The Stacey Matrix: Cynefin's Cousin That Asks About People, Not Just Cause and Effect

I wrote a while back about using Cynefin to sort a problem by whether cause and effect are knowable in advance. The Stacey Matrix, which Ralph Stacey developed in the late 1990s — earlier than Cynefin, and one of the frameworks Dave Snowden has said informed his own thinking — asks a related but genuinely different question: not "how knowable is this," but "how much do the people involved agree, and how certain are we, about what to even do." It's less about the problem's inherent structure and more about the social and epistemic state of the people trying to solve it.

Two axes, not one continuum

The first axis is certainty: how confident are we that a given action will produce the result we expect, based on precedent, cause-and-effect knowledge, or established practice? The second axis is agreement: how much do the stakeholders involved actually agree on what the goal even is, or what a good outcome looks like? Cross them and you get the same four-ish zones people intuitively recognize — Simple, close to both, where a known playbook applies cleanly; Complicated, where the technical path is uncertain enough to need expert analysis but the goal isn't in dispute; Complex, where you have to probe and adapt because neither the path nor sometimes the goal is fully settled; and Chaos, far from both, where nobody agrees on the goal and nobody has any confidence a given action will produce a predictable result.

The distinction Cynefin doesn't make explicit

Here's the thing Cynefin genuinely doesn't foreground the way Stacey does: two systems can look identically "complex" from a pure cause-and-effect standpoint and still be completely different problems to actually manage, because one has a team that agrees on what success looks like and one doesn't. Cynefin's domains are about the problem. Stacey's are about the problem and the room — the people who have to agree on what you're even trying to do before any technical approach matters.

This matters more than it sounds. I've sat in AI project reviews where the retrieval architecture was genuinely complex in the Cynefin sense — probing and experimentation were the right technical approach — but the actual blocker wasn't technical uncertainty at all. It was that product, legal, and engineering didn't agree on what "acceptable answer quality" even meant. That's a Stacey "far from agreement" problem wearing a Cynefin "complex" costume, and if you only have Cynefin in your toolkit, you'll reach for probing experiments to solve a disagreement that experiments can't resolve, because no amount of technical evidence settles a dispute about goals.

Where the two frameworks genuinely diverge

Cynefin is precise about a claim Stacey doesn't make: that the four domains reflect different underlying cause-and-effect relationships in the system itself, and that misreading which domain you're in is a category error, not just a difficulty level. Stacey's matrix is more of a spectrum in practice — Stacey himself later moved away from treating the zones as discrete categories at all, preferring to talk about the "edge of chaos" as a place organizations productively sit rather than a box to identify and exit. If you want a framework that tells you which technical process fits a problem — best-practice checklist, expert analysis, probe-sense-respond, or immediate stabilizing action — Cynefin is the sharper tool. If you want a framework that surfaces whether your real blocker is technical uncertainty or human disagreement, Stacey is the one built for that question, and it's worth running both lenses over the same problem before deciding what kind of meeting you actually need to schedule.

Using them together on an AI project

In practice I run a quick version of both. Cynefin tells me whether to trust a runbook, bring in a specialist, or start probing with small experiments. Stacey tells me something Cynefin can't: whether the room actually agrees on the target, which changes who needs to be in that room before any technical approach gets chosen. A "complex, low agreement" AI initiative — say, defining what "good" looks like for a new agentic feature nobody's built before, across teams with different incentives — needs a facilitation and alignment process before it needs an eval harness. A "complex, high agreement" one — everyone agrees what success looks like, the technical path is just genuinely uncertain — can go straight to the kind of probing, reference-architecture-backed experimentation I've written about elsewhere on this blog. Same Cynefin domain, completely different first move, and Stacey is the framework that tells you which one you're actually in.