The Verification Venue · a fear pointed at the wrong number
The Ruler That Only Bent
The news says inflation came back down. Your receipt says nothing came down. Both are right, because they are reading two different rulers. The rate is how fast prices climb; the level is how high they have already climbed. When the rate fell, the level did not: it just climbed more slowly. Here is the whole thing in one instrument.
In June 2022 the U.S. inflation rate hit 9.1%, the fastest in forty years. A year later it was 3.0%. That drop is real, and it is not the same as prices falling. A level is the running total of a rate, so a rate that is smaller but still positive keeps adding to the total. Grab the playhead below and drag it from mid-2022 into mid-2023: watch the blue inflation dial fall while the gold price line keeps ticking up. Every number here is recomputed live from the actual monthly Bureau of Labor Statistics series bundled into this page, not quoted from a headline.
The two rulers · BLS CPI-U, all items, U.S. city average, 1982-84=100, not seasonally adjusted · Jan 2019 to Dec 2024
Drag across mid-2022 into mid-2023: the rate dial falls from 9.1 to 3.0 while the level keeps climbing.
All items never fell. Energy and gasoline are the honest exception: their level did dip from the peak (though even they stayed well above 2021).
The level then reads as cumulative percent since your chosen base month.
The counterfactual asks what it would take for prices to actually come down. The wage overlay shows the real fix: pay catching up.
-
-
-
-
The reorder never happens. Flip to Rate only and you see a mountain that crashes back down. Flip to Level only and you see a line that bends its slope but never turns down. Same data, two rulers, and that is exactly why the two get confused: people hear the mountain coming down and picture the line coming down with it. It does not, because the level is the running integral of the rate, and the rate never went negative.
So we just need prices to fall? That wish has a name
The natural next thought is: fine, then bring the level back down. Turn on the Deflation counterfactual. To erase the 2021-2024 surge, prices would have to fall at roughly the same 9% pace they rose, for about two years without a break. That is deflation, a sustained negative rate, and mainstream economics fears it. This is why the Federal Reserve targets 2% inflation, not 0% and emphatically not below. The fear is concrete: when prices fall, the real weight of every fixed-dollar debt (mortgages, student loans, business loans) rises, so people cut spending to service it, which pushes prices down further (Irving Fisher called this the debt-deflation spiral in 1933); big purchases get postponed because they will be cheaper next month, so demand and jobs fall; and central banks lose their main tool, because nominal interest rates cannot go far below zero. Falling prices is not the happy ending. It is the thing the 2% target exists to prevent.
Then what actually fixes high prices? Pay
Turn on the Wage overlay. During the surge, prices outran pay: nominal average hourly earnings kept rising, but the price line rose faster, so inflation-adjusted (real) pay fell through 2021 and 2022, and the typical paycheck bought less even as the dollar figure on it grew. From about mid-2023 that flipped: pay began rising faster than prices, and real earnings started climbing back. The repair for an elevated price level is not the level coming down, it is wages rising past it until the higher prices become affordable again. That turn has begun in the aggregate. Whether a particular household is whole again is uneven and depends on which pay measure you read, which is the honest scope, spelled out in the check below.
The check — every number recomputed in front of you, from the bundled BLS series
1. The peak and the year after. The 12-month rate is level(t) / level(t-12) - 1, computed straight from the bundled monthly index:
2. The core identity. Over the exact span in which the rate fell from 9.1% to 3.0%, the level did not fall, it rose:
3. Cumulative, and by basket. All items rose about 20.6% from Jan 2021 to Dec 2024 and never returned to that base. Energy is the honest exception: its level did fall hard from its 2022 peak, yet even energy stayed well above 2021.
| basket | Jan 2021 | Dec 2024 | cumulative | peak → later trough |
|---|
4. Disinflation is not deflation. The right test for "did prices fall" is whether the 12-month rate ever went negative. From the June 2022 rate peak through Dec 2024 it never did:
5. The level is the running integral of the rate. Rebuild the level from scratch by compounding each month's growth from the base, and it reproduces the actual index to the last decimal:
6. Real wages: fell, then recovered. Real average hourly earnings = nominal pay divided by the price level. Deflating the bundled earnings series shows the 2021-2022 dip and the post-2023 climb (direction, not a single magic magnitude, is the robust claim):
The offline verifier recomputes all of this from the same numbers and checks it against what this page shows. Run it: node research/why-prices-stay-high-when-inflation-falls/verify-why-prices-stay-high-when-inflation-falls.mjs
The free choices and the honest scope
Which series, exactly. The rate and level use the CPI-U, all items, U.S. city average, 1982-84=100, not seasonally adjusted (series CUUR0000SA0). This is the series whose 12-month changes BLS reports as the headline 9.1% and 3.0%. A seasonally adjusted, annual-average, core (ex food and energy), or PCE measure gives a different peak (PCE peaked lower). We name the series and do not silently swap it.
Rounding. Computed from the index, the June 2022 rate is 9.06% and the June 2023 rate is 2.97%; to one decimal these are the 9.1% and 3.0% BLS publishes. The level change over that span is +2.97%, which we describe as "about 3%."
"Prices never fall" is true for the aggregate, not every item. Turn the basket to energy or gasoline and the level visibly dips: energy fell about 21% from its June 2022 peak, gasoline about 37%. Yet even those stayed roughly 30% above January 2021, so "prices came back down" is false even there. Food at home and shelter never fell back to their pre-surge base at all. The month-over-month all-items line does tick down in about 8 of 30 months across the disinflation, but those are small seasonal dips (autumn gasoline); year over year the level was never once lower than a year before. Seasonal wiggles are not deflation.
Real wages are measure-dependent, and we do not overclaim. We bundle nominal average hourly earnings of all employees, total private (seasonally adjusted, series CES0500000003) and deflate by the bundled CPI-U. That gives a real decline of about 1.4% from Dec 2021 to Dec 2022 and a rise of about 1.3% from Jan 2023 to Jan 2024. When BLS first reported these in early 2023 and 2024 the headlines were about -1.7% and +1.4%; subsequent benchmark revisions to the payroll data have nudged the magnitudes, which is exactly why we recompute from the current series and lead with the direction (down then up), not a single figure. The magnitude also depends on the measure: average versus median hourly earnings, average hourly earnings versus the Employment Cost Index, the start month, and composition effects all move the answer. "Real wages recovered in the aggregate" is defensible; "every household is made whole" is not, and we do not claim it.
The deflation counterfactual is a transparent projection, not a forecast. It takes the last bundled month and projects the level downward at a steady annual pace equal to the June 2022 peak rate (about 9%), then marks where it would cross the January 2021 base. It is there to show how much sustained deflation "undoing" the surge would require, not to predict it.
Deflation-is-harmful is the mainstream view, not a theorem. Debt-deflation (Fisher 1933), deferred demand, and the zero lower bound are the standard case for a positive inflation target. There is a minority academic argument that mild, productivity-driven "good deflation" can be benign. We present the mainstream rationale and do not claim every price decline is automatically catastrophic.
What this page is not about. It is about the arithmetic of level versus rate, not about assigning causes (supply shocks, stimulus, corporate pricing). Those are real debates and separate from the load-bearing fact here, which is that a falling positive rate still raises the level.
Data vintage. The bundled series were retrieved from the BLS public API on 2026-07-17 and span January 2019 through December 2024. CPI values are revised and rebased over time; the verifier pins the exact index values so any future drift is caught.