Headline CPI prints first at 8:30 ET on release morning and lands on every financial-news front page within minutes. The Committee reads three other numbers before it reads the headline. The Cleveland Fed trimmed mean CPI, the Cleveland Fed median CPI, and the core services ex-housing line (the “supercore”) all release on the same morning, derived from the same BLS microdata. Each was built to strip a different kind of noise out of the headline print. Reading all three together is closer to what the FOMC staff brief actually contains than any single number is.
The trimmed mean CPI cuts the top 8 percent and bottom 8 percent of the monthly category-level price changes by expenditure weight, then averages the rest. The mechanic answers a specific question: what is the central tendency of price change this month, after removing the categories where supply shocks or one-off moves dominated. In May, the Cleveland Fed trimmed mean printed at 2.9 percent year over year, down from 3.1 in April. That is a one-tenth move, but the year-over-year series has now declined for six consecutive months, which is a cleaner disinflation signal than the headline or core series have produced over the same window. The trimmed mean is the measure that responds least to airfare swings, used car cycles, and energy pass-through. It is the measure that responds most to broad-based wage and rent pressure. When the Committee says it is watching whether inflation is “broadly” easing, this is the line.
The median CPI takes the same category-level distribution and reports the price change of the median expenditure-weighted category. It throws out even more of the distribution than the trimmed mean does. The mechanic is more conservative: the median sits inside the central 50 percent of category moves, and it changes only when the middle of the distribution moves. In May, the median CPI printed at 3.4 percent year over year, down from 3.5 in April. The median has been the stickiest of the three measures across the 2024 to 2026 disinflation, and it remains the measure furthest above the 2 percent target. The signal the Committee reads off the median is whether the disinflation is making it through the shelter and services categories that anchor the middle of the distribution, not just the goods categories at the tails.
The supercore (core services ex-housing) is constructed differently. It is not a trim or a median. It is a direct sub-aggregate of the headline CPI services index, computed by removing both energy services and the shelter components (rent of primary residence and owners’ equivalent rent). What is left is the services line items most tightly tied to wage pass-through: medical services, transportation services excluding airfare, recreation services, education and communication services, and other personal services. The supercore is the closest CPI proxy for the labor-cost transmission the Committee actually targets. In May, the supercore printed at 3.7 percent year over year, down from 3.9 in April. The three-month annualized rate, which is the rate the Committee weighs more than the year-over-year on the supercore line specifically, printed at 3.1 percent in May, down from 3.4 in April.
The relationship between the three measures is where the signal lives. When all three move in the same direction by similar magnitudes, the underlying inflation process is moving and the Committee can treat the headline print as confirming the direction. When the three diverge, one of them is being driven by a category-specific shock and the others are showing the underlying state. The May release clustered the three more tightly than at any point since early 2024: trimmed mean down 0.2, median down 0.1, supercore down 0.2. The headline core CPI print, by comparison, was down 0.1 year over year and showed a 0.18 percent month-over-month change, neither of which by itself would have moved the September pricing. The cluster of the three is what moved it.
The trade-offs across the three measures matter for how to read them on any given month. The trimmed mean is the most responsive measure: it picks up directional changes one to two months earlier than the median, and it can swing sharply in months where the underlying distribution shifts even if the median does not. The median is the most stable measure: it confirms what the trimmed mean signals, with a lag of one to three months, and is the harder number to move. The supercore is the most policy-relevant measure: it is the line the Committee minutes reference more than any other CPI sub-aggregate when the Committee is discussing whether wage growth has stopped feeding inflation. When the supercore three-month annualized rate sits below the year-over-year rate, the underlying trajectory is improving relative to the trailing twelve, which is the condition the Committee has been waiting for to anchor a cut.
The May release moved each of those conditions in the same direction. The trimmed mean confirmed the six-month disinflation trend. The median printed its second consecutive decline after three sticky months. The supercore three-month annualized rate (3.1 percent) fell decisively below its year-over-year (3.7 percent) for the first time since November 2025. Fed-funds futures repriced the September cut from 76 percent to 82 percent across the morning, and that repricing was driven mechanically by the cluster of the three measures, not by the headline core print itself. The 10-year auction at 1:00 ET cleared with a 0.4 basis point tail and indirect bidders at 67.2 percent, both of which are inside the cut-confirming side of the November versus September auction history. The PPI services release Thursday is the next variable in the sequence, and the June 18 dot plot is the variable the whole sequence resolves into.
The one-line read for the morning: the headline number is what the news leads on, the three measures are what the Committee leads on, and on this release the three measures agreed.