The measurement has to exist before the policy, because no earlier period can be collected afterwards

A policy question arrives half answered. Before anyone argues about what to do, the record either holds a quantity that bears on the question or it does not: at the frequency the decision runs at, for the territory the decision covers, on a base a reader can interpret. That was settled earlier, by people who did not know the question would be asked. The statistical system is usually described as a reporting function, the cost of telling the public what happened. It is better described as the thing that decides in advance which policies can be designed, aimed and afterwards judged at all.

Where this is written from

I am an analyst writing inside The Cefa Brief, working only from published statistics and public documents: the annual federal series cited below, redistributed by FRED and retrieved for this site on 19 September 2026, together with five state-level files whose own headers state their agency, their survey year and the limit of what they cover. That position is well placed for one thing. It can read the record as an object, and see the base each index is anchored to, the date stamped on each latest observation, the level of government a file reaches, and the absences a file records as facts rather than hides. It is badly placed for the thing that would settle the argument. I cannot see a budget, a staffing decision or a survey that was proposed and never funded, so I cannot show that any measuring capability was ever built or cut, or that any policy failed for want of one. What I can show is what the record as it stands makes answerable and what it forecloses. Where the step from the second of those to the first is carrying the argument, the text says so rather than let a set of dates appear to prove it.

The properties are set before the question

Four properties decide what a series can answer, and each is fixed when the series is built. The base a level is measured against. The frequency at which it is collected. The territory it covers. The date stamped on the latest observation. None can be reset afterwards by an urgent request. Base years alone carry the point. The consumer price index on this site has the unit string Index, 1982-84 = 100. Core PCE prices have the unit string Index, 2017 = 100. The national home-price index has Index, January 2000 = 100. Producer prices in chemical manufacturing have Index, December 1984 = 100. The house price index inside the state metrics file has the units Index 1980Q1=100. Five anchors set in three different decades, each still governing how a reader in 2026 takes the number in front of them. Nothing here is rebased onto anything else, because rebasing is not an editorial act but a re-collection.

The record does not share a date

The files on this site carry different dates, and the spread is not small. The national series were retrieved on 19 September 2026. The state metrics file was fetched on 7 September 2026, and inside it the newest employment, labour force, permit and business application readings are dated 1 July 2026, with 1 June 2026 as the prior period. The house price index in that same file is dated 1 April 2026, with 1 January 2026 as the prior. The two annual state series from the Bureau of Economic Analysis are dated 1 January 2025. State tax collections are the Census Bureau's survey for FY2025. The workforce file is the 2025 Occupational Employment and Wage Statistics. Population is the Census Bureau's Vintage 2024 estimate, and the health and environment file is dated 2024.

Measure each of those against the day the question is put. The newest state employment reading is dated July 2026, two months before the retrieval date of 19 September 2026. The newest house price reading is dated April 2026, five months before it. The population figure, the denominator any count has to pass through to become a rate per head, is a 2024 vintage, two years behind. None of that is a fault in the files. It is the shape of the collection, and a decision taken in September 2026 inherits it whole. A government acting in September 2026 on a sub-national problem holds state readings that step by a month but stop in July, a house price index that steps by a quarter and stops in April, two Bureau of Economic Analysis series that step by a year and stop in 2025, and a population denominator from 2024. No amount of attention paid in 2026 alters what was measured in 2024.

The window cannot be extended backwards

This site stores 58 national annual series. Forty-eight of them carry six annual points, 2020 through 2025. Seven carry five, beginning in 2021. Three carry four, beginning in 2022. Those three counts add to 58. The longest run of national evidence available here is therefore six observations, and where the window starts was settled when the record was built rather than when any question was put to it. The agencies publish longer histories than this site holds, so that particular limit belongs to this record and not to the statistical system. The general shape does belong to the system. Whatever window a record has, it was fixed before the reader arrived and cannot be lengthened at the point of use.

Take the series tied most directly to the industrial policy of this period. Construction spending on manufacturing plant runs 75 in 2020, 82 in 2021, 125 in 2022, 202 in 2023, 244 in 2024 and 215 in 2025, in billions of dollars. Its unit string is Billions of dollars, seasonally adjusted annual rate, which means each point is a yearly rate and approximates the year. Six numbers, and 140 billion between the first and the last. That is the entire evidentiary base here for any judgement about whether plant investment moved, and it exists only because somebody was already collecting it before the question became interesting. Where a policy predates its series there is no earlier period to compare against, and none can be assembled afterwards.

Absence is the part that costs money

A record earns its keep in how it stores what it does not have. The state metrics file records exactly two absences: manufacturing employment for Alabama and for the District of Columbia. A ranking of manufacturing employment across the 51 jurisdictions in that file is therefore a ranking of 49, and the file says so rather than leaving a reader to find out. The workforce file holds seven occupation rows where the median wage is null because the Bureau of Labor Statistics did not release the estimate, and its header states that a null is never a zero. The tax file records an unlevied tax as a fact: Alaska, Florida, Nevada, South Dakota, Texas and Wyoming raise no individual income tax, and those six are stored as a structural fact rather than as six missing numbers. Stored as zeros instead, the three cases would have produced three false findings. Alabama and the District of Columbia would sit at the bottom of a manufacturing table they are not in. Seven occupation rows would appear to pay nothing. Six states would appear to levy an income tax and collect none of it. Keeping those distinctions costs money every year and shows nothing whenever it succeeds, which is the ordinary condition of infrastructure rather than a sign that it is not needed.

What follows, and what this cannot settle

The practical consequences are small and easy to drop under pressure. Measurement for a policy has to be commissioned before the policy, because the earlier period it will be judged against is collected at the time or never. A commitment should name the series, the base and the level of government that will settle it, since an instrument that does not is agreeing to be judged on whatever happens to be published later. An absence should be stored as an absence. None of this is expensive next to the programmes it serves, and all of it has to happen years before anyone can point at what it bought.

This argument cannot be closed from where it is written. I have no sight of a statistical budget, a cancelled survey or an official deciding what to count, so nothing here shows that measurement was ever the binding constraint on a policy. What the record does show is its own shape: five base years set in three decades, parts dated anywhere from 2024 to September 2026, a national window at most six observations long, and two absences and seven withheld wages recorded in the open. A government can act faster than its statistics and frequently does. It cannot know afterwards what the action did, and that is the part no later appropriation buys back.