There is a particular kind of disaster that, in hindsight, nobody can quite believe was allowed to happen. The warning signs were visible. Someone, somewhere in the building, had run the numbers and gone pale. And yet the machine rolled on until it couldn’t. When we reach for explanations we usually reach for people: someone was greedy, someone was reckless, someone looked away. That story is satisfying because it gives us a villain. It is also, most of the time, wrong about the mechanism.

The more useful explanation is structural, and the 2011 film Margin Call is the cleanest dramatisation of it I know. Treat the film here strictly as a parable — it makes no claim about any real institution, and nothing in it should be read as a description of the actual events of 2008. What it gives us is a clean model of a single problem.

The gap, dramatised

In the film, a junior analyst finishes a piece of work his departed boss never completed. The maths is not complicated once you see it: the firm is holding positions that, under conditions now arriving, could exceed the value of the entire company. He understands exactly what he is looking at. He has almost no power to act on it.

So the information climbs. At each level it is re-explained, compressed, and translated for an audience that understands a little less of the detail and holds a little more of the authority. By the time it reaches the top of the building, the person with the power to decide openly asks to have it explained as if to a child. He is not stupid. He is simply standing where the decision rights are — which is precisely not where the comprehension is.

That is the comprehension gap: in a steep hierarchy, understanding concentrates at the bottom and authority concentrates at the top, and the two are connected only by a long, lossy chain of translation. The gap is not an accident of these particular characters. It is the default geometry of large organisations.

Why it is structural, not moral

It is tempting to moralise this. Margin Call deliberately refuses to. There is no cartoon villain in the film, and that refusal is the whole point. Each person behaves rationally given their position, their incentives, and the partial picture available to them. The harm is produced by the arrangement, not by appetite.

This matters because the moral reading leads to the wrong remedy. If the problem is bad people, the fix is to find better people — more ethical, more careful, more honest. But replacing the individuals does not change the geometry. The next set of people will sit in the same seats, receive the same compressed signals, and face the same incentive to be first out the door. The gap regenerates.

Consider how information actually degrades on the way up. Each layer must summarise, because the layer above has less time and less context. Summarising means deciding what to drop. The things most likely to be dropped are the caveats, the conditionals, the “this holds unless” clauses — exactly the parts that carry the risk. By the top of the chain, a hedged, uncertain, alarming analysis has often become a confident headline. The signal didn’t fail to travel. It travelled, and was sanded smooth on the way.

The standard non-fix

The instinctive organisational response to a near-miss is to put smarter people at the top, or to add a reporting layer whose job is to “surface” risk. Both responses leave the geometry intact. A smarter executive still cannot hold the granular detail of every position; that is not a failing, it is a physical limit on attention. And an extra reporting layer is one more translation step — one more place for the caveats to fall out.

The deeper trouble is that the standard fix treats comprehension as something that should flow upward to meet authority. But comprehension is heavy and detailed and local. It does not compress well. Authority, by contrast, is light: it is permission, not knowledge. The cheaper move is to send the light thing down to meet the heavy thing.

The structural cure

This is the idea at the centre of L. David Marquet’s Turn the Ship Around!, an account of his command of the nuclear submarine USS Santa Fe. Marquet’s framing is deliberately plain: move authority to where the information is, rather than moving information to where the authority is. He calls it leader-leader rather than leader-follower, and the operational signature of it is a change in language — people stop asking permission (“Request permission to…”) and start declaring intent (“I intend to…”).

The shift sounds small and is not. “I intend to” forces the person closest to the information to think one full step further than a request does. A request offloads the judgement upward; a declared intention keeps the judgement local and merely invites a veto. Over time the vetoes mostly stop coming, because the person at the edge is reliably right — they were always the one who understood the situation.

Translate that back to the comprehension gap. The junior analyst in the parable did not need a better way to explain the risk upward. He needed the authority to act on what he already understood, bounded by clear constraints. The cure is not better communication across the gap. It is moving the decision to the side of the gap where the understanding already lives.

A caveat the engine requires me to keep: Marquet’s is a single case, a specific submarine with a specific crew under specific constraints. It is a powerful illustration and a usable principle, not a proof that the same move works identically everywhere. Treat it as a design pattern to adapt, not a formula to copy.

The cost, and the limits

Here is the honest complication, and the reason “just decentralise” is not a finished answer. Moving authority to the edge only narrows the comprehension gap if two things are also true at the edge: competence and shared context.

Competence, because authority handed to someone who does not in fact understand the situation does not close the gap — it relocates it and removes the brake. The submarine worked because Marquet paired the transfer of authority with a relentless investment in technical competence and clarity about the goal. Strip those out and “give control” produces confident, fast, local error, which is worse than slow central error because nothing checks it.

Shared context, because a decision made at the edge still has to serve the whole. The person declaring “I intend to” needs to understand not just their own instrument but what the organisation is for. Without that, you get a thousand locally rational decisions that add up to incoherence — which is its own kind of institutional failure.

So the cure has a price. It demands that you spend, continuously, on competence and on making the organisation’s purpose legible at every level. That spending is the actual work. The slogan is free; the conditions that make the slogan true are expensive. Most organisations want the decentralisation without paying for the competence, and then conclude that decentralisation doesn’t work.

What to do with this

The practical residue is short. First, look for the comprehension gap by asking a single diagnostic question: for our highest-stakes decisions, how many translation steps sit between the person who understands the situation and the person who is allowed to act on it? If the answer is more than one or two, the gap is doing its quiet work.

Second, resist the reflex to close it by improving communication upward. Improving the translation chain makes the chain longer-lived, not shorter. Ask instead what authority could be moved down to meet the understanding, and what competence and context you would have to guarantee at the edge to make that safe.

The disasters that everyone half-saw coming are rarely failures of seeing. They are failures of arrangement — the understanding and the authority were never in the same place. You cannot legislate that gap away with better people. You can only design it down.