How the three composite scores — Momentum, Forward Stress, and Inflation Regime — are built, scored, and read, and where the framework can fail honestly.
The State of the Economy is the most detailed feature on USMacro.com. It reads where the US economy stands each month, on a monthly-to-quarterly horizon: growth, forward stress and inflation, as three gauges, and the stage of the cycle they place it in. It asks one question of that reading: what stage are we in, and which way is it shifting. Its record is every stage period since the gauges begin and what followed each one, shown as counts and ranges on the Regimes page. It says what the record shows; it does not say what anyone should buy or sell.
It reduces a large body of economic data to three scores, each on a 0–100 scale, plus a set of views that show where those scores have been and what historically tended to follow.
Why three composites and not one number. A single "economy score" cannot tell apart two situations that demand opposite responses. Take a clean recession where the Federal Reserve can cut interest rates aggressively (2008), and a stagflation where the Fed is trapped by high inflation (2022). On a growth-and-risk reading alone the two can look nearly identical, yet they are completely different for an investor. It takes a third dimension, the inflation environment, to separate them. Hence three scores:
We don't curve-fit, and the framework can fail honestly (the Falsifiability Principle). Every input to every composite has to be defensible on its own merit. We do not add "modifiers" or "fudge factors" to make a composite look like it works. If a composite fails a historical test, the rule is to find out which underlying assumption was wrong and fix the design — not to patch the output until it looks right. The reason is simple: a model held together by patches can't teach you anything when it fails, because you can't tell which part was wrong. A clean composite that fails honestly is more valuable than a tweaked one that appears to work. (This mirrors how the Conference Board maintains its Leading Economic Index — clean methodology, no period-specific patches.)
The lines you'll see on the charts at 40 and 60 are the gauges' own bands: fixed marks on the 0–100 scale that divide a gauge's readings into low, middle and high. The math is never adjusted to make a score land on one. They were set in advance, as working hypotheses, and tested only later, on the first-published record — at least 85% first-published from Sep 2016; earlier years partly use today's revised data, read only once it had been published (how much). Above 60 did not mark better years for stocks on any gauge: after a month at 60 or above, the next year's median return ran from 4.3 points worse to 0.9 points better than after other months. Below 40 came before recessions on no gauge (Momentum none of five, one false alarm; Forward Stress one of five, three false alarms; Inflation Regime none of five, eight false alarms).
A note on transparency: every weight is disclosed — the structure, the individual indicator weights, and their effective contribution to each score. The same effective weights appear on the site's component heatmap. What stays private is the operational implementation, not the methodology: how the framework is built, maintained, and evolved. The reasoning is here to be judged — that openness is the point.
Every composite is built the same way at its core. For each indicator, we ask: where does today's reading sit within its own history? An indicator at its 80th percentile is higher than it has been 80% of the time in the past. This converts every input — whatever its native units, a rate, an index, a dollar figure — onto a common 0–100 scale. A composite score is then a weighted average of its inputs' scores. For Forward Stress and Inflation Regime, 50 means every input at the median of its own look-back period — not the midpoint of the gauge's own record, which sits above 50: since 1981, Forward Stress's median reading is 55 and Inflation Regime's 66. Momentum ranks its inputs around zero growth instead, as set out below.
The look-back window for that ranking is 10 years for Forward Stress, recent enough to reflect today's structural era. It is 20 years for Inflation Regime and Momentum: inflation regimes evolve slowly, and Momentum's rare-contraction scoring rewards long memory. Because the window trails, a given score's meaning adapts to the era: an 80 means "high relative to the last decade (or two)," not relative to all recorded history. This is deliberate: the alternative, ranking against fixed full history, would let long-dead regimes (say, 1980s rate levels) distort what counts as extreme today.
Where an indicator is bad when high (credit spreads, for instance, where wider is worse), its percentile is inverted before entering the composite — so that across all three scores, higher always means healthier.
Built from 9 coincident indicators across 4 pillars, weighted toward services — 45% — because services are roughly four-fifths of the US economy and the part whose breakage separates real recessions from sector scares.
| Pillar | Indicator | Effective weight |
|---|---|---|
| Services · 45% | Private services payrolls ◆ | 18.00% |
| Services aggregate weekly hours | 13.50% | |
| Real PCE services (consumption) | 13.50% | |
| Goods · 20% | Industrial Production ◆ | 13.00% |
| Goods-producing payrolls | 7.00% | |
| Household · 20% | Real Retail Sales | 11.00% |
| Real personal income ex-transfers | 9.00% | |
| Labor stress · 15% | Sahm Rule (real-time) ◆ | 8.25% |
| Continuing jobless claims | 6.75% |
◆ required anchor — the composite refuses to publish without it. Growth inputs enter as 3-month annualized rates; the Sahm Rule enters as a level.
The zero-anchored scorer — why Momentum ranks differently. A plain percentile rank fails here: services growth is rarely negative, so ranking growth rates puts a mild dip and a collapse in the same bottom tail. At worst, services payrolls shrank at about a 2% annual rate in 2001. In 2008–09 they shrank at more than 5%. We measured exactly that failure and redesigned the scorer around the zero line, the one boundary with real economic meaning. Positive growth is ranked 40-100 against the past 20 years of positive readings, while contraction is ranked 0-40 against every contraction in the series' full history. As a result, a mild dip reads mild and a collapse reads like the collapses it actually resembles. The score of 40 is therefore the expansion/contraction anchor, not a historical median.
How to read it against Forward Stress. Forward Stress is the leading gauge. Forward Stress's 12-month change fell 15 points or more between 10 and 12 months before each of the four recessions its record reaches — and five times with no recession after it (1995, 1996, 1998, 2015, 2022). Momentum is the coincident gauge. Momentum was already within 10 points of its lower band (40) three months before three of the five recessions, then fell below 30 within four months each time. The grammar: above 60 — solid expansion. 40–60 — the middle band; the regime's Slowdown line sits inside it, at 50. Below 40 — weak; a fast fall here, with Forward Stress also low, meets the Contraction rule (Section 10). Below 30 — the economy is contracting now. Momentum has been below 30 for two months running ten times since 1964, every time inside a recession or in the month after one ended — no false alarm, on a record fully first-published only from 2013 (how much).
The most complex composite: 15 leading inputs across 5 pillars, combined in two levels (within each pillar, then across pillars). Built entirely from forward-looking signals.
| Pillar | Indicator | Within pillar | Effective |
|---|---|---|---|
| Monetary & Financial Conditions · 30% | Yield curve (10yr–2yr) ◆ | 45% | 13.50% |
| Excess Bond Premium (EBP) | 30% | 9.00% | |
| Baa–10Y credit spread | 10% | 3.00% | |
| Business-loan delinquency (DRBLACBS) | 9% | 2.70% | |
| Business-loan charge-off (CORBLACBS) | 6% | 1.80% | |
| Banking & Credit · 12% | Bank lending standards (SLOOS) | 70% | 8.40% |
| Commercial paper – T-bill spread | 30% | 3.60% | |
| Housing · 20% | Building permits ◆ | 65% | 13.00% |
| Housing starts | 35% | 7.00% | |
| Business & Labor leading signals · 25% | Real capital-goods orders (NEWORDER) ◆ | 45% | 11.25% |
| Manufacturing weekly hours (AWHMAN) | 30% | 7.50% | |
| Initial jobless claims (ICSA) | 25% | 6.25% | |
| Global Conditions · 13% | Trade-weighted dollar (DTWEXBGS) | 50% | 6.50% |
| OFR EM financial-stress index | 35% | 4.55% | |
| G7 leading index (CLI) | 15% | 1.95% |
◆ required anchor — the composite won't compute without it.
The pillar weights reflect a deliberate ordering: the Monetary & Financial Conditions pillar, anchored by the yield curve, carries the most weight because it holds the most established leading recession signals. The within-pillar weights follow the same logic — each pillar's anchor indicator is weighted heaviest, with supporting signals scaled beneath it.
Why it's inverted. Forward Stress measures risk, so a rising reading is bad. To keep all three scores oriented the same way, it is displayed inverted: a high Forward Stress score means low stress. When the score falls, stress is building.
Capital-goods orders are deflated to real terms. New orders are reported in nominal dollars; during the 2022 inflation surge, nominal orders printed near record highs while real orders were weak. Read nominally, that would have looked like strength when the real signal was stress. The composite divides new orders by a capital-equipment price index so it reads the real signal — a correction whose effect on the historical score is genuine and material.
The composite most often misread, so read this section carefully. Built from 5 indicators:
| Indicator | Weight | Role |
|---|---|---|
| Core PCE, 12-month change | 35% | Regime anchor — the Fed's preferred measure |
| Core PCE, 3-month annualized | 20% | Recent momentum / inflection detection |
| Cleveland Fed Median CPI | 20% | Persistence / breadth (filters tail noise) |
| 5-year breakeven inflation | 15% | Market-implied expectations |
The score also uses one input, weighted 10%, whose data this site does not publish.
Rolling up: 55% realized inflation (level + recent momentum), 20% persistence, 25% expectations. The realized weighting reflects that realized inflation explains the majority of Fed rate decisions, with expectations explaining the rest. Core PCE is used rather than Core CPI because the Fed's 2% goal is stated in PCE terms, and because CPI's much larger shelter weight distorts the signal during housing-driven inflation.
The U-shape — the key idea. Unlike the other two composites, Inflation Regime is not scored "higher is better" or "lower is better." It is scored on distance from the Fed's 2% target. Inflation near 2% scores high (healthy); inflation far from 2% in either direction scores low (concerning).
So the most useful way to read this composite is not "how high is inflation" but "does the Fed have room to cut?"
To be clear about scope: Inflation Regime is not a forecast of future inflation and not a display of the current inflation rate. It integrates several dimensions of the inflation environment into one reading of the Fed's room. The page names that reading in two parts: a level, from “unusually calm” to “severely strained”, and a direction — rising, stable or easing.
Throughout the page, plain-language descriptors are tied to specific score ranges so the words always mean the same thing. Momentum and Forward Stress share one ladder of five words: “unusually strong”, “healthy”, “neutral”, “weak” and “near historic lows”. Inflation Regime has its own: “unusually calm”, “calm”, “neutral”, “strained” and “severely strained”. Each word is one of the gauge's color bands. These are conventions, not opinions — the same score always produces the same word.
In July 2026 the State of the Economy's regime layer was tested against its own pre-registered criteria — written down, with kill conditions, before any measurement ran. The full protocol and every measured row are in the project's decision records; the essentials are here.
Recognition. Across the four NBER recessions the point-in-time timeline can be tested on (1990, 2001, 2008, 2020), the engine entered a Contraction label within one to five months of every NBER peak. It took five months for 1990 and one to two for the other three. By comparison, a placebo catches at most two of four by chance. Over the full 45-year timeline, the adjudicated register shows zero Contraction entries that were genuine false alarms: every entry either recognized a recession, occurred inside one, or preceded one.
Lead. The regime label itself is a recognizer, not a predictor — it follows the peak by one to five months. The leading behavior lives a layer up. On the point-in-time record, the trajectory reading turned "deteriorating" in the year before every one of the four peaks, and Forward Stress crossed below 45 before each. That record is only partly first-published for the three earlier recessions (how much).
Against simple rules. The framework was compared, symmetrically, to three public single-variable rules. The Sahm rule had comparable timing, with 1 false episode (July 2024) to our zero. A false episode is a firing — the rule reaching 0.50 after a month below it, since 1981 — neither inside a recession nor followed by one within twelve months. The 2s10s yield-curve inversion never fired within the test's 12-month window on these four recessions — its classic leads run longer than the window. An OECD leading-indicator rule was earlier on three of four recessions, at a cost of 18 false episodes to our zero. That is the July 2026 test's frozen count (decision record C3: a three-month fall in the G7 indicator), under the same twelve-month window. Longer leads can be bought; the price is false alarms.
Limits, stated plainly. Only the 2020 recession falls in the fully vintage-clean window — the earlier three are measured on progressively thinner point-in-time coverage (stated per row in the records). Recent entries are scored against NBER dates as currently declared, and NBER declares with about a year's lag. And the credit-stress input relies on bank-loan data that is known to understate market-led, sectoral, and policy-masked credit episodes (2001, 2015–16, 2020). That is a structural limitation this validation documents rather than repairs.
What happens next. Every quarter, what the site published is scored against what happened, as counts, and recorded. The first sitting, for the third quarter of 2026, was held on October 1–2, 2026. It graded the stress signal and the trajectory calls, and not the eleven retired labels. A historical validation, not a forecast.
The site presents the same underlying data through several deliberately separate views. No single view is "the signal" — the signal lives in the combination (see Section 7). Here is what each shows.
Reading Today (top panel). A short paragraph saying how each of the three gauges has moved over the last three months, in plain English. It is generated mechanically from the scores — the same inputs always produce the same text, by design (no AI writing, nothing that could drift or invent).
The trajectory chart — "Where are we in the cycle?" A 2D plane: Momentum on the vertical axis (low at bottom, high at top), Forward Stress inverted on the horizontal axis (high stress at left, low stress at right). Today's reading is a dot; the connected dots behind it trace the last 12 months. The plane divides into four quadrants — Strong · calm, Strong · stressed, Soft · stressed, Soft · calm. What each quadrant has coincided with on this site's record, and how the dot has moved between them, is set out in its own guide. The Four Quadrants → (The chart depends on Momentum, live since July 2026; it is on the site now.)
The long-horizon history — "The three gauges as first published" Each composite's full monthly history on the 0–100 scale, each month as it was first published — at least 85% first-published from Sep 2016; earlier years partly use today's revised data, read only once it had been published (how much), with the gauges' bands at 40 and 60 marked. Recessions are shaded, and the lines for each composite are overlaid. This is where you see whether a score is near a historically meaningful level and which way it's been trending over years, not months.
The historical-analogs view is no longer on the page. It showed the five past periods that looked most like today and what the stock market did next. We tested whether those matches said anything about what followed, and they did not; Section 5 sets out the test.
The component heatmap — "What's broken (and what's working)" Every individual indicator as a colored cell (the same color bands used elsewhere on the site), grouped by composite, over recent months. This is the "what's actually driving the score" view — it lets you see which specific inputs are stressed and which are calm, rather than only the rolled-up number.
Data freshness. Economic indicators release on different schedules: daily market series are always current, while monthly and quarterly series are published weeks after the period they cover. The inputs are therefore at different "ages" on any given date. A small grey tag next to an input's name (for example · 2mo) marks how many months old that input's latest data was as of the date you're viewing: the picked date in the Time Machine, or today in live mode. No tag means the input was current (zero months old) as of that date. That is always true for daily series, and also true for any monthly or quarterly series that happened to be up to date. The tag appears only on monthly or quarterly inputs that were genuinely a month or more behind. It rounds out the point-in-time picture — not just what each input read at that date, but how fresh that reading was. Hovering an input's name shows the exact data period. The freshness tag is separate from the * marker, which flags an input that fell back to revised data because its point-in-time archive didn't reach that far back. "Stale" and "fell back" are different signals, so on a starred row the freshness tag is omitted.
Carried inputs. An input reads carried when its source has gone a full publication period without producing anything new. For a monthly series, that means its latest observation is older than the month before the one being scored; for a quarterly series, older than the prior quarter. A carried input keeps its value and its full weight, because it is still the newest observation that exists and dropping it would make the reading less accurate, not more. What changes is that we say so: the input is labelled carried with the month it was last observed. And the composite's coverage figure counts only the weight backed by a current observation, so coverage stops reading 100% over data that has not moved.
The page tracks not just the level of each score but how it's moving.
This watches the one-month change in a composite. The firing line sits at about the fastest 8% of Forward Stress's monthly moves. The extreme tier sits at about the fastest 1%. Momentum's lines are derived from its own history by the same rule; Inflation Regime uses Forward Stress's, and because it moves more, they fire for it in about one month in five.
Separate from velocity, this tracks the three-month direction — deteriorating, stable, or improving. Forward Stress's threshold (±5 points over three months) and Momentum's (±10) each come from that gauge's own history; Inflation Regime uses Forward Stress's. It is the layer that catches grinding declines that never trigger the sharp-move velocity flag. Using only the data available then (how much was first-published), the two layers split the work. The 2020 episode was a cliff: Momentum's velocity fired at the extreme tier, down and then up. The 2007–08 episode was a grind: velocity stayed silent on Momentum and Forward Stress, while Forward Stress's trajectory read deteriorating in most months from September 2007. The 2022 episode was mostly a grind too: Forward Stress's velocity fired only once, in May. Both matter, and they read very differently.
Investor sentiment, such as the AAII bull-bear spread, is not an input to any composite, and the page shows no sentiment badge. It's a poor signal at market tops (euphoria can persist for years) but a strong confirming signal near bottoms (capitulation is sharp). Because of that asymmetry it's kept out of the composite math, with its real use reserved for the separate Capitulation Monitor (live — six oversold gauges with the regime as context).
Equity valuation, such as CAPE or the Buffett indicator, is not a composite input either, and the page shows no valuation badge. It is read on its own pages, under Valuations.
The analogs view asked "when has the economy looked like this before, and what happened next?" This section says how it found its matches, and then what happened when we checked whether the matches were any use.
The fingerprint. Each month in history is reduced to a point in 6D space: the three composite scores — where the economy stood — and each score's 3-month change — where it was heading:
fingerprint(month) = ( Momentum, Forward Stress, Inflation Regime · ΔMomentum₃ₘ, ΔForward Stress₃ₘ, ΔInflation Regime₃ₘ )
The distance. Position and heading are measured separately — each as a straight-line (Euclidean) distance across its own three dimensions — and then added. The two blocks live on different scales, so each is first divided by its own frozen median distance, and they are weighted half and half:
distance = 0.50 × (position distance ÷ median position distance) + 0.50 × (heading distance ÷ median heading distance)
Inside each block the three dimensions are weighted equally — there's no a-priori reason to favor one. The heading dimensions are each divided by their own composite's typical 3-month move first, so an ordinary swing in Momentum counts the same as an ordinary swing in Inflation Regime. Those scaling constants were measured once, over the whole universe, and are frozen in the engine: they are not recomputed as new data lands.
From distance to a "% match." The view turned each distance into a rank against every eligible past month, so a "99.7% match" meant closer than 99.7% of them. That made every match look nearly identical, which is one of the reasons the view is gone: a rank says how a match compares with other months, not how alike the two months are.
The rules the matching followed:
What it matches — and what it deliberately does not. The engine matches the point and the heading into it. Two months can sit at identical scores and still rank far apart here if one is improving into the position and the other deteriorating into it. That is the case the earlier point-only engine got wrong often enough to be retired. What the engine does not match is the full path — the whole shape of the year before. That variant was built and measured alongside this one and rejected: weighting every lag equally buries cliffs, and it paired March 2020 with March 2016. A three-month heading is the span the evidence supported, and the limit is a stated one rather than a hidden one.
Before looking at any result, we fixed the bar the matches had to clear. The middle of their five outcomes had to land closer to what actually followed than the middle of 5 months picked at random, in at least 60% of months. Their average miss had to be smaller than that of all 500 sets of random picks. And both had to hold in every decade on its own.
We then re-ran the view for every month from November 1986 to August 2026 — 477 months — using data as first published — at least 85% first-published from Sep 2016; earlier years partly use today's revised data, read only once it had been published (how much), and only past matches whose outcome was already known at the time.
| looking ahead | months tested | closer than random picks | average miss, ranked among 501 (1 = smallest) | decades that cleared the bar | closer than the middle of every past month |
|---|---|---|---|---|---|
| 3 months | 474 | 49.1% | 328 | 0 of 5 | 39.7% |
| 1 year | 465 | 50.4% | 27 | 0 of 5 | 39.4% |
| 3 years | 441 | 55.5% | 1 | 1 of 5 | 48.5% |
No horizon cleared the bar. The last column is the plainer comparison: taking the middle outcome of every past month, with no matching at all, came closer than the matches more often than not at every horizon.
The stage. For each month we also asked whether a Contraction began within the next 12 months, and compared the matches' answer with what the stage the economy was in had already shown. The matches scored worse — an error score of 0.127 against 0.124, where lower is better. Whatever the matches knew about a coming Contraction, the stage already knew.
So the matches the view would show for today have no tested meaning, and the view is no longer on the page.
Measured September 2026.
The trajectory chart reads against a four-quadrant plane (Momentum vertical, Forward Stress inverted horizontal). The quadrants — Strong · calm, Strong · stressed, Soft · stressed, Soft · calm — are split at 50 on both axes. The guide reads each month on the point-in-time record — at least 85% first-published from Sep 2016; earlier years partly use today's revised data, read only once it had been published (how much)(where the data published about five weeks after the month began put it) and counts every move from one quadrant to another.
A second way to count the moves. A dot that crosses a line for a single month and comes back counts as two moves. Counting only the moves where the dot then stayed at least 3 months leaves 43 of the 88: 25 counter-clockwise, 14 clockwise and 4 across the middle, against 38, 40 and 10 counting every move. The guide counts every move, as ruled before the count was taken; this reading is reported, not used.
A few deliberate choices shape everything above:
The signal lives in the combination, not any one view. No single gauge, chart, or flag is "the answer." The framework intentionally spreads signal across several views and asks you to synthesize — that's what the Reading Today panel and the quadrant structure help you do. A reading is strongest when multiple views agree.
The framework is more precise at bottoms than at tops. This is intentional. The framework is built to be confident near bottoms and appropriately humble near tops, rather than pretending to precision it doesn't have at the top.
No buy/sell recommendations — historical context instead. The page never says "buy" or "sell." It says what regime you're in and what has historically followed from similar regimes, as counts and ranges. This is a deliberate stance: the honest output of a regime tool is context for your own decision, not an instruction. And the historical record is genuinely too wide (turning points span many months) to support a timing call.
Per the Falsifiability Principle, the limits are stated plainly, not buried.
Where the site reads the gauges “as first published” or “point-in-time”, each month is scored from the data as it stood about five weeks after the month began — where an archived copy of that data exists. Where none exists, that part of the month is scored from the latest figure that had been published by then, at today's revised value — never from a figure that had not yet come out. So the older record is only partly first-published in its values, though right in its timing; the recent record is first-published almost throughout.
How much, by era. Of the weight each gauge actually used, the share that was first-published — every gauge stays at or above 85% from Sep 2016:
| Years | Momentum | Forward Stress | Inflation Regime |
|---|---|---|---|
| 1964–80 | 35% | — | 0% |
| 1981–89 | 35% | 52% | 8% |
| 1990–99 | 33% | 52% | 11% |
| 2000–09 | 38% | 65% | 74% |
| 2010–14 | 84% | 76% | 80% |
| 2015–now | 100% | 90% | 97% |
The daily record reads the same, era by era (within 3 points). The coverage figure the Time Machine shows on a worked-out day is a different measure — the share of each gauge's full design weight. It counts inputs that are never revised as first-published even in years before they existed, and any other input that did not exist yet as not first-published. So for the oldest years it reads higher than this table for Forward Stress and Inflation Regime, and lower for Momentum.
Which inputs lack first prints, and until when.
What it means for the older record. A reading from the years the table shows as partly first-published is partly today's understanding of that time, not only what a reader could have seen then. Comparisons between a month's first-published reading and its revised reading are narrower there than they would be with complete first prints, because part of both is the same revised data.
Two mechanics. A settled monthly score keeps the revised values its missing inputs had when it was computed — a later revision to those inputs does not move it. The daily line's last 13 months are rebuilt every night, so their missing inputs move with new revisions until the month ages out of that window.
Two Forward Stress inputs have no archive of their first-published values: the Federal Reserve's excess bond premium, and the OECD's leading indicator for the G7 before July 2018. Both are re-estimated over their whole history as new data arrives. For those two, every month uses today's version, so rebuilding the history can move older Forward Stress scores by a few tenths of a point. Settled months on this site are not rewritten.
The broad dollar index, one of the Forward Stress inputs, is published once a week: the Federal Reserve releases each week's daily readings on the following Monday. Daily scores from September 1, 2025, and monthly scores from October 2026, use a dollar reading only nine days after its date, by which time it has been published. Earlier scores were rebuilt in September 2026 with each day's dollar reading treated as known on that day, up to eight days before it was published. Measured against the nine-day rule, some Forward Stress scores since December 2010 differ by a few tenths of a point (at most 1.4 points on a day, 0.7 on a month). Those scores are not rewritten.
Different surfaces deliberately use different data bases. The map:
Updates. The composite scores are recomputed by a scheduled job that runs twice each morning and twice each weekday evening (US Eastern), as new data is pulled. What you see is the value from the most recent recompute — the page serves a saved result, it doesn't recompute on each visit.
Because of how the scores are built, it's the current month's reading that moves, and it moves as often as its fastest-changing input. Forward Stress and Inflation Regime both include market-based inputs that update every business day (the yield curve, credit spreads, the dollar, and inflation breakevens). Their current reading therefore shifts essentially daily, with larger step-changes when the monthly and quarterly economic releases land (jobs, inflation, permits, lending standards, and so on). Momentum steps far more slowly: eight of its nine inputs are monthly releases, and the ninth — continuing jobless claims, 6.75% of the composite — is weekly. Its current reading therefore moves on a weekly-to-monthly cadence rather than a daily one.
Prior months don't move. Once a month is complete, its score is fixed; it changes only if the underlying data is later revised, or during a periodic full recomputation of history. This is what lets the historical chart be a stable record rather than a line that quietly rewrites itself.
Related features. Two companion features are documented on their own pages:
Revisions to this page. This methodology page is versioned. When a decision that affects what you see on the site changes, this page changes with it. The framework's weights and structure also undergo a formal review about once a year; any change is reflected here in the same release that ships it, so the page and the live site stay in lockstep.
The banner at the top of the page names the economy's stage — one of four, in the order a cycle runs them — with an inflation reading beside it. It is produced by fixed mechanical rules over the gauges' levels and their direction: the same numbers always produce the same stage, with no discretion involved.
Why four. The engine that stood here until September 2026 printed eleven labels. Its own record retired them: ranked by what followed, 44 of the 55 label pairs could not be told apart, and the labels that could were separated by inflation, not by growth. The four stages are what the record can defend, and each is a place on the clock face rather than a description of the weather.
The stage is read from Momentum — its level and its three-month change — and from nothing else. Forward Stress enters at one place only, inside the Contraction rule, where it was already; it does not name a stage of its own. The rules are checked in order and the first whose condition holds names the month.
The inflation overlay is not a stage. It rides beside the stage as a level and a direction: the Inflation Regime gauge's own band word, and rising / stable / easing from its six-month change. So the banner reads, for example, Expansion · inflation strained, stable. Inflation is a condition the economy is in, not a place in the cycle, and the record is clear that mixing the two is what produced eleven labels for four states.
Persistence, scaled to severity. A month's noise cannot move the stage. A Momentum reading below 30 enters Contraction immediately; the compound contraction rule, a sharp slowdown and a recovery confirm over two months; everything else over three. While a stage is forming, the page says so rather than hiding it — "Slowdown forming, 2 of 3" — and the standing stage is the one it prints.
The regime reads this feature's own gauges and nothing else. Not housing, not breadth, not the market. Each feature on this site stands on its own principles, and a synthesis of what they mean together is a separate surface with its own rules. The law is written into the regime's decision record, and the synthesis it points at is the landing page's reading.
Stress is a flag, not a stage. Forward Stress was tested for a classifying role against three outcome tiers and six candidate rules, and none met the bar that was set before the test ran. Read instead as a fall from its own twelve-month high, it carries information the growth stage does not. So it ships as a signal beside the stage, with its own register, its counts, and each occasion it fired without a recession following named in the record.
The full record is on its own page. Every period since 1981 with what followed, each stage's own table, the stress register, and the rules in the decision record's own words: the Regimes methodology →. The page that shows the record itself is Regimes.
One caveat applies in the Time Machine's worked-out view: the stage at a past date is computed from only the data available on that day. The caption says so, and the month's final reading may differ once later data arrived. That is the point of the mode, not a defect of it.
Each composite is one weighted blend of its inputs' own-history scores — score = Σ weighti × scorei, the weights summing to 100%. The weight is the input's actual weight for the month: its design weight after renormalization over the inputs that contribute (within a pillar, and across the pillars that clear the coverage floor). A carried input keeps its full weight. Nothing is smoothed: the number on the card is that sum, printed to the whole point, and its change to 0.1.
So two identities hold exactly, and the card's expander prints them. vs neutral (50) — each input's weight × (own score − 50), summing to the score minus 50. This month — each input's weight × its own-score change since last month, summing to the card's delta. The "why" line under the delta is the second column in words: the largest movers in the delta's direction, the largest against it, and everyone else as one net. Rows are ordered by the size of this month's contribution — an ordering of the arithmetic, not a ranking of what matters.
The this-month column sums to the delta by itself only when the same inputs contribute at the same weights in both months. When an input enters (its window matured, or its source resumed), leaves (the staleness gate fired), or the weights renormalize because another input did, the difference is a reweighting term. That term is the change in an input's weight × (its prior own score − 50), plus an entering input's full vs-neutral contribution, minus a leaving one's. The table names it by cause and by input — never lumped, never hidden — and in the common month it is exactly zero. The only other difference is rounding: the card prints the score to the whole point and its change to 0.1 (the Time Machine's snapshots store the score at 0.1); the sums are exact on the stored inputs.
Momentum's own scores are the DR-41 zero-anchored ranks (40 = no growth), Inflation Regime's the distance-from-2% scores (Section 2). The identity is the same algebra for all three because every composite is the one weighted blend. In the Time Machine's worked-out view both columns are computed from the point-in-time snapshot's inputs and weights — the why as it was knowable that day, as far as the archives reach. On a recorded day the table is the one the live page showed.
The "What changed" card under Reading Today reads four rows at three calendar months — the regime and the three composites — each from the table this page's own engine serves. Live, it reads today's revised data, as the page does; in the Time Machine, only the data available on the picked date. The label row shows the displayed stage of each month, never a recomputation. The change column wears the site's sign rule at each composite's trajectory threshold (5 points; 10 for Momentum), and a month is a calendar month, not 90 days.
A single place defining every technical term used above. Each entry is a term this page uses; a term the page does not use is not defined here, and the list grows when the page does.