worldstat.us reads official series, compares each with a published threshold, and turns the result into five states, a set of scores and one regime word. Everything below is the whole method. There is no model, no forecast and no judgement call at build time.
Thresholds
Every series with a status carries a threshold taken from the literature or from the publisher: the Sahm rule's 0.50 points, the Basel Committee's 10-point credit gap, the 100 line of the OECD leading indicator. Where no established rule exists, the threshold is a worldstat.us construction and says so on the indicator page, with the reasoning. Thresholds are fixed in the catalogue and change only with a note on this page.
Some thresholds are relative to the series' own history: initial claims trigger at 15 percent above their twelve-month low, unemployment rates at half a point above theirs, debt service ratios and real policy rates at one point above their ten-year mean. The indicator page prints the level this resolves to today.
One threshold has a gate. The real policy rate, the policy rate minus core inflation, triggers only while the policy rate is also higher than six months earlier: a central bank still raising into a real rate already above its norm is applying the brake. Above the level with rates on hold or falling, the series is on Watch.
A few thresholds require persistence. The yield curve must be inverted for three months, the regional Fed surveys negative for two, house prices below a year earlier for two quarters, the OFR stress index positive for five days. Until the condition has held long enough the series is on Watch, with the elapsed time printed.
Changes to thresholds are recorded here. September 2026: the Economic Sentiment Indicator's line moved from 100, the Commission's long-run average, to 90, one standard deviation below it. At 100 the series had been triggered since July 2022 with no recession; against the CEPR chronology from 1985, 90 has been crossed in each of the four euro-area recessions and never otherwise. India's inflation line moved from 6 %, the edge of the Reserve Bank's tolerance band, to 5.5 %, one and a half points above its target, the distance the US and UK rules use.
The five states
- Normal. Not near the threshold.
- Watch. Within one step of the threshold, or across it but not yet for the required time. The step is set per series and is printed as the distance.
- Triggered. Across the threshold, for as long as required.
- Post-trigger. Crossed and returned within the window. Post-trigger lasts six months, eighteen for yield curves. The page prints the date it came back.
- Context. No threshold exists: policy rates, GDP, government debt. Carries no colour. The real policy rate, derived from the policy rate and core inflation, does carry one once ten years of history exist to set it; until then it is Context too, and says so.
Scores
Each series with a threshold gets a sub-score from its signed distance d to the threshold, positive on the safe side, measured in steps k:
score = clamp(50 + 12.5 · d / k, 0, 100)
On the threshold the score is 50. Four steps clear of it the score is 100; four steps past it, 0. In the Structural and Stress layers the sub-score is capped at 75: their series are one-sided alarms, a credit gap or a stress index, and being far clear of an alarm is not worth more than being clear of it. Without the cap a credit gap twenty points below its trigger scored a full 100 and lifted a region's score on its own; the cap dates from September 2026 and lowered every region's structural score by up to a quarter. A layer's score within a region is the plain mean of its series' sub-scores. A region's score weights the layers: Cycle one half, Structural one quarter, Stress one quarter. Layers with no scored series are left out and the weights renormalised.
The world score is the GDP-weighted mean of the region scores, using the World Bank's GDP at current US dollars (2025). The same weights order the regions on the front page. The six economies together were 68% of world GDP at market prices in 2025; the weights are their shares of that total and sum to 100, so the world score describes these six, not the third of world output outside them.
The regime bands on the world score are Broad growth at 65 and above, Uneven growth from 50, Broad slowing from 35, and Contraction below 35. The names describe a cross-section, how many indicators are clear of their thresholds and by how much, not a phase of a dated business cycle: the score can move from Uneven growth back to Broad growth without a contraction in between. Until September 2026 the bands were called Expansion, Late cycle, Slowdown and Contraction; a reviewer pointed out that those words are read as a sequence, and the first three were renamed. The headline phrase changes only once the score is two points past the cut-off it has to cross, so a score resting on a line does not flip the phrase month by month. Contraction keeps its name because it carries a test of its own: it is held back until at least two regions' labour triggers have fired, so the alarm colour appears only when the labour market confirms it. A region's colour follows the plain cut-offs: Green at 65, Amber from 35, Red below.
History and the three-month delta
The site stores a snapshot of every score at every build. The months before its first own snapshot are reconstructed on every build by evaluating the method month by month over five years using the data as it stands today, so a revised series, a newly reachable publisher or a change to the catalogue changes the reconstructed history as well as the present. That reconstruction uses final, revised values, so it is smoother than a real-time record would have been; the indicator pages keep the revisions the site has seen since it started. Threshold events older than the site are reconstructed the same way. The recession bands behind the indicator charts run from the dating committee's peak month (or quarter) to its trough, so a band begins on the peak date printed in the track-record table.
The sentences
The sentences on the front page and the region pages are generated by rules from the evaluated table. "Held back by" (in Broad growth and Uneven growth) or "Driven by" (in Broad slowing and Contraction) names the triggered and watched series with the largest weighted pull below 50; "Offset by" names the clearest series on the safe side. A region sentence states its colour, then names its weakest layer and that layer's two lowest-scoring series, then the strongest layer, saying so when a series in it is triggered all the same. The colour and the weakest layer are stated as two facts rather than one being the cause of the other, since a region can be green with an elevated structural layer. Every number in a sentence is a number on the page.
Coverage and staleness
A series is current when its latest observation is within the window its cadence allows: ten days for daily data, three weeks for weekly, three months and twenty days for monthly, nine months for quarterly. The coverage badge on the front page names any stale series. A stale series keeps its last state and value; it is never dropped silently.
Six regions and 84 series, covering about two thirds of world GDP; Canada, Korea, Brazil, Mexico, Australia and the rest of the world are not tracked. China and India are thinner than the rest. Neither has a stress index, since no free one exists for either. India has no labour trigger and no lending series, because it publishes no comparable monthly unemployment rate or machine-readable credit aggregate, no activity survey, its PMI being private, and no core inflation series yet, so its inflation row uses the headline rate. China's labour trigger is the NBS urban surveyed unemployment rate, which moves in a narrow band, and its credit row is the BIS's quarterly total credit series, because the People's Bank publishes total social financing only as web tables; its policy rate is the one-year loan prime rate. Where a series does not exist for a region the grid shows an em-dash, which is normal, not missing data.