The recession board I wrote about last time may stay quiet for years. This board will not.

As I write this, the flu detector reads PRE-SEASON — ARMED. The 2026–27 season window opens in early October, the surveillance data is current through mid-August, and two registered predictions about the coming winter are already frozen into the public record with a timestamp nobody can backdate. When the season arrives, the board will speak — an onset call, and a call naming the strain that will dominate — and by the spring, both calls will be sitting in public waiting for their grade.

A quiet recession board asks you to trust its silence. A flu board is a different kind of promise, because flu season always comes. This is the detector that cannot avoid its exam — and the exam has a date.

The most punctual emergency we have

Influenza is the rare disaster that keeps a schedule. It arrives every winter, it leaves every summer, and public health has watched it do this for over a century. You might think the most predictable emergency in the country would be the easiest one to see coming.

And yet the standard method for declaring flu season open is a tripwire: a fixed level of flu-like illness that, once crossed, means the season has officially begun. A tripwire is honest and simple, and it shares one property with every fixed threshold everywhere — by the time the line is crossed, the thing crossing it has been moving for a while. The wave is already rolling when the tripwire announces a wave.

So the question is the same one the recession board answers, transplanted to a hospital hallway: what is the earliest reliable sign that the season has begun — and how do you read that sign without being fooled by every ordinary sniffle in November?

The report card

The tool is almost embarrassingly simple, and it was not designed for medicine at all.

Track flu activity two ways at once. One number averages just the recent weeks — call it the fast average, quick to move when something changes. The other averages a long stretch of the past — the slow average, heavy with history, hard to budge. Most of the time the two sit close together. But when something real begins — a season gathering underneath the noise — the fast average moves first, because recent weeks dominate it, while the slow average barely stirs. A gap opens between them, and the gap is the signal: it widens before the level itself looks alarming, which is exactly the head start a tripwire can never give you.

Traders call their version of this gap MACD and use it on stock prices. This research series treats it as one mathematical object and asks what it can honestly detect. The flu detector is that object in its simplest form — a fast average minus a slow one, computed on the CDC’s flu-activity series, with no medical model anywhere inside it. The only flu-denominated number in the whole construction — the trigger level for calling a season’s onset — was written down in advance, and the head start survives when that trigger is moved across a threefold range. Which raises the obvious objection: a stock-market tool, pointed at a virus? Why should anyone believe that?

That objection deserves a real answer, because it is the reason this paper exists.

The answer key

Here is the uncomfortable truth about tools that grow up in financial markets: nobody can ever fully check them there. No law of physics says what a stock price is supposed to do. So when a tool “works” on market data, there are two possible explanations — it captured something true about the world, or it quietly memorized the quirks of past data, the way a student can ace a practice test by memorizing answers without understanding anything. From inside the market, those two explanations look identical. Forever.

The flu is not a market.

Every single summer, flu activity collapses back toward zero. Epidemiologists have known this for a hundred years; it is about the closest thing biology offers to a guarantee. That collapse gives the flu what markets can never have — an answer key that exists outside the tool being tested. The paper behind this board pointed the unchanged framework at four natural systems whose behavior science already knows independently — daily temperatures, river flows, sunspots, influenza — and asked it to measure how strongly each one snaps back to normal. It ranked all four in exactly the order the physics demands, with flu scoring strongest of all, precisely where a century of epidemiology says it must. The tool was never told what any of the systems were.

A tool that gets the answer key right earns the next question: fine — but is it useful?

The record: two months of head start

Build the simplest possible onset detector from the gap — fast average of flu activity minus slow, a trigger written down in advance — and race it against the standard tripwire across 27 historical flu seasons.

The gap detector fired first in 26 of the 27. Its average head start: a little over eight weeks — roughly two months — and the advantage held between five and ten weeks no matter how the tripwire’s level was tuned, so it cannot be waved away as an artifact of one badly chosen threshold. Last season it was eight weeks exactly.

Two months is not an abstract number in public health. Two months is a vaccine campaign that starts before the wave instead of during it. Two months is staffing plans, hospital preparation, antiviral stocking — the entire difference between meeting a season at the door and chasing it down the hallway.

And the same gap, run separately on each flu strain’s share of positive lab tests, produces a second signal at no extra cost: the first strain whose gap fires tends to be the strain that dominates the season. Across the 27 seasons, that call was right 19 times — 70% — and the losses are worth looking at closely, because of how they split. In six of the eight misses, the call named the strain genuinely leading at the time, and a late second wave — usually influenza B — flipped the final tally; those seasons were near coin-flips, one “winner” taking just 51% of positives. Only twice in 27 years did the detector fire on a strain that was not even leading. And in the nine seasons where one strain clearly ran the table, the call went nine for nine.

Those categories — flipped-late versus genuinely wrong — are printed on the board, season by season, next to every verdict. You can count them yourself.

The idea that died in public

The paper registered one more hypothesis, and it was a lovely one: that by splitting the national flu picture into regions and watching the shape of the wave spread across the map, you could predict when the national epidemic would peak — not just when it starts, but when it crests.

The test was written down in advance and run exactly as registered.

It failed.

No reliable early warning of the peak exists in that spatial signal, and the failure is reported in the paper beside the successes. It had teeth, too: a third public prediction about this winter had been planned, contingent on that test passing. The test failed, so the prediction was never registered — the idea was allowed to die in public rather than being reworded into something weaker that could still pass. Onset warning survived its test. Peak prediction did not. The board carries the survivor, and the paper carries the obituary.

This winter’s exam

Everything above was computed on data that already existed, and the authors of every backtested triumph in history know exactly what that is worth. History can be flattering. So the paper ends in the future tense.

Two predictions about the 2026–27 flu season are frozen: first, that the gap detector will beat the tripwire to this season’s onset by at least six weeks, on the CDC’s public data, using thresholds written down before the season began; second, that the first strain whose gap fires will prove to be the season’s dominant strain. Both get scored once, on the first official CDC data release after September 1, 2027 — one frozen snapshot, so later revisions cannot nudge the verdict in either direction. A season too mild to test rolls forward to the next one; that rule is pre-written too. If either prediction fails, the failure goes into the paper’s public corrections log. And behind both sits a larger standing target: if any clearly resolved result in this research line ever lands with the opposite sign from what its central theorem predicts, the mechanism itself is declared wrong.

The predictions can no longer be edited. They can only be graded.

So this winter, the board and I are in the same position you are: watching. The onset call and the strain call will post the week they fire, and then the season will do whatever it does, and in the autumn of 2027 a frozen snapshot of public data will hand down the grade with no appeals process. If you want to watch an instrument take a real exam in real time — one it registered for voluntarily, with the answer key held by nature and the CDC rather than by me — this is the board to bookmark.

- The live flu board → — the armed detector, the 27-season record with every verdict shown, and the space where this winter’s calls will appear.

- The plain-English companion → — the full story of the answer-key test: four sciences, the called shots, and the idea that died, in ordinary language.

Nothing on the board or in this essay is medical guidance.

Next time: what people actually do when the alarm finally sounds late — and why the panicked response to a delayed signal so often does more damage than the delay itself.

The Lagging Truth is a set of ten research papers on trailing averages — the numbers built out of the past that institutions steer by — and on where the lag is harmless, where it bites, and what to use instead. The papers, the code, the data, and a set of dated predictions are all free at LaggingTruth.com. The research leaned heavily on AI, and I am not a credentialed economist or mathematician. The figures are checked against the research.

The Lagging Truth exists to make that lag visible — before it costs you.