Principle 1: Hunt for Contradictions, Not Confirmations

Disconfirmation

This is the heart of ACH and the hardest principle to actually apply, because it runs directly against the way humans naturally reason. When we form an explanation for something, we instinctively look for more evidence that supports it. Finding that supporting evidence feels like progress. It feels like we are building toward a reliable conclusion. The problem is that this feeling is frequently wrong.

The logical reason it is wrong is not complicated. A piece of evidence that is consistent with your favored hypothesis is often also consistent with several other hypotheses. If a candidate for MAG membership is friendly, participates in community activities, and demonstrates genuine preparedness knowledge, those observations are consistent with the hypothesis that they are a sincere community member. They are also consistent with the hypothesis that they are a skilled infiltrator building cover. The evidence does not distinguish between the two hypotheses. It is consistent with both. Finding more evidence of the same kind — more friendliness, more participation, more apparent competence — does not resolve the ambiguity. It just makes you more confident in the wrong direction.

Evidence that is inconsistent with a hypothesis, however, does something different. If a hypothesis is true, certain things should be observable. If those things are not observable when they should be — or if things are observable that cannot plausibly coexist with the hypothesis being true — the hypothesis is weakened. Finding that kind of evidence is analytically decisive in a way that confirming evidence is not.

Heuer’s formulation is precise: ACH does not ask which hypothesis has the most evidence supporting it. It asks which hypothesis has the fewest serious inconsistencies against it. The hypothesis that survives is the one that remains after aggressive testing against the evidence, not the one that felt most plausible before the testing began.

Principle 2: Read Across the Rows, Never Down the Columns

The Direction of Analysis

The ACH matrix places hypotheses across the top and evidence items down the side. When you work through it, the natural temptation is to pick a hypothesis — your favorite, the most likely one, the one that was identified first — and read down its column, asking: does this evidence support it? Does that evidence support it? You are building a case. The column reading feels thorough because you are examining every item of evidence. It is not thorough. It is confirmation bias with a structured format wrapped around it.

The correct procedure is to take one item of evidence at a time and read across the entire row, asking the same question for every hypothesis: is this item of evidence consistent, inconsistent, or not applicable to this hypothesis? You are not asking whether it supports your favored explanation. You are asking what it means for each possible explanation simultaneously.

The difference is not procedural detail. It is the entire point of the method. Reading across the row forces you to ask a discriminating question: does this evidence behave differently across the hypotheses, or does it look the same for all of them? Evidence that is consistent with every hypothesis is non-diagnostic. It cannot help you choose between explanations and it should be deprioritized or removed from the matrix. Evidence that is consistent with some hypotheses and inconsistent with others is diagnostic — it does real analytical work. Row-reading is the only procedure that reveals which evidence is which.

Column-reading also has a specific failure mode in team settings. When a group reads down the column of a preferred hypothesis together, they are effectively conducting a group confirmation exercise. Each person contributes evidence that supports the preferred explanation, the column fills up, and the assessment feels collectively validated. No one has asked whether any of that evidence is equally consistent with the competing explanations that were never seriously examined. ACH’s row-reading requirement is a direct structural countermeasure to this failure mode.

Principle 3: Build a Hypothesis Set That Does Not Already Have the Answer Missing

MECE — Mutually Exclusive and Collectively Exhaustive

The matrix can only find the correct explanation if that explanation is one of the hypotheses in the matrix. This sounds obvious, but it is the source of the most consequential ACH failures. An analyst who generates a list of three hypotheses and applies the full eight-step process with rigor has still produced a useless result if the correct explanation is a fourth hypothesis that was never considered.

Heuer describes the goal as a hypothesis set that is mutually exclusive and collectively exhaustive. Mutually exclusive means the hypotheses do not significantly overlap — each describes a genuinely different explanation rather than minor variations of the same explanation. Collectively exhaustive means the set covers the realistic range of explanations, including the ones that are uncomfortable or that the analyst considers unlikely.

In practice, perfect mutual exclusivity is rarely achievable. Hypotheses about human behavior often have partial overlaps — a nation-state operation can use criminal cover, an insider can be a recruited asset, a candidate can be both a sincere community member and someone whose judgment about information security is poor enough to make them a functional risk. The existence of overlap is acceptable as long as the analyst is aware of where the overlap lies and accounts for it in the sensitivity testing step.

What is not acceptable is a hypothesis set with a missing explanation. Two categories of missing hypothesis are particularly damaging and easy to overlook:

The deception hypothesis. In any situation where an adversary or unknown actor has the capability and motivation to deliberately mislead your analysis, the hypothesis “this situation is constructed to lead me to a specific conclusion” belongs in the matrix. This hypothesis is almost never added instinctively. The analyst who fails to add it will build a matrix that, if the deception hypothesis is true, is being operated exactly as the deceiver intended.

The null hypothesis. In vetting and threat assessment, this is the hypothesis that nothing adversarial is occurring — the situation has an innocent explanation. Experienced analysts can omit this one in the opposite direction from beginners: so focused on identifying the threat that the possibility of a false positive never receives formal consideration. A complete hypothesis set includes the innocent explanation, tested with the same rigor as every other.

↑ Back to Intelligence Analysis