Minutes, not meetings
Use for a local, low-consequence issue with direct evidence and little dispute.
- One person can complete it
- A reversible test is available
- The propagation path is short
LEARN · A 12-MINUTE INTRODUCTION
Diagnosis before prescription.
EDF is a way to slow down one specific mistake: acting on the first plausible explanation. It gives a team a shared language for moving from what happened to what can responsibly be done next.
LESSON 01
THE MENTAL MODEL
Three distinctions change the quality of a diagnosis.
NOT THE SAME
Outcome ≠ explanation
“Customers abandoned checkout” is an outcome. “The new payment library caused customers to abandon checkout” is an explanation. Keeping them separate prevents a guess from entering the record disguised as a fact.
NOT THE SAME
Origin ≠ control point
A cause may begin somewhere you cannot change. The best control point may be later in the path: a review gate, an alarm, a handoff, or the authority to stop work. EDF asks where future outcomes can actually be influenced.
NOT THE SAME
Confidence ≠ certainty
Confidence is an explicit, evidence-relative judgment. In the current EDF draft, labels such as Low, Medium, and High are qualitative, not statistically calibrated probabilities. “Medium confidence, sufficient for a reversible test” is a claim about whether the evidence is adequate for that action, not certainty that the diagnosis is true.
The habit to practice
Whenever someone says “the problem is…,” ask: Is that the observed outcome, an interpretation, an origin, or a proposed control?
LESSON 02
THE SEVEN MOVES
A diagnosis is a chain of accountable reasoning.
- 01
Declare the system context
Name the primary system, the larger context that materially shapes it, and the aspect you are examining. This is an honesty statement about where your claim is valid.
- 02
State the outcome and manifestations
Describe the result and its observable signals. Remove blame, causes, adjectives like “broken,” and embedded solutions.
- 03
Build an origin network
Generate more than one plausible contributor. Include technical, human, organizational, environmental, informational, and economic origins when relevant.
- 04
Trace propagation
Connect contributors to the outcome. A list says what existed; a path explains how the system converted conditions into effects.
- 05
Evaluate evidence and confidence
Pair each important claim with support, contradiction, source quality, and remaining unknowns. Then state confidence without inflating it.
- 06
Identify and rank control points
Compare candidate interventions by influence, controllability, cost, reversibility, timing, and confidence.
- 07
Judge diagnostic sufficiency
Decide whether the model is good enough for the next responsible action. The answer may be act, investigate, or reframe.
LESSON 03
CHOOSE THE SMALLEST SUFFICIENT LEVEL
Rigor should match the situation.
Make alternatives visible
Use when several causes are plausible or the consequences justify explicit evidence.
- Record competing explanations
- Preserve material unknowns
- Explain why one control ranks first
Model distributed control
Use when multiple systems, institutions, or disciplines interact.
- Map several propagation paths
- Separate stakeholder perspectives
- Calibrate across analysts when useful
Start with EDF–0. Expand only when uncertainty, disagreement, consequence, or system interaction demands it.
LESSON 04
DIAGNOSTIC SUFFICIENCY
Stop at responsible action—not perfect knowledge.
A diagnosis is sufficient when its confidence, unknowns, and proposed action are appropriately matched. A reversible experiment can tolerate more uncertainty than an irreversible, high-consequence decision.
The next move is bounded and testable.
Proceed, state what you expect to change, and return to observe the result.
A material unknown could change the action.
Gather the smallest piece of evidence that best separates competing explanations.
The current system context cannot explain the outcome.
Change the focus or widen the context instead of forcing the evidence into the model.