company recently laid anyone off. A skeptic could argue that the post-layoff "spikes" are just the seasonal pattern asserting itself, and the temporal proximity to layoffs is coincidence.

Fair objection. We built company-specific baselines to test it. For each company with a post-layoff H-1B spike, we computed their average quarterly petition count across all non-layoff quarters, then measured the Q+1 and Q+2 counts against that baseline rather than against the depressed Q0 trough. We also flagged which calendar quarter Q+1 and Q+2 fell in, to separate lottery-season effects from genuine H-1B application spikes.

The result: the strongest post-layoff surges survive seasonal adjustment decisively, and they cluster outside lottery season. Amazon's Q+1 after its 2024-Q2 layoffs came in at 3.73x its own baseline, landing in Q3 (July through September, not lottery season). Intel hit 3.0x baseline, also in Q3. IBM hit 3.0x. APA Corporation, 2.8x. Microsoft, 2.4x. Meta, 2.3x. Tesla, 2.2x. Every one of these spikes lands in Q3, which is transfers and extensions filed year-round, not cap-subject lottery petitions. The seasonal calendar does not explain them.

The Q1-layoff cohort tells a different story. When a company lays off in calendar Q1 (Jan - Mar), Q+1 falls in calendar Q2 (Apr - June, lottery season), and many of those spikes partially deflate when compared to baseline. Apple's 29,533% increase from Q0 drops to 2.85x baseline at Q+2, still elevated but within range of normal seasonal variation. Salesforce's 41,300% spike from Q0 becomes 1.94x baseline at Q+2. These are still signals, but weaker. The seasonal adjustment doesn't eliminate the pattern. It clarifies which instances are strongest, and the strongest instances are the ones that can't be explained by the calendar.

The Same Roles

This is the finding that moved the needle most for me, because it addresses the question the data was weakest on: are companies hiring for different roles after layoffs, or the same ones?

Across all companies in the dataset, we tracked 37,621 pre-layoff H-1B petitions spanning 203 SOC codes, and 26,338 post-layoff petitions spanning 198. The role diversity barely changed. What changed was the volume.

The single most common occupation code on both sides of the layoff divide is 15-1252, also known as Software Developers. Before layoffs, 5,067 petitions. After layoffs, 4,104. Companies are laying off workers classified under Software Developer and filing H-1B petitions classified under Software Developer. The same SOC code, appearing on both sides of the same company's layoff quarter, is the closest the data can come to showing same-role replacement without individual worker matching.

They are cycling the same roles through the H-1B system.¹²

The exceptions are genuinely interesting. PayPal's post-layoff AI role share jumped from 16.7% to 66.7% after its 2023-Q3 layoffs, and from 6.1% to 14.9% after its 2023-Q1 layoffs, with wages shifting toward higher brackets. PayPal scored "Genuine AI Buildout" on all seven of its layoff quarters. It appears to be the one company in the dataset that is actually doing what the others only claim to be doing. Meta showed a meaningful AI share increase to 15.3% post-layoff in its 2023-Q2 window, but simultaneously narrowed from 63 SOC codes to 16, suggesting consolidation rather than transformation. RTX shifted from 4% to 14.3% AI roles with five new SOC codes appearing, alongside an average wage jump from $150,000 to $262,000. These cases exist. Though they are outnumbered.

The Wages

The H-1B program requires employers to file a Labor Condition Application (LCA) specifying, among other things, the wage they intend to pay. This wage is supposed to be at or above the "prevailing wage" for the occupation and geographic area. The prevailing wage is set by the Department of Labor at four levels,¹³ and Level 1 (the entry level) sits at roughly the 17th percentile of local occupational wages.

This means a company can legally hire an H-1B software engineer at a salary below what 83% of domestic software engineers in the same area earn, and the Department of Labor will certify the application within seven business days, because the DOL is required by statute to approve LCAs unless there are "obvious errors or inaccuracies," and paying someone at the 17th percentile of their occupation is not, legally, an error.¹⁴

The wage story turned out to be more complicated than the rest of the data, and the complication matters.

A caveat before the numbers: the LCA data tells you what companies offered to pay visa workers before layoffs and what they offered to pay visa workers after layoffs. It does not tell you what the laid-off domestic workers were earning, because those salaries live in payroll systems that are not public. So the comparison that follows is H-1B to H-1B across the layoff divide, not H-1B to displaced domestic worker. The second comparison is the one that would settle the question, and the data cannot make it.

Other people's data gets closer. George Borjas, matching H-1B petition records to American Community Survey data, found in 2026 that visa workers earn roughly 15 to 16 percent less than comparable domestic workers after controlling for education, occupation, age, and location.¹⁵ The DOL's own March 2026 proposed rulemaking, analyzing over 3 million LCAs, put the gap at approximately $19,000 per worker per year. The EPI documented internal records from HCL Technologies showing H-1B workers paid 13 to 64 percent less than U.S. citizens in the same IT roles. None of these studies can match individual laid-off workers to their specific visa-sponsored replacements. All of them point in the same direction: the H-1B wage floor sits well below what domestic workers in the same occupations earn, and the gap is structural, not incidental.

With that context: at most large tech companies, the post-layoff H-1B wage distribution held roughly steady. Apple's petitions clustered in the $120K to $200K range before and after its 2024-Q1 layoffs, though the $200K-plus tier nearly tripled from 88 to 216 petitions. Amazon's wage distribution was nearly proportionally identical before and after. Microsoft, similar. Whatever cost advantage these companies gain from the visa system, it does not show up as lower posted salaries on the LCA filings.

It shows up somewhere the LCA data cannot reach: reduced bargaining power. An H-1B worker in the green card queue cannot negotiate a competing offer without risking their immigration status. They cannot threaten to leave. They cannot organize. They cannot wait for a better deal. Immigration attorneys market this to employers as a "retention benefit."¹⁶ Even at identical posted wages, a worker who cannot walk away is structurally cheaper than one who can, because the option to walk away is itself a form of compensation that never appears on a pay stub.

Where the wage data does tell a sharper story is at specific companies. IBM's 2020-Q2 window is the clearest case in the dataset: 88 pre-layoff petitions became 493 post-layoff, with the sub-$80K bracket doubling from 48 to 100 petitions and the $80K to $120K bracket exploding from 36 to 265. IBM added 22 new SOC codes while losing zero of its pre-layoff codes. AI role share actually dropped from 10.2% to 4.3%. The word "transformation" requires, at minimum, that something transform. More workers, lower wages, same or broader role mix, fewer AI-specific hires: that is a downgrade. IBM repeated the pattern in 2023-Q2, going from 37 petitions to 580, adding 17 new SOC codes.¹⁷

The Financial Vitals

If a company is laying off workers because of genuine financial distress, you would expect it to show up in the quarterly numbers. Declining revenue, maybe. Compressing margins. Rising debt. Some financial signature that explains why the headcount had to change.

What the data shows instead is that most companies conducting layoffs had healthy and stable financials before, during, and after the layoffs. Median operating margins and FCF margins dip at Q0 (the layoff quarter, when restructuring charges hit) and recover by Q+1 and Q+2. Revenue was trending up in the vast majority of cases. The pre-layoff financials at Q-2 and Q-1 were already healthy.

The sector breakdown makes this concrete. Information Technology leads with 129 layoff events with a median H-1B change of +225% from Q0 to Q+1. Financials had 52 events, with a +158% median H-1B change. Consumer Staples is the control group: 28 events, zero H-1B replacement signals, zero wage replacement signals. These are companies doing straightforward cost-cutting with no pretense and no visa-sponsored follow-up.¹⁸

A caveat, because the inference deserves one: "the company was not really in trouble" is my reading of the quarterly numbers, not testimony from anyone inside the building. Financially healthy companies restructure all the time. Priorities shift. Roles become redundant. Products get sunset. All of this is true, and none of it requires distress.

What makes the pattern harder to explain as routine business adjustment is the combination: healthy financials, followed by layoffs, followed by H-1B spikes in the same SOC codes, with AI role share that is flat or declining, at companies that never cited AI as a reason for the cuts. Each piece is individually explainable. Together, they describe something that the word "restructuring" does not quite cover.

The Silence and the Secrecy

Which brings us back to Adam Smith. Not for decoration; because he described this exact mechanism in 1776, in language so precise it reads like he had access to the LCA database.

Masters too sometimes enter into particular combinations to sink the wages of labour even below this rate. These are always conducted with the utmost silence and secrecy, till the moment of execution, and when the workmen yield, as they sometimes do, without resistance, though severely felt by them, they are never heard of by other people."¹⁹

The silence and the secrecy. Smith needed one phrase to describe what the modern H-1B system needs five federal agencies, four wage tiers, a 60-day deportation clock, and a 50-year green card backlog to accomplish. The masters combine. The combination requires no conspiracy, no coordination, no smoke-filled room. It requires only shared incentives and a structure that makes resistance expensive.

The "silence" is the Labor Condition Application, which the Department of Labor is required by statute to approve within seven business days unless it contains "obvious errors or inaccuracies." The DOL has no authority to investigate whether the employer actually needs the worker, whether domestic workers are available for the role, or whether the offered wage is genuinely competitive. The DOL Inspector General called this a "rubber stamping" process in 1996 and noted that it "does not protect U.S. workers' jobs or wages." That was thirty years ago. The process has not meaningfully changed.²⁰ The wage gap documented in the previous section (15 to 16 percent by Borjas's measure, $19,000 per worker by the DOL's own estimate) persists because every institutional actor benefits from maintaining the fiction that the prevailing wage is prevailing.

The "secrecy" is the 90-day gap between a layoff announcement and an LCA filing, which is long enough to break the narrative connection and short enough to serve the same headcount plan. No single agency sees both sides of the transaction. The DOL certifies wages but cannot deny petitions. USCIS adjudicates petitions but does not systematically receive WARN Act notices or financial filings. The SEC has the quarterly numbers but no immigration mandate. Congress has the authority to change the rules but remains paralyzed by a debate that, unusually, does not split along party lines: both parties have constituencies on both sides, which is why the cap has been stuck at 85,000 since 2004 despite repeated attempts to move it from either direction.²¹

The Ecosystem

And behind each agency, an ecosystem that depends on the current arrangement continuing, and that has arranged itself, with the quiet efficiency of water finding its level, so that every participant benefits from the silence.

NASSCOM, India's IT industry body, lobbies Washington as aggressively as any domestic trade group, framing H-1B restrictions as trade barriers equivalent to tariffs; the Indian government raises visa access in bilateral negotiations the way other countries raise market access for goods.²² The Information Technology Industry Council and the Partnership for a New American Economy (whose co-chair, during the Disney layoffs, was Disney CEO Bob Iger) fund research and political campaigns on both sides of the aisle. American universities operate as the credentialing pipeline: international students pay full tuition (often two to three times the domestic rate after financial aid), satisfy the "specialty occupation" bachelor's degree requirement upon graduation, and enter the workforce through OPT, a program that lets employers avoid FICA taxes on their labor for up to three years, before transitioning to H-1B sponsorship. The universities profit at every stage and lobby accordingly, framing their advocacy in terms of global competitiveness rather than the revenue model that underwrites it.²³

Smith, who also observed that "people of the same trade seldom meet together, even for merriment and diversion, but the conversation ends in a conspiracy against the public," would recognize the structure instantly. The masters combine. The combination is tacit. And the workmen yield, "without resistance, though severely felt by them," because the domestic workers are silenced by severance agreements and the visa workers are silenced by deportation risk, and by the time the Q+1 filings arrive the Q0 layoffs have already left the news cycle.

The Compassion Problem

I want to end with something I have not figured out, which is the compassion problem at the center of all this.

The layer the data makes visible is the simplest one. The H-1B worker from Hyderabad who accepts a position at $90,000 for work that the domestic market would price at $130,000 is being underpaid relative to their peers. They are also experiencing a roughly 500% increase over what they could earn in their home country. Both of these things are true at the same time. The worker is being exploited by the program's wage structure and experiencing a life-changing economic opportunity through the program's existence. The company is taking advantage of an international wage differential and providing employment to a skilled person who wants it.

The domestic worker who is laid off has a legitimate grievance. The visa worker who is hired into a system that will bind them to their employer for decades through the green card backlog has a legitimate grievance. The system makes compassion for either group deployable as an argument against the other. Advocate for the domestic worker and you are painted as anti-immigrant. Advocate for the visa worker's rights and you are painted as indifferent to domestic displacement. The framing works so well because it contains enough truth in each direction to feel unanswerable. Jay Gould's boast about hiring half the working class to kill the other half was not a business plan; it was a description of a system that pits two vulnerable groups against each other so that neither can organize against the employer.²⁴

The second layer is harder, and I've been circling it for a while, and I think the honest thing is to walk into it directly rather than keep circling... but it feels like touching a wet outlet.

There is a principle, old enough and widespread enough that it probably qualifies as a norm of political philosophy rather than a partisan claim, that a government's first obligation runs to its own citizens and residents.²⁵ France protects French workers. Japan protects Japanese workers. The H-1B's own prevailing wage requirement exists because Congress accepted this premise, at least on paper. Every version of the program has included displacement attestations and wage floors. The entire preceding section of this essay is about how those protections work in practice.

But the United States is also, in a way that France and Japan are not, a country whose identity is constitutively immigrant. The phrase gets printed on mugs and sewn onto throw pillows, but underneath the sentimentality sits a structural fact about how the country populates itself, replenishes its labor force, and generates the dynamism that its own economic mythology depends on. Roughly 45 million people currently living in the United States were born in another country. Their children, the second generation, number another 12 million. Together, immigrants and their immediate descendants account for roughly one in five Americans.²⁶ To say the government owes its first obligation to "its citizens" is to say something that sounds clean until you ask who counts, and when they started counting, and what happened to them in the decades between arriving and belonging.

And this is where the H-1B system creates a category of person that the clean formulations cannot accommodate. An Indian-born software engineer who arrived on an F-1 visa at 22, graduated from a U.S. university, worked under OPT, transitioned to H-1B, married, had children who are U.S. citizens by birth, bought a house, paid federal and state taxes for 15 years, and is currently 134,000th in line for a green card (an applicant from Iceland would face no meaningful wait; an applicant from India, which supplies 70 to 75 percent of H-1B approvals, faces a timeline a human lifespan may not contain²⁷): is this person a "foreign worker" whose interests the government may subordinate to those of citizens? Their children are American. Their mortgage is American. Their entire adult life has been American. The legal system classifies them as a temporary guest. Their children who turn 21 while the application is pending "age out" and lose their derivative status, potentially facing deportation from the only country they have ever known.

So the compassion problem has at least three sides. The domestic worker displaced by an employer exploiting the visa system's structural advantages. The newly arrived visa worker bound to an employer by deportation risk. And the long-resident visa holder trapped in a bureaucratic purgatory that makes a mockery of the word "temporary," whose interests are distinct from both the newcomer's and the domestic worker's, and whose existence the policy debate barely acknowledges because they fit neatly into neither the "protect American workers" frame nor the "welcome immigrants" frame. The employer, whose name appears on the LCA and the WARN notice and the quarterly earnings report, benefits from each of these three grievances and appears in none of them. The system's most reliable feature is that the conversation about who gets hurt never arrives at who profits.

They are already here. They have been here. The question of whether to admit them was answered fifteen years ago. The question the system has not answered is whether to recognize what their presence already means.

The essay form wants a resolution here. The reader, reasonably, wants one too. But the compassion problem does not resolve into policy. It resolves, if it resolves at all, only into moral clarity, which is a different and less useful thing: The domestic worker's grievance is legitimate. The visa worker's exploitation is real. The long-resident's limbo is unconscionable. Saying all three sentences in a row does not produce a fourth sentence that starts with "Therefore." It produces the recognition that the system generating all three outcomes simultaneously is working exactly as its accumulated incentives dictate, which is the most dispiriting sentence in this piece.

Dena Moore is somewhere in Orlando, working a different job for less money in a field where her experience has been devalued by the same arbitrage that eliminated her position. She trained her replacement. She signed the NDA. Her government had a statutory obligation to prevent exactly what happened to her. It certified the paperwork instead, within seven business days, because that is what the statute requires.

And somewhere in the queue, number 134,000 or so, there is the person who replaced her, or someone like her, who has now lived in this country for fifteen years, paid taxes into a system whose benefits they cannot fully access, raised children who are citizens of a country that considers their parent a guest, and built a career that their employer can end at any time by simply not renewing a piece of paper. The program calls them temporary. The word has been technically accurate for longer than some members of Congress have held office. Their children will turn 21 eventually, and when they do, the system will look at these American kids who grew up in American schools and speak with American accents and have never lived anywhere else, and inform them that their status was derivative all along, and the thing it was derived from has expired, and they should make arrangements.

Moore and the person in line have never met. They have no reason to think of each other as allies and every structural incentive to see each other as adversaries. This is the system's most durable achievement: two people, both worse off than they would be under an honest labor market, each visible to the other only as the cause of their problem, neither able to see the employer between them because the employer was never required to file both transactions in the same place. Martin Foran wanted to protect them both. Demi Moore would call it betrayal. The person in line would call it the cost of staying. Somewhere there is a quarterly earnings call where an executive describes the same arrangement as a strategic rebalancing of the global talent portfolio. The phrase contains a termination and a sponsorship and neither of their names.

Everyone who could say something knows. The Department of Labor knows. The executive knows. The immigration attorney knows. The HR director who wrote the severance agreement with the confidentiality clause knows. The silence has a headcount and it has excellent retention benefits.