September 30, 2025. Midnight. The government shuts down.
The next morning, roughly 73 million Americans — one in five people in this country — are waiting on a Social Security check whose size will be determined by a number the Bureau of Labor Statistics (BLS) can no longer produce. The CPI (Consumer Price Index) report that calculates their annual cost-of-living adjustment (COLA), the one that tells the Social Security Administration (SSA) how much prices rose to the hundredth of a percentage point, is delayed. The Federal Reserve has no official inflation data to act on. Markets are rotating between ADP (Automatic Data Processing), Revelio, and Truflation, asking which private number to trust, knowing none of them are the instrument the law requires.
This is not a financial crisis. It’s a Thursday morning when a $700 million government agency went dark — and the question of who pays for the silence got answered in a way most people missed.
The consensus has two stories, and they’re both missing the same thing
The mainstream narrative comes in two flavors.
The first is efficiency. Federal statistical agencies are slow, methodologically outdated, and chronically underfunded relative to what the private sector can now produce. Revelio Labs aggregates 100 million employment profiles in near-real-time. ADP processes actual payroll data rather than survey estimates. The market is building the economic measurement infrastructure of the 21st century, and good for it.
The second is politicization. Trump fired the BLS commissioner hours after a weak jobs report, nominated a Heritage Foundation economist who has called BLS data “BS” and described Social Security as a Ponzi scheme — a nomination later withdrawn amid Senate opposition — and the whole sequence is a coordinated attack on a data source that inconveniently shows what it shows.
Both frames have real evidence behind them. Private data has genuinely improved. The political pressure on the agency is documented and alarming. Give each its due.
Here’s what both miss: the structural logic underneath either story. Neither frame asks who pays when the public instrument degrades. Neither frame explains why the private companies filling the gap are, to a firm, arguing that the public instrument needs to survive.
The working hypothesis: Defunding public data infrastructure doesn’t eliminate the cost of producing that data. It transfers and concentrates it — to taxpayers paying more per unit as public capacity is replaced by private contracts, to benefit recipients absorbing hidden cost through measurement error, and to institutional investors paying premium prices for data the public once got free. The NOAA weather contract record shows this mechanism already completed one full cycle. The BLS is the live version, higher stakes and still unfolding.
The proof of concept: what already happened to weather data
The mechanism has already run to completion once, and it ran on weather.
The National Weather Service costs every American about $4 a year. Return on that investment, according to the same American Meteorological Society analysis: 73:1. Every private weather app, every airline rerouting decision, every insurance actuarial model runs on NOAA data as its baseline. Private forecasting companies built businesses on top of that free public layer — not instead of it. AccuWeather, the company most historically associated with calls to commercialize the National Weather Service, publicly stated it does not support full privatization. It needs the free public data to exist.
Peter Neilley, Senior Vice President of Global Forecasting Services at The Weather Company, described the whole arrangement as a three-legged stool: government, academia, and private sector. “Weaken any leg,” he said, and an “informed and resilient society” is in jeopardy. John Dean, the CEO of WindBorne — a venture-backed company now literally filling NOAA (National Oceanic and Atmospheric Administration) data gaps under contract — was blunter still: “We’re not ready to fully replace radiosondes now, and it’s not obvious that we ever will be. I don’t think you’d see any kind of private company wanting to see NOAA services degraded.”
Dean said this while announcing that WindBorne was expanding to plug holes left by the NWS, which had suspended or reduced weather balloon launches at eleven locations after DOGE (Department of Government Efficiency) cuts eliminated over 880 NOAA employees in an initial round, with further reductions bringing total planned cuts to roughly 1,900. The company filling the gap said, in the same breath, the gap shouldn’t exist.
Here’s the number that makes the mechanism concrete. Spire Global (NYSE: SPIR) held a $3.8 million NOAA contract for atmospheric satellite data in September 2024. One year later — as NOAA’s internal capacity was being cut — that same data category renewed at $11.1 million. Nearly three times the price. In twelve months.
The government didn’t stop paying for atmospheric data. It started paying a private company three times as much to get it back.
There’s a condition buried in some of these contracts that makes the cost transfer permanent rather than temporary. NOAA has historically made all its data freely available to anyone — forecasters, researchers, foreign weather services, the public. When it buys commercial data instead, it can’t always do that. The first Trump administration signed a 2020 contract that prevented NOAA from publicly releasing a commercial hurricane forecasting model for five years. The private company sold the government data for its internal use and retained the right to sell that same data to private buyers. The government paid for it. The public didn’t get it. That’s not an edge case in how the transition from public to private data works — it’s the mechanism by which a public infrastructure becomes a tiered product, and the tier you’re on depends on what you can afford to pay.
There’s a second cost hiding inside this one. Traditional NOAA weather balloons cost roughly $200 per launch and fly for two hours. WindBorne’s AI-guided balloons are cheaper per observation and stay aloft for months — a genuinely useful technology. But they measure differently. Radiosondes collect a precise vertical profile from a fixed location, the same way, twice a day, since the 1930s. That consistent historical record is what makes climate research possible. The private substitute is cheaper per data point in isolation and can’t replace the network in practice. You’re paying more for coverage that’s different, not equivalent. The word for that, from a taxpayer standpoint, is less.
The live version: BLS and the instrument that governs $1.5 trillion
The weather data story is complete. You can see the before and after in a single contract renewal. The Bureau of Labor Statistics story is still running — same mechanism, different instrument, higher stakes, and no Spire contract to point to yet because the transition hasn’t finished.
The Bureau of Labor Statistics produces the Consumer Price Index. That number governs Social Security cost-of-living adjustments for roughly 73 million Americans — one in five people in this country, including about 87% of everyone over 65. It also governs Treasury bond interest rates, IRS tax bracket adjustments, SNAP benefit levels, and the Federal Reserve’s rate-setting decisions. Its 2025 budget: $700 million. Its staff: roughly 2,000 people, down nearly 25% by year’s end.
When field economists can’t reach stores, can’t conduct surveys, can’t collect prices in person, the BLS fills the gap with two flawed proxies: statistical imputation — educated guesses derived from prior data patterns — and automated web scraping. The agency discloses this. The share of imputed and scraped figures in the CPI rose drastically in 2025.
Each method bakes systematic error into the measurement in a different way. Imputation assumes the past is a reliable guide to the present; it fails whenever the economy shifts quickly. Web scraping is subtler and more insidious. An algorithm scraping a retailer’s website sees that a box of cereal costs $4.99 this month and $4.99 last month and records no price change. It misses that the box is now 14.1 ounces instead of 16. The BLS has its own research program for tracking this — what it calls “shrinkflation” — and it explicitly notes the methodology carries larger measurement error than standard collection. Manufacturers change package sizes precisely because consumers are more sensitive to sticker price than net quantity. A scraper catches the sticker. It can’t weigh the box.
Nearly a third of the measurement that determines Social Security checks is now built on interpolated assumptions and blind digital scrapes rather than actual prices collected at actual stores by actual people who no longer work there.
Now the error sensitivity. A 0.1 percentage point error in the CPI — just one-tenth of one percent — misallocates $1.5 billion per year in Social Security payments. The Social Security Administration paid about $1.5 trillion in benefits last year. That leverage ratio — a tenth of a percent of measurement error against $1.5 trillion in obligations — is the number that should focus the mind.
Here is where it becomes visceral.
If the CPI understates inflation — if the instrument running on 30% imputed data misses the actual rise in prices — 73 million Americans receive a cost-of-living adjustment that doesn’t cover the actual cost of their lives. Their checks fall further behind real prices every year. A COLA that comes in 0.2 percentage points below actual inflation means a retiree gets 2.5% less in purchasing power after a decade, 5% less after twenty years, 7.5% after thirty. The compounding is invisible until it isn’t — until a senior who was comfortable at 75 is choosing between prescriptions and groceries at 85.
If the CPI overstates inflation, taxpayers fund benefits above actual cost-of-living increases, drawing down the Social Security Trust Fund faster than the actuaries projected. Both errors are made more likely by an agency running at 30% imputed data with a quarter of its staff gone.
The BLS budget that governs this instrument: $700 million. The annual payout it calibrates: $1.5 trillion. That’s a 2,000:1 leverage ratio on measurement error. The proposed 8% budget cut — $56 million — doesn’t save $56 million. It exposes $1.5 trillion in annual obligations to the degradation cost.
The private companies aren’t measuring the economy. They’re measuring BLS.
When the government shut down in September 2025, institutional investors didn’t go without data. Apollo Global Management published a 75-page guide to private alternatives. The Fed acknowledged leaning on ADP and other private sources. Revelio Labs — which launched its public labor statistics product the day before the August BLS jobs number dropped, explicitly timed to the commissioner firing — positioned itself as the institutional alternative.
Claudia Sahm, creator of the Sahm Rule recession indicator and a former Federal Reserve economist, wrote the structural problem plainly: private companies “will tend to reflect the needs of their clients — they won’t capture the economy as a whole.” ADP covers what its clients are, and ADP’s clients skew heavily toward large formal employers. Revelio harvests online employment profiles — company websites, LinkedIn, Indeed. The restaurant worker, the home health aide, the seasonal contractor, the small-town retailer — none of them appear in Revelio’s dataset. These are precisely the workers whose labor conditions most directly shape Social Security enrollment, SNAP eligibility, and the downstream benefit calculations that BLS data governs.
But the structural problem runs deeper than coverage gaps.
Revelio’s headline marketing number is that its dataset has a 0.74 correlation with BLS establishment survey figures. That correlation is the product’s credibility claim. The Federal Reserve Bank of Richmond found it in practice: private estimates diverged significantly from official BLS figures in three of every twelve months — accurate enough when the economy is stable, unreliable exactly when you need it most. The Richmond Fed found private estimates “tend to be less precise during unusual events such as the pandemic,” which is precisely “when a sharp policy response may be required.”
Here’s what that 0.74 actually means. Revelio built its model against BLS outputs. It validated its methodology against BLS revisions. It publishes monthly figures explicitly indexed to what the BLS would show. The private alternative is calibrated against the public instrument it’s supposedly replacing. As the BLS degrades — more imputation, smaller samples, fewer field economists — the benchmark Revelio is correlated against degrades with it. The private data ecosystem claiming to replace public infrastructure is dependent on that infrastructure remaining functional enough to serve as its own calibration floor.
And then there’s the ADP pullback.
In October 2025, ADP stopped providing client-level data to the Federal Reserve Board. Not because of the shutdown — the timing was concurrent but the cause was separate. ADP made a commercial decision about data access after a Fed governor publicly referenced the data in a way ADP found objectionable. No statute governs this. No public accountability mechanism exists for it. The central bank had built a dependency on a private alternative — and the private alternative exercised its right to pull back, for reasons entirely its own.
This is the self-sealing loop. Cut public data. Private alternatives fill the gap. Private alternatives calibrate against the degraded public signal. The degraded signal becomes the floor, not the ceiling. Cut public data further, and the private alternatives lose their anchor. At some point there is no reliable benchmark — not for the Fed, not for the SSA, not for the hedge funds paying into a $14 billion alternative data market that is, at bottom, an elaborate derivative of the public instruments it claims to be replacing.
Who pays, and how much
The costs are real but they fall differently depending on where you sit.
For every American taxpayer, the NWS costs $4 per person per year and returns $9 to $73 for every dollar spent. Spire’s NOAA contract for atmospheric satellite data jumped from $3.8 million to $11.1 million in a single contract year — a 192% increase — for data the agency previously generated internally. But the dollar amount is only part of the cost. Once public infrastructure is dismantled, the government becomes a captive buyer. It can no longer walk away from the contract, produce the data itself, or negotiate from strength. That’s not paying twice. That’s paying a premium markup to a private vendor while surrendering the leverage that would let you stop.
For the 73 million Social Security recipients — one in five Americans — the exposure is $1.5 billion per year in benefit miscalculation per 0.1 percentage point of CPI error. They have no private alternative. They cannot subscribe to Truflation and ask the SSA to use it instead. They are fully dependent on the accuracy of an instrument now running at 30% imputed data, administered by an agency with a fourth of its staff gone and no confirmed commissioner.
For institutional investors, the cost is a $14+ billion alternative data market that exists largely because public data is degrading. Revelio raised $15 million in its Series A — with Barclays as an investor — to build a product that packages a degraded public good and sells it back to institutional clients at subscription prices. A bank is financing the privatization of something it will resell to the same clients the public data once served for free.
For the Federal Reserve, the cost is structural. It is legally mandated to target inflation using measures derived from BLS data. During the shutdown it was reduced to supplementing with private proxies that correlate 0.74 with an instrument it could no longer access. That 0.74 is not a replacement. And when it needed it most, the proxy’s provider turned it off.
One more number worth sitting with. The monthly CPI series had been published continuously since January 1921. The unemployment rate since January 1948. October 2025 is now a permanent partial blind spot in both records. Not a delay. Not a revision that will arrive later. The data collection didn’t happen, and it cannot be collected retroactively. Seventy-seven years of unbroken CPI history. Gone — not because the measurement was impossible, but because the agency measuring it was sent home.
The investor companion to this piece, The Captive Buyer Trade, covers the market implications for Spire Global’s government contract revenue.
What would change my mind
1. BLS imputation rates in the CPI return to below 15% within 18 months of a new confirmed BLS commissioner taking office — evidence that new leadership restored collection capacity rather than further degrading it.
2. Spire Global’s NOAA contract value decreases year-over-year in FY2027 while NOAA’s internal headcount stabilizes above 2024 levels — evidence that the private substitution dynamic is reversing rather than accelerating.
3. The private employment data sources — ADP, Revelio, and LinkUp — maintain or reduce their divergence from each other over four consecutive quarters through mid-2027. The Richmond Fed publishes this comparison monthly. If the private sources stay coherent with each other even as BLS degrades, the calibration anchor is holding. If they start fragmenting — disagreeing not just with BLS but with each other — the baseline has slipped, and what I’m describing here is already happening.
If you found this useful, the best thing you can do is forward it to one person who would push back on it. I’d rather be wrong in public than right in private.