Searching for Meaning

Walls of string

There is a wall in most conspiracy films. Photographs, newspaper clippings, index cards, and red string connecting them all. The wall is always accurate. The people in the photographs really did meet, the payments really were made, the dates really do line up, and the string is true.

This piece was inspired by listening to a Guardian podcast on a conspiracy about missing scientists. On it, the investigative journalist Phil Tinline makes a distinction that explains why the wall proves nothing. The role of the investigative journalist, he says, is not to find the connections; it is to find whether they mean anything. The conspiracist and the journalist do the same first job: gathering connections. The professional difference between them is the second job: testing whether the connections carry weight and being willing to conclude that they don't. The journalist who finds nothing and says so has done journalism.

I have spent forty years drawing walls of string, but have I been a journalist or a conspiracy theorist?

We call them dependency maps, capability models, and current state architectures. We walk the seams of the organisation, find the connections, and the connections are true. This system really does feed that one; this process really does depend on that team, and everything touches everything. The diagram proves it, and therefore, by implication, the diagram matters. An architect who maps connections without judging them is doing organisational conspiracism, and the politeness with which our deliverables are received is not respect.

It is what you give someone who has shown you the string but not the story.

Other people's second jobs

Every discipline whose work is finding things has had to build a second practice for weighing them. It is worth looking at two of them because they suggest that what architecture is missing is not talent, but vocabulary and permission.

Radiology finds so much that it needed a word for real findings that mean nothing. An incidentaloma is a mass discovered on a scan that is there, visible, but clinically insignificant. The word exists because chasing every finding harms the patient, through cascades of further scans, biopsies and interventions, each carrying its own risk. So, the discipline built protocols for leaving things alone. Noticing that some findings meant nothing was not the hard part, but the achievement was giving the judgement a name and a protocol.

Every dependency map I have ever produced was full of incidentalomas. Architecture has no word for them, nor protocol or permission to leave them alone.

Intelligence analysis contributes to this as well. An analyst who writes "we assess with low confidence" has done something professionally legitimate. The confidence level is part of the finding, not an apology for it. An architect who says "this part of the model is deliberately vague" has made a finding against themselves. Same epistemic honesty but opposite career consequences.

We already weigh. We just do it off the books.

Here is where I should be careful, because the obvious conclusion is wrong. Architects do weigh connections. It is arguably the thing good architects most obviously do. The judgement happens in the conversation after the review board, in the corridor advice, and in the choice of which risks to escalate and which to leave. The desire is not missing, and neither, mostly, is the action.

What is missing is somewhere for the judgement to live. The model has slots for connections but no slots for their weights. There is no deliverable that records "this seam is load bearing, with moderate confidence" or "we looked at this one and it is inert, spend your attention elsewhere". So, the weighing lives in the architect's head and maybe gets transmitted in conversation. The weighing practice of architecture is itself a piece of fabric: informal, doing the real work and invisible to the formal structure.

There is a reason it stays informal, and the system made it happen. Consider the two ways an architect can be wrong. Write "this connection does not matter", and if that seam ever fails there is a document with your name on it, sitting next to the incident report. Produce a model that turns out to be incomplete, as every model does, and the wrongness is diffuse: no single failure, blame shared with everyone who reviewed and approved it. Both are wrong but only one of them can be blamed in a meeting. Governance cultures punish the named judgement far more reliably than they reward it, so the rational architect presents the connections and lets the reader infer the weight.

It is built by ordinary corporate machinery working exactly as designed. Incident reviews that hunt for a name before a mechanism. Objectives that count what can be counted: models delivered, reviews passed, domains covered. Risk logs where entries are added and not retired, because adding a risk is diligence and retiring one is exposure. And maybe cynically but poetically, nobody has ever been promoted for concluding that something means nothing (feel free to send over your examples if I am wrong).

The system does not want the judgement. It wants the artefact, because artefacts can be counted and judgements can be quoted. Look at what review boards test: coverage, conformance, consistency of views. Nobody at the board asks which three of these forty connections matter, and which we can all stop thinking about.

The weighing is not rejected. It is never requested, so it never becomes work.

The missing third act

For a while, I thought that was the whole argument: find, then weigh, and we only do the first in public. Then I tried to write down what a weighed connection leads to and reached for a vocabulary that was not there. Defer, build, stop, or something else. I did not know enough to say which, and not for lack of judgement.

In every discipline that weighs, the weighing exists to produce a disposition, a decision about what happens next and when. The journalist's weighing ends in publishing, continuing to dig, or dropping it. Medicine, again, is furthest ahead. Clinical practice has a formal disposition vocabulary: treat now, active surveillance, do not treat. The middle one is the revelation. Active surveillance is a formal, respectable, protocolised decision to do nothing yet, with a review date attached. It is neither neglect nor deferral by omission. The patient knows, the record knows, and there is a date on which the judgement gets re-examined.

Architecture maybe has one disposition: build. Everything on the map is either in scope now or it silently is not, and the silent category has no name, no owner, and no clock. "Deferred" in most organisations means "fell off the roadmap without a decision being recorded". There are few examples of an architectural equivalent of active surveillance, a stated decision that this seam matters, needs nothing yet, and will be looked at again in March.

This is also, I think, what the recurring complaint about architects means. The complaint is that architects just draw a picture and move on, they never get their hands dirty. The proposed cure, delivered with great confidence in a thousand LinkedIn posts and more than a few restructures, is that architects should write code.

Run that cure through any other discipline that finds and weighs. A radiologist who hands you the scan, points at the shadows, and leaves the room is not practising medicine, but the remedy is not to send radiologists into surgery. An air accident investigator who never reaches a finding is failing, but the remedy is not to have them fly the next plane. An auditor who signs nothing is useless, but the remedy is not to have them keep the books. In every discipline whose job is weighing, we understand that judgement is the dirty work, and that pushing the judge into fabrication does not add accountability but removes the one person positioned to weigh the whole.

The complaint is right, and the cure is wrong. Architects who never get their hands dirty are failing, but the dirt was not in the code. The architect's hands get dirty at the moment of disposition, when they put their name to a set of consequential, falsifiable decisions: build this now, watch that until the third quarter, stop this, and these six connections mean nothing.

A vocabulary we should have been building for thirty years

So, the chain has three parts. Find the connections. Weigh them. Dispose of them, in the surgical sense of disposition rather than the bin. Architecture industrialised the first, runs the second in private, and, what is probably the challenge here, it never built the third.

The start of a disposition set is not complicated. Act. Watch, with a date. Stop. Ignore, with the reasons recorded. Four words that would change what an architecture deliverable is, because each one is a judgement someone signed, not a shape someone drew.

I want to be clear about the state of this, because pretending the four categories are settled would be wrong. They are the start of a vocabulary. Radiology took decades to develop its protocols for leaving things alone, and it had the advantage of bodies that are broadly alike. Organisations are not. What counts as load-bearing in one will be adjacency in another, and the confidence we can attach to any judgement will be lower than we would like and should be stated anyway.

But the test of whether a connection means anything is available today and does not require a framework. A connection means something if severing it, or letting it drift, would change an outcome someone would lose sleep over. Otherwise, it is adjacency, and the most valuable sentence an architect can offer, the one no deliverable currently has room for, might be this one: we looked, and it means nothing, look elsewhere.

And we need to find a way for that to show value, so we can be promoted for concluding that something means nothing.