I built a red bar across the top of my site last month.

Full width. White text. High contrast — I checked the numbers, 7.3 to 1, comfortably past the accessibility threshold. It was, by every measure I could apply, impossible to miss.

My regular readers missed it.

Not because they’re careless. Because it was still there. It had been there on their last four visits, saying the same thing, and their brains had done what brains do with anything that never changes.

They filed it as furniture.

And the moment I understood that, I realised I’d been watching the same failure for thirty years in rooms where the stakes were considerably higher than a newsletter signup.

The version that costs money

Walk into any network operations centre and look at the alarm board.

Count the amber.

Now ask how long each one has been amber.

I have had that conversation more times than I can reconstruct, and the answer is almost never “since this morning.” It’s “that one’s been like that since we migrated.” It’s “oh, that’s just the link to the old site, ignore it.” It’s, occasionally and memorably, “I don’t actually know, it was like that when I got here.”

The board is not showing the operator what’s wrong. It’s showing them what has been wrong long enough to become normal.

Every one of those permanent ambers is doing something worse than nothing. It’s raising the floor. It’s teaching the room that colour doesn’t mean urgency, that the board is decorative, that judgement happens somewhere other than the screen everyone paid for.

Post-incident reviews in process industries have been saying this for decades: when something goes genuinely wrong, the alarm that mattered arrived in a flood of alarms that didn’t, and the operator had already learned to discount all of them.

The failure isn’t inattention.

The failure is that we asked people to ignore things, they learned to, and then we were surprised.

Why permanence kills signal

Here’s the part that took me years to state properly.

Information lives in change, not in presence.

A warning that is always on carries no information. It’s a constant. You could delete it and the operator’s picture of the world would be identical, because they’d already deleted it themselves — just internally, where you can’t see them do it.

This is not a discipline problem. You cannot train it away. Attention is a filter that adapts to its environment, and the environment you built taught it that red means nothing.

Which produces the ugly conclusion:

Your most urgent channel is your most fragile one.

The louder the mechanism, the faster it burns out under permanence. A red banner is more susceptible to this than a plain sentence, because red is a promise. Break the promise a few times and the reader stops reading the colour, and now you’ve spent something you can’t easily buy back.

I spent it on my own site in three weeks.

The bit that connects to everything else I write

Every permanent alert is a decision somebody didn’t make.

That’s the whole thing.

Nobody decided the link would stay degraded. Somebody deferred fixing it, then deferred deciding whether it mattered, and the amber became the record of that deferral — a small monument to a choice postponed.

Multiply by the forty of them on the board, and the alarm system stops being a monitoring tool and becomes an archive of everything the organisation has quietly agreed not to resolve.

That’s the complexity tax in its most visible form. Not an abstraction. A screen full of it, in colour, that everyone has stopped reading.

And it arrives with a second bill: the next real incident costs more, because the mechanism that should have caught it early has been taught to be ignored.

Three rules I now apply to anything that shouts

Every permanent signal gets an expiry date.

Not “we’ll review it.” A date. When the date arrives, either the condition is fixed or the alert is deliberately downgraded to something quieter — a log line, a report entry, a dashboard the right person actually opens. What is not allowed is for it to simply continue.

My red bar now has a calendar entry. Six weeks, then it changes or it goes.

Count what you’re asking people to ignore.

This is the single most useful audit I know, and almost nobody runs it. Not “how many alerts do we have” — how many are we asking a human being to look past every day?

If the number is above about five, you don’t have a monitoring system. You have a training programme in not looking.

Test by removal.

If you switch it off and nobody notices for a week, it was never doing its job. That’s not an argument for leaving it on. That’s the evidence you needed to take it out and put the attention somewhere it will be spent.

The part I got wrong

I want to be honest about what my instinct was when I saw the banner wasn’t working.

Make it brighter.

That’s the reflex — the alert isn’t landing, so increase the volume. Bigger type. Stronger red. Maybe animate it.

It’s the same reflex that adds a fifth severity level to an alarm system that already has four nobody respects. It escalates the format instead of fixing the reason, and it always buys a few days before the new level gets absorbed into the furniture too.

The answer was never more contrast.

The answer was fewer permanent things, each of which changes when something actually changes.

Go and look at your board

Whatever your board is.

The alarm console. The risk register nobody has reordered since the audit. The compliance dashboard that has shown the same two ambers since the NIS2 assessment. The banner on your own website.

Count the things that have been in the same state for more than a month.

That number is not a measure of your monitoring.

It’s a measure of how much your organisation has agreed to stop seeing.

And the thing about what you’ve stopped seeing is that it’s still there, still costing, still perfectly visible to whoever eventually writes the incident report.