The Pixel Nobody Dares Remove

Chesterton's fence appears in The Thing (1929), in an essay on the drift from domesticity. A reformer finds a fence standing across a road, sees no purpose in it, and proposes to clear it away. Chesterton's response is that he may not, not yet: he must first establish why it was built, and only once he understands it may he be permitted to remove it. The passage is usually deployed against junior engineers as a caution against enthusiasm, which is a smaller use than Chesterton intended but not an unfair one.

The rule assumes something it never states, which is that the answer is recoverable. Chesterton's fence stands in a landscape with parish records and living neighbours. Investigation is expensive but finite. You can, in principle, find out.

Chesterton knew this, up to a point. Later in the same essay he allows that a reformer who has established how the fence arose and what it was for may fairly conclude that those purposes have since lapsed. The permission is conditional on recovery. Whether his actual opponents ever lacked that knowledge is worth doubting, since most of them understood the institutions they wanted altered perfectly well and simply valued them less, but the structure of the argument survives the quarrel.

Of course this is ALL about Google Tag Manager - isn't it obvious?

Open a Google Tag Manager container that has been in continuous use for six years and you will find the assumption has failed.

There is a tag called GA - conv - NEW v2 (FINAL). It is paused. There is another called GA - conv - NEW v2 (FINAL) copy, which is not paused, and which fires on a trigger referencing a URL path that returned a 404 for three years before somebody put a redirect on it for unrelated reasons. There is a script from a personalisation vendor that was acquired in 2021 and wound down the following year, still loading, still failing silently, still costing you eighty milliseconds on every page because the request has to time out before the browser gives up on it.

Nobody knows why any of this is there, and carelessness has very little to do with it. The person who built it left, and their successor left, and the agency responsible for the layer beneath that was replaced during a rebrand. The reasoning existed once, in a Slack thread on a workspace whose retention policy deleted it at ninety days, and in the head of a contractor whose email now bounces. Every record of it lived in a person, and people leave.

Management research has a name for the condition. Martin de Holan and Phillips distinguish accidental organisational forgetting from the intentional kind, observing that the accidental sort reliably destroys the knowledge a firm would have chosen to keep. A tag container is accidental forgetting with timestamps on it.

So Chesterton's rule, applied, produces an unexpected result. You must not remove the fence until you understand it. You cannot understand it. Therefore the fence is permanent. A rule intended to enforce humility before removal has become a rule of indefinite accumulation, and it has become that rule precisely because it is being followed correctly.

Whoever removes it owns it - Tag You're It

Consider the two errors available. Remove a tag that mattered, and the failure is immediate and attributable. Marketing screams into its morning lattes over a vertical cliff on the dashboard, and the git blame equivalent points directly at you.

Keep a tag that did nothing, and the cost spreads thinly across every page load and every future person who has to read the container without being able to tell which half of it is live. No single person absorbs it. It appears in no report as a line item.

One error is career-visible and the other is thermodynamic. Given that distribution, preservation is correct play, whatever it looks like from the outside. The container grows monotonically because everyone in it is behaving sensibly, which is the least satisfying sort of explanation and usually the true one.

Not touching that fucker

The industry's response to this is the container audit, generally sold as a fixed-scope engagement with a spreadsheet at the end. I have both bought and delivered these, so I am implicating myself rather than anyone else.

The spreadsheet has a column headed Recommendation. Beside anything the auditor recognises, it says remove or consolidate. Beside a great deal of the container it says something in the register of unknown purpose, retain pending review, which is the professional formulation of "I am not touching that fucker."

The auditor arrives with the same information as the client, which is none, and a worse risk position, being an outsider who would be blamed twice over. The engagement is priced at a few days. Establishing the purpose of one undocumented tag can consume most of one. So the audit reliably certifies the container rather than reducing it, the client pays for the certification, and the container is now formally blessed as necessary. The fence has been surveyed. The fence remains.

What the accumulation actually costs

The performance argument is the one usually made and it is the weaker one. The scripts do cost load time, and the Web Almanac's analysis of third parties has put average median render-blocking time for the ten most widely used providers at around 1.4 seconds, a figure that ought to cause more alarm than it does. Call it a tax and move on.

The expensive part is that the container stops being legible, and an illegible system changes which questions people are willing to ask.

Somebody wants to know which event fires when a customer opens the size guide, and whether those customers convert differently. In a container you understand, that is a lookup, five minutes, answered before the meeting ends. In a container nobody understands, it is archaeology. It requires opening every tag and correlating triggers against a dataLayer implementation that has been modified by three parties since anyone documented it, and it arrives at an answer hedged with caveats. Two days of skilled work, and skilled work at a two-day price does not get spent on curiosity,which is where most of my useful work for clients has started.

The instinct at this point is to propose a cleanup, and cleanups fail for the reason already established. You cannot clean what you cannot read.

What is missing is provenance. The fence problem exists because the reason was never attached to the artefact, and the industry's substitute for attaching it is the naming convention, which is documentation constrained to sixty characters and written in a hurry by somebody who is, statistically, about to leave.

A tag carrying its own origin (who requested it, against which decision, to measure what, and what result would have told you it was working) answers Chesterton's question at the moment of asking, rather than requiring an excavation four years later. Removal becomes cheap because understanding is cheap. The fence has a plaque on it.

Almost nothing in the martech stack works this way, and the reason is organisational. The request lives in a message or email or meeting while the implementation lives in a tag manager and the outcome lives somewhere else entirely, and these have never been formally introduced. The tag arrives in the container stripped of everything that would explain it. It is born anonymous, and it is then preserved forever on the strength of that anonymity.

Chesterton had already worked out the general shape of this by 1905, in a column called On Mending and Ending Things, where he argues that repair requires knowing what a thing was for, and that the modern habit is to patch indefinitely whatever ought to have been ended. He was writing about institutions. Your GTM container most certainly qualifies.

Ending, rather than mending

Cleanups fail and audits certify, which leaves the unpopular option. End the container. Build a new one from what the business currently needs, attach the reason to every tag as it goes in, and give the old one a date it dies on.

The objection is that you might lose something. You might. You are also paying rent on an asset nobody can read, the objection is that you might lose something. You might.

Chesterton's reformer at least had the option of going to find out. Where the option has gone, so has the prohibition, and what remains is the position he took in 1905.

Some things want ending rather than mending.