Reading Culture from Meetings

Overview

Structure is on a chart. Culture is not, and it decides what the structure actually permits.

This topic has to earn its place against the suspicion that it is soft-skills material, so it is organised around two hard questions: how do you classify a culture from evidence rather than impression, and which of your architectural practices does each type break?

Three types

Westrum’s typology, from research on organisational safety, classifies cultures by how they handle information:

Pathological — power-oriented. Information is a resource to be hoarded. Messengers carrying bad news are punished. Novelty is crushed because it threatens the people invested in the present arrangement.

Bureaucratic — rule-oriented. Information travels through channels. Messengers are ignored rather than punished. Novelty creates problems because it does not fit the process.

Generative — performance-oriented. Information is actively sought. Messengers are trained and rewarded. Novelty is implemented because the goal is the outcome rather than the procedure.

Most organisations are not uniformly one type. A company can be generative about technical decisions and pathological about commercial ones, and the mix often varies by which executive you are standing near.

Six observables

The rubric for this module demands evidence rather than vibes, so here is the instrument. Each question has three characteristic answers.

What to observe	Pathological	Bureaucratic	Generative
Someone brings bad news to a senior person	Blamed, or quietly sidelined	Politely noted; nothing happens	Thanked; something changes
Who speaks when a senior person is present	Only seniors	By role and turn	Whoever knows the answer
How a past failure is discussed	Not discussed, or someone is named	In terms of which process was not followed	In terms of causes, openly
An idea from a junior engineer	Dies unless adopted by someone senior	Needs a sponsor and a document	Gets tried
Getting information from another team	Requires permission	Requires a formal request	You just ask
Is there something everyone knows and nobody says	Yes, and naming it is dangerous	Yes, and naming it is awkward	Rarely

Two notes on using this.

Observe occasions, not personalities. The unit of evidence is “in the planning meeting on the third, X happened” — not “people here seem defensive”. Specific occasions are checkable and survive disagreement; impressions do not, and an analysis built on them will be dismissed the first time someone objects.

The last row is the most diagnostic single question. An organisation with a widely-known unmentionable has told you most of what you need to know, and the identity of the unmentionable tells you the rest.

What each type does to your practices

This is the part that makes culture an architectural concern.

graph TD
  A["<b>Pathological</b><br/>bad news is punished"] --> B["Nobody will carry a risk<br/>report. Build mechanisms<br/>that surface problems<br/><i>without a messenger</i>"]
  C["<b>Bureaucratic</b><br/>process is the currency"] --> D["Artifacts carry decisions.<br/>Use the process rather than<br/>routing around it"]
  E["<b>Generative</b><br/>outcomes are the currency"] --> F["You can defer, experiment<br/>and reverse — which is what<br/>staged bets assume"]

  classDef path fill:#D9695B,stroke:#B4483A,stroke-width:2px,color:#ffffff
  classDef bur fill:#E8A33D,stroke:#C4832A,stroke-width:2px,color:#ffffff
  classDef gen fill:#7D9A4E,stroke:#5F7A38,stroke-width:2px,color:#ffffff
  class A,B path
  class C,D bur
  class E,F gen

Three cultures, three different sets of practices that work. The same architectural proposal needs different machinery around it in each.

In a pathological culture, you cannot rely on people raising concerns. Risks stay hidden until they are incidents, because carrying bad news is personally costly. The architectural response is to build mechanisms that surface problems without requiring a messenger: an automatic regression gate that blocks a release on its own authority, a cost alert that fires at a threshold, a rung-attribution metric in a monthly report nobody has to volunteer. Module 6’s automatic gate is worth considerably more here than in a healthy culture, and that is a design decision driven by culture rather than by engineering.

In a bureaucratic culture, artifacts are the currency. Decisions travel in documents and are legitimised by having followed the process. Routing around the process fails even when you are right. The response is to use it deliberately — which makes Module 9’s artifact set a political instrument rather than administrative overhead, and makes a well-formed proposal the fastest path rather than the slowest.

In a generative culture, you can defer, experiment and reverse. Which is exactly what Module 3’s strategy-as-bets assumes. That assumption is worth checking: a staged plan with explicit reversal conditions reads as rigorous in a generative culture and as indecisive in the other two, where committing to a plan is how credibility is earned.

The cost of misreading it

Specific and expensive: you propose the right practice for the wrong culture, it fails, and the failure is attributed to you rather than to the mismatch.

A staged bet with reversal triggers presented to a bureaucratic organisation reads as not having decided. An automatic gate proposed in a generative one reads as distrust of colleagues. The content was right both times.

One fact, read carefully

The most diagnostic fact available is a single sentence from the case: in 2023 a machine-learning feature degraded over months and was switched off, quietly, without an announcement. Nobody brings it up in meetings. Everybody remembers.

Run that against the six observables.

How is a past failure discussed? It is not. There was no postmortem, no announcement, no published account. Is there something everyone knows and nobody says? Yes, explicitly — and that is the last row of the table, the most diagnostic one.

What it is not. There is no evidence anyone was punished, so this is not clearly pathological in the strong sense. Nobody appears to have been blamed; the feature simply disappeared.

What it is. A failure that cannot be discussed is at minimum strongly bureaucratic on the information axis, with a pathological edge: the organisation protected itself from the conversation rather than from the recurrence. A generative organisation would have produced a published postmortem, and the absence of one is the evidence.

Three consequences for the architecture, which is the point of the exercise:

1. You cannot rely on anyone raising a repeat. If the new platform starts degrading, the person who notices will weigh the cost of saying so against the memory of 2023. So Module 6’s automatic regression gate is not a nice-to-have — it is the mechanism that substitutes for a messenger, and that is now its primary justification.

2. Ivan Petrov’s resistance is rational and informational. He has seen a system degrade quietly and watched the organisation decline to discuss it. His scepticism is not conservatism; it is the only available form of risk reporting in a culture that does not reward the explicit kind.

3. Your proposal needs an artifact, not a conversation. In a culture where the failure was never written down, a written record that this decision was made for these reasons, with this reversal condition, is both unusual and valuable. Module 9’s ADR practice lands here as a cultural intervention.

What to write in the analysis: the classification, the specific evidence (the unexamined 2023 failure, the absence of a postmortem, the unmentionable), and the three architectural consequences. That is evidence-based, not vibes — which is exactly what the rubric asks for.

Feeds the workbook

This is the second half of your organisational analysis, and the rubric requires the culture classification to cite observable evidence. Use the six-row table, name occasions, and finish with the practices you are relying on that this culture would break.

Check your understanding

1. Run the six observables against an organisation you know. Which row was most revealing?
2. Name a practice from Modules 3–7 that assumes a generative culture. What would you substitute in a bureaucratic one?
3. There is something everyone knows and nobody says where you work. What is it, and what does it tell you about how to propose change?

---

→ Structure and culture are a snapshot. One force changes both, and it is why you were hired. — What changes with size