The light turned red. Here is exactly which two switches did it.

Where we are

The market-risk headline reads RED. It turned on the September 2 daily read, having read YELLOW on every observed weekly row since August 6. Two things had to be true at once for that to happen, and this week both were. The University of Michigan consumer survey printed 51.7 on Friday, August 28 — down 3.5 points from 55.2, the first new reading in four weeks, and below the 55 line three editions of this brief had been watching. Then, on this week's read, global central-bank liquidity was reclassified from neutral to draining. Sentiment under 55 AND liquidity draining is the published definition of RED. Both switches on; the colour follows mechanically, with no judgement from anybody.

Everything else is steadier than the headline. The business cycle stays in SUMMER — the quadrant where the economy grows and prices rise at the same time — confirmed on 15 consecutive daily reads with high confidence. It flipped from FALL on August 24 and has held every day since. Inflation is still the dominant risk under the read: core PCE and core CPI both rising, and the market's own five-year inflation forecast at 2.35%, up from 2.30% a week ago, so the one input that had gone flat last week is climbing again. All three inflation inputs point the same way.

So the honest one-line summary of the week is not 'markets got scarier'. Credit is calm, volatility is low, the dollar is still classified supportive. It is that the two specific measurements this system uses to decide the colour — what households say, and whether the big central banks are adding or removing money — both crossed their lines within five days of each other, after weeks of sitting close to them. The rest of this edition is about how much weight those two numbers can carry.

Two of two conditions

The headline colour is not a score. It is a gate with two switches, and it is published methodology rather than a judgement call: the light turns RED only when consumer sentiment is below 55 AND global central-bank liquidity is classified as draining. Either one alone leaves it YELLOW. Last week's draft of this edition said, in these words, that if liquidity were reclassified as draining while sentiment stayed where it was, the colour would go RED mechanically. That is what happened, and it happened within the week.

The order matters. Friday flipped the first switch: sentiment at 51.7 satisfies the RED condition and has since August 28. At that point the whole headline was resting on one input — the liquidity classification — which had carried the neutral label since August 10 after reading draining on every observed weekly row from June 29 through August 3. Monday's draft argued that the most recent weekly liquidity gain was a currency move rather than a real expansion, and that the classification was therefore being asked to hold the colour out of RED on a week it could not really vouch for. On the September 2 read it stopped holding.

A reclassification like this is a threshold being crossed, not a verdict. The measure behind it is the change in central-bank balance sheets over recent weeks, and the next section shows that this week's step across the line is itself partly a currency move — in the other direction from last week's. Which is exactly why the colour, on its own, is a thin thing to trade on, and why this brief spends more words on what is behind it than on the colour itself.

The warning got quieter because the news got worse

Here is the part I would rather not publish, which is the reason it gets a section rather than a footnote.

After the August 3 correction — when a routine data revision moved the survey across the 55 line and this brief was wrong a day later — I added a proximity warning. When the sentiment reading sits within two points of a threshold, the read is stamped on-boundary and every surface carries a note saying a small revision could flip the colour. For three editions it fired correctly. Last week's edition led with it: YELLOW, two-tenths from RED.

This week that warning reads stable. Not on-boundary. Stable. It switched off because the survey fell to 51.7, which is 3.3 points from the 55 line, outside the two-point margin. The number got materially worse, one of the two RED conditions went from unmet to met — and the instrument built to flag fragility responded by reporting calm. Four days later the colour turned RED, and the warning never made a sound, because the switch that turned it was liquidity, and the proximity warning has never watched liquidity.

It is not a bug in the sense of code doing something other than what it says. It measures exactly what it claims to measure: how close the survey sits to a numeric line. The defect is in what it was silently taken to mean. Distance from a threshold on one variable is not distance from a change in the headline, because the headline is a gate on two variables. Once sentiment crosses under 55, moving further under 55 does nothing at all to the colour — the colour's entire remaining sensitivity transfers to the other switch. This week is the cleanest demonstration of that I could have asked for and did not want.

I am recording it here rather than quietly fixing it and publishing a clean-looking edition, for the same reason the August 3 correction was printed in the open: an instrument being wrong in a way nobody noticed is the exact failure this brief exists to catch, and it does not stop being that when the instrument is mine. The engineering fix — watch both switches, not one — is straightforward and is queued. What makes it worth a section is the general lesson: a warning system can go quiet because the risk receded, or because the risk moved somewhere the warning does not look, and those are indistinguishable from the outside unless someone says so.

The same survey, two answers

One more thing about Friday's number that will look like a contradiction on the page, because it is two measures rather than one.

The growth side of the cycle read is still fully positive — three of three components — and one of those three components is consumer sentiment, scored improving. That sits in the same read as a headline treating the same survey as pessimistic at 51.7. Both are correct, because they measure different things. The colour reads the level: 51.7 is below 55. The growth component reads the three-month trend: 44.8 in late June, 49.5 through July, 55.2 in August, 51.7 on Friday. Down on the month, still up across the quarter.

They are one survey answering two questions, and it currently answers them differently. If the trend rolls over at the next print, the cycle quadrant loses a component while the colour is already RED — and those would not be two independent pieces of bad news either. They would be the same survey, counted twice.

Last week the dollar flattered liquidity. This week it did the opposite.

Measured across the same three central banks at both ends — the Fed, the ECB and the Bank of Japan, so a bank entering or leaving the count cannot fake a move — global liquidity is $17.62T, down $90B on the week from $17.71T. Over twelve weeks it is down 0.9%, from $17.78T. All three shrank in dollar terms: the Fed by $20B to $6.73T, the ECB by $50B to $6.86T, the Bank of Japan by $20B to $4.03T.

Last week's draft made a point that has to be applied symmetrically or it was never honest: the dollar fell 0.7% that week, and the entire weekly liquidity gain was roughly the size of that currency move applied to the European balance sheet. This week the dollar rose 0.6%, and the arithmetic runs the other way. Six-tenths of a percent of the ECB's $6.91T is about $41B; the observed fall was $50B. The same fraction of the Bank of Japan's $4.05T is about $24B; the observed fall was $20B. So in their own currencies, the ECB shrank a little and the Bank of Japan may not have shrunk at all. The Fed's $20B is in dollars and is real.

The Chinese leg proves the mechanism again from the other side: it moved from $7.47T to $7.46T on a statement that has not changed — the July figure, 64 days old. A frozen local-currency number that moves in dollars is an exchange rate and nothing else.

So the read is: the switch that turned the headline RED flipped on a week whose decisive step was partly a stronger dollar deflating foreign balance sheets, just as last week's reassuring step was partly a weaker dollar inflating them. What the currency cannot explain away is the twelve-week trend on a constant basis: down 0.9%, with the Fed shrinking steadily in its own currency throughout. The drain is real over the quarter. This particular week's crossing of the line is real plus noise, and the classification does not distinguish the two. Neither should a reader pretend it does.

The pillars stopped easing

Financial conditions still read loose, but the direction changed this week from easing to stable, and the components explain why. The extra interest riskier companies pay to borrow ticked up to 2.65% from 2.63%. For safer companies it rose to 0.81% from 0.79%. Expected stock-market volatility rose to 16.3 from 14.5. The dollar rose 0.6%, though it stays classified supportive. And the ten-year government bond yield after subtracting expected inflation — the one measure that had been rising while everything else eased — jumped to 2.45% from 2.36%, still classified restrictive, and now 0.29 points higher than twelve weeks ago. The plain ten-year yield behind it is 4.79%, up from 4.67%.

None of those are large moves, and the composite remains loose: credit is not stressed and volatility is not elevated. But for the first time in three weeks, none of the five pillars moved toward easier. Two weeks ago four of five did. The calm is intact; the easing has paused. That is a smaller change than the headline colour, and it may be the more informative one, because it is measured in market prices rather than in a survey or a classification threshold.

That leaves the tension this brief has carried for three editions, now sharper. Markets are calm — credit relaxed, volatility low, dollar supportive. Households say they feel worse, and central banks are on balance removing money. The cheerful reading is measured in prices; the gloomy ones in what people say and what institutions do. Which of them is early is not knowable from this week's data, and any edition claiming otherwise would be inventing the answer.

What to watch

• Whether the liquidity classification holds. It flipped to draining on a week whose step was partly the dollar; a week of dollar weakness could flip it back to neutral and turn the headline YELLOW again with nothing real having changed. If that happens, this brief will say so in those words, because a colour that follows the exchange rate is not a colour worth acting on.
• The next consumer survey print, which now decides two separate things: the level, already past the RED condition, and the three-month trend, still positive and one of three components holding the cycle in SUMMER. A second consecutive fall puts both on the same side.
• The ten-year real yield at 2.45%, restrictive and rising while everything else stays loose. It is the one pillar that has moved the same direction all quarter.
• The market's five-year inflation forecast, back to rising at 2.35% after one flat week. All three inflation inputs are climbing again; the crack noted last week closed.
• The Chinese statement, still the July figure at 64 days old and the oldest input under the headline. The release watch confirms nothing newer has been posted; the August figure is due mid-September. Until it lands, the global liquidity series is a three-bank story with a fourth leg that only moves on exchange rates.
• Data quality is clean: all 27 sources current, nothing stale, the collector healthy with its last refresh this morning. Ten of the thirteen weekly rows behind this edition were recorded live; the three at the old end are reconstructed from point-in-time data — each series as it was published on that date, not as later revised — following a fix shipped on August 30 and recorded in the public changelog. One defect stays open and undiagnosed: the August 17 edition described that week's conditions direction as easing, and the stored row still reads stable with the change dated August 24. Unchanged from the last two disclosures; if it resolves as an error in a published edition it will be corrected in public with the reason.

The frame, not the trade

The specific way to be wrong this week is to read RED and treat it as a verdict. It is a gate that closed because two thresholds were crossed, one by a survey that fell 3.5 points and one by a classification whose deciding week was partly a currency move. Both are real measurements. Neither is a forecast, and the second could reverse on the exchange rate alone. The right response to the colour is to look at what is behind it, which is what the sections above are for.

The larger point is about instruments, and it generalises past this dashboard. My proximity warning went quiet the week the risk was highest, because it watched one switch and the risk moved to the other. Anyone running a monitoring system of any kind — alerts on a portfolio, stop levels, a rule that says review this position if it falls below that — owns some version of that failure, because a threshold alarm answers 'is this near the line' and is routinely read as answering 'is this dangerous'. Those come apart the moment the line is crossed, and they come apart completely when the danger has a second dimension the alarm never looked at.

So the question for your own book this week is not what to do about a red light. It is: which of your rules depend on two things being true at once, and do you have a warning on both of them, or only on the one that was closest to its line last time you looked? Then ask whether any of your conclusions are resting on a single measurement that could move for reasons that have nothing to do with the thing it is supposed to measure — the way one classification here can move on the dollar. If you cannot tell a real move from a revaluation in your own numbers, your alerts cannot either.

None of this is advice and none of it is a forecast. It is a week's data, the reasoning applied to it, and an honest account of a headline that changed for two reasons, one of which is stronger than the other.

Correction — September 14

This edition dated the turn to RED to "the September 2 read". That was wrong. The daily record shows both conditions met on the September 1 read, and the headline stayed RED for six days, September 1 to 6, before returning to YELLOW on September 7. The date was wrong because the turn was checked against the weekly history, which samples one day in seven and skipped the first RED day; the daily table was not consulted for the date. Nothing else in the edition changes: the two switches, the values and the reasoning stand as published.

Corrected on September 14 with Pedro's authorization and printed here rather than edited silently, the same way the August 3 correction was. The regime line above now carries the corrected date and the marker, so link previews and the feed show it too. The underlying defect, a weekly history that can hide a whole colour change between samples, is recorded as an open instrument issue in the editions that followed.