Pick the question first

TL;DR Most decision frameworks answer one of two questions: should I commit to this, or how should I think about it under uncertainty? Confusing the two is the most common failure I see. 37signals answers the first. Annie Duke answers the second. Pick the question first. The right framework follows.

A founder messages me: "I need a decision framework." I ask what the decision is. They cannot tell me. They want the framework first.

The framework is comfort. The decision is the work.

Two centres of gravity

Decision frameworks come in two centres of gravity.

• The first is a commitment framework. Its question is structural: should this decision happen at all, who should make it, and how do we bound the process of making it so the cost of being wrong stays small?

• The second is an epistemic framework. Its question is cognitive: how should I think about this under uncertainty, what would tell me I was right, and what conditions will tell me to walk?

The two share primitives at the boundary. Duke's kill criteria are an epistemic output that takes commitment-like form. 37signals' circuit breaker is a commitment device with epistemic effects. The point is not that any tool sits cleanly on one side.
The point is that the primary work of each framework is different.
A commitment framework cannot tell you the probability of a payoff. An epistemic framework cannot tell you who in the room owns the call. Apply the wrong one to the wrong question and you get the symptom every senior leader recognises: long meetings, neat decks, no decision.

37signals as commitment framework

Jason Fried and David Heinemeier Hansson, across Rework (2010) and It Doesn't Have to Be Crazy at Work (2018), and Ryan Singer in Shape Up (2019), built the cleanest commitment framework I know. Its primitives are appetite (a fixed time budget, typically six weeks), pitch (a shaped problem with an explicit no-gos list), betting table (a small council that decides which pitches run), default-to-no (raw ideas get a soft no by default because time is fixed), single decider (the council argues; one person calls it), and circuit breaker (if it does not ship in the cycle, it ends; it is not extended). 37signals' more recent catalogue, How We Make Decisions (2022), adds 38 reflective questions that work as a triage on whether a decision needs to happen at all.

The framework refuses to talk about probability. It is silent on cognitive bias. It will not help you evaluate a bet on its expected value, only whether to commit to it on its strategic fit and how to commit cleanly once you have. That is a narrower kind of "good" than Duke's. The discipline is in the narrowness. Most decision frameworks leak into everything. This one stays narrow on purpose.

Duke as epistemic framework

Annie Duke, in Thinking in Bets (2018), How to Decide (2020), and Quit (2022), built the sharpest epistemic framework in popular circulation. Her primitives are different. Every decision is a bet on a future state. Decision quality and outcome quality are independent: her term for the bias of confusing them is resulting. Probabilities are expressed as calibrated ranges, not point estimates. Kill criteria are pre-committed states plus dates for when you will walk away. The truthseeking pod is a small group bound by Mertonian CUDOS norms. Premortems, backcasts, and 10-10-10 are tools for mental time travel.

She borrows correctly. The premortem is Gary Klein's (Harvard Business Review, September 2007). The inside/outside view comes from Daniel Kahneman and Phil Tetlock. Monkey-and-pedestal is Astro Teller's, from X. Duke's original contribution is the integration of these tools into a single operating system, plus the term resulting and the operationalisation of kill criteria as states plus dates.

This framework refuses to tell you who should make the call. It assumes you are the right person and the decision belongs to you. It will help you think well about the bet, log the reasoning, and learn from the outcome. It will not help you decide whether the bet should be made in the first place by you.

The wrong framework on the wrong question

The pattern I see most often: a founder running an epistemic framework on a commitment problem. They are five meetings into mapping probabilities and payoffs for a decision that is small, reversible, and not actually theirs to make.
The framework feels rigorous. The decision goes nowhere.

The mirror image is rarer but worse.
A corporate innovation leader runs a commitment framework on an epistemic problem. They commit to a six-week cycle on a project whose central uncertainty has not been named. The team ships. The pedestal is built. The monkey was never tested.
They mistake the discipline of bounded commitment for the discipline of bounded learning. These are not the same thing.

Neither of these failures is irrational at the individual level. The founder doing epistemics on a triage question is being conscientious. The innovation leader doing commitment on an uncertainty question is shipping on time. The structural verdict is harder: good people, running real frameworks, on the wrong question.

The companions

I built two Socratic chat companions, one for each framework, because the diagnostic step keeps being the bottleneck.

• The 37signals companion (Claude Artefact) asks whether the decision needs making, who owns it, whether it can be smaller, whether it is reversible.

• The Duke companion (Claude Artefact) asks for the base rate, runs a premortem, forces a kill criterion in the form of a state and a date. They are siblings, not competitors. The choice between them is the upstream decision. The choice within them is what each framework was built to handle.

Use both, in order. Triage with 37signals: is this even a decision?
Then, if yes and the call is yours, think with Duke: what is the bet, what would tell you it was right, when will you walk?

The layer underneath

Neither framework, on its own, tells you which layer of the critical path the decision sits on.

That is its own diagnostic: the layer of the company where the bottleneck actually lives, not the layer where the symptom shows up. Both kinds of decision exist at every layer.

What changes by layer is which one is the bottleneck.

• Early, at the market and problem layers, the epistemic question (is this real?) is the limiter, even when commitment is also at stake.

• Later, at the commercial engine and scale layers, the commitment question (how do we bound it, who owns it, when do we ship?) is the limiter, even when uncertainty is also at stake.

• The wrong tool at the wrong bottleneck is the source of most stuck deliberation I see in founder and corporate innovation work.
  
  The companions help once you know the layer. The layer is a separate question, and it is the one I usually start with.

Close

If you are running a framework on a decision and the framework feels heavy, check the question first. Most of the time the framework is fine. The question is wrong. Fix the question, and the right tool announces itself.