The CIA's Method: Analysis of Competing Hypotheses (ACH)
Analysis of Competing Hypotheses (ACH) is a structured way to decide which of several explanations is most likely. It was developed by Richards J. Heuer Jr. (1927–2018), a career CIA officer who spent the later part of his career on analytic methodology, in the late 1970s and 1980s. He set it out in full in Psychology of Intelligence Analysis (CIA Center for the Study of Intelligence, 1999), chapter 8. It became the foundational technique of the CIA's Structured Analytic Techniques — the family of formal methods the Agency uses to get reasoning out of an analyst's head and onto paper where it can be checked — and it is still taught there.
ACH exists to fight one failure: confirmation bias. An analyst who picks a favorite explanation and then collects supporting evidence will almost always find some, because most evidence fits more than one story. ACH reverses the work. You generate several plausible explanations up front and then try to disprove each one. The hypothesis that survives with the fewest inconsistencies is treated as most likely — not the one with the most supporting points.
The core question
Every cell of the analysis asks one thing:
"If this hypothesis were true, would I expect to see this evidence?"
Two consequences follow, and they are the heart of the method:
- Inconsistency is more powerful than consistency. A single solid piece of evidence that cannot coexist with a hypothesis damages it more than ten pieces that merely fit it. Consistent evidence rarely proves anything; inconsistent evidence can eliminate. One highly reliable item that is strongly inconsistent with a hypothesis can remove it from contention entirely.
- Diagnosticity is what matters. Evidence is only useful if it helps you tell the hypotheses apart. Heuer's own illustration is a high fever: it is fully consistent with a patient having the flu, but it is equally consistent with many other illnesses, so it does almost nothing to tell a doctor which one it is. Evidence that fits every hypothesis equally is non-diagnostic and should carry no weight in the ranking, however vivid it is.
The official form: the ACH matrix
The working artifact is a matrix:
- Hypotheses across the top, one column each.
- Evidence and arguments down the side, one row each — including assumptions and the things you would expect to see if a hypothesis were true.
- Each cell marked for how consistent or inconsistent that row is with that column. The common notations are C / I / N (consistent, inconsistent, not applicable or neutral) or + / – / NA. Many teams add CC and II for strongly consistent or strongly inconsistent, and a weight for each row's credibility (how reliable the source is) and relevance.
The ranking is read from the inconsistency count, weighted by credibility, not from the number of pluses.
Illustrative matrix
The example below is a recent open-source use of ACH discussed on X: a reservoir dam failure, tested against three explanations. It is shown here only to demonstrate the layout and the scoring logic.
| Evidence / argument | Credibility | H1: Drought stress | H2: Sabotage | H3: Neglected maintenance |
|---|---|---|---|---|
| Structure failed during a dry period | High | C | N | C |
| No claim of responsibility, no device residue found | Medium | C | I | C |
| Prior engineering warnings about the spillway on file | High | N | N | CC |
| Technical alerts logged and not acted on | High | N | I | CC |
| Failure point matches the component named in the warnings | High | I | I | CC |
| Similar reservoirs under the same drought did not fail | Medium | I | N | C |
| Inconsistencies (lower is stronger) | 2 | 3 | 0 |
The first row fits two hypotheses and cannot choose between them, so it is barely diagnostic. The last four rows do the work. Maintenance survives because the specific prior warnings and the ignored alerts contradict the other two explanations — the same conclusion the thread reached.
Heuer's eight steps
- Identify all reasonable hypotheses. Brainstorm with people who hold different perspectives. Include the obvious explanation, at least one you believe is wrong, and a null hypothesis ("nothing special is happening" or "this is exactly what it looks like"). Word the hypotheses so they are mutually exclusive: if one is true, the others are false. Overlapping hypotheses let the same evidence count for both and blur the ranking. A hypothesis that is never written down can never win, which is how the most important explanation is most often lost.
- List the significant evidence and arguments. Include assumptions, logical deductions, and the evidence you would expect to see under each hypothesis — including evidence that is conspicuously absent. Absence of something that should be present is often the most diagnostic item on the list.
- Build the matrix and score each item against every hypothesis. Work across the rows, one piece of evidence at a time against all hypotheses, rather than down one hypothesis at a time. This forces the question "does this help me tell them apart?" and exposes non-diagnostic evidence.
- Refine the matrix. Reword hypotheses that turned out to be vague or overlapping, add any that the evidence suggests, merge those that cannot be distinguished, and drop rows that are consistent with everything.
- Draw tentative conclusions by trying to refute, not confirm. Rank the hypotheses by how much evidence is inconsistent with each. The least-refuted hypothesis is the tentative leader; the goal is to reject, not to accumulate support.
- Test sensitivity. Ask which few items drive the ranking, then ask what happens if one of them is wrong, misread, or deliberately planted. If the conclusion rests on one source, and that source could be mistaken or deceptive, the conclusion is fragile and must be reported as fragile.
- Report the relative likelihood of all remaining hypotheses, not only the winner. Put the weight of the report on why each alternative was rejected — which evidence knocked it down — rather than only on why the leader was chosen. Keep an audit trail of how every cell was scored so others can check and challenge the reasoning.
- Identify milestones and indicators — specific future events or disclosures that would show things are unfolding differently than expected, and would change the ranking if they appeared.
Why it works against deception
ACH was built by an intelligence service for an environment where the other side is actively trying to mislead. Heuer wrote on cognitive factors in deception and counterdeception, and on the internal CIA dispute over the defector Yuri Nosenko, whose account of Lee Harvey Oswald divided the Agency for years ("Nosenko: Five Paths to Judgment," Studies in Intelligence, 1987). The lessons of that work are built into the method:
- Put the deception hypothesis on the matrix explicitly. If a planted story is possible, it must be a column, or it will never be tested.
- Treat evidence supplied by an interested party as a sensitivity question. Step 6 asks what happens to the ranking if a key item was fabricated or staged. A conclusion that collapses without one official exhibit is weaker than it looks.
- Distrust evidence that is consistent with everything. Deceivers rely on vivid, emotionally persuasive items that fit the story they are selling but would also fit the truth.
Common mistakes ACH is designed to catch
- Satisficing — accepting the first explanation that seems good enough and stopping.
- Counting supporting evidence instead of looking for refuting evidence.
- Anchoring on the first account announced, then treating every later item as a question of whether it can be squared with that account.
- Ignoring the absent evidence — the video that should exist, the record that should have been kept, the residue that should have been tested for.
- Letting one analyst's judgment stay invisible. The matrix exists so disagreement is about specific cells, not about whose instinct to trust.
Tools and later literature
- PARC ACH — a free software version of the matrix developed by the Palo Alto Research Center with Heuer's involvement in the mid-2000s, used inside parts of the intelligence community and by outside analysts.
- Structured Analytic Techniques for Intelligence Analysis (Heuer and Randolph H. Pherson, CQ Press, 2010; later editions) places ACH among dozens of related techniques, including key-assumptions checks and deception detection.
- CIA, A Tradecraft Primer: Structured Analytic Techniques for Improving Intelligence Analysis (2009) lists ACH among the Agency's standard diagnostic techniques.
How people use it on X today
Recent posts treat ACH as a practical, current tool rather than historical trivia, and data-analytics communities recommend it as a guard against "analytics rabbit holes" — following one attractive explanation deeper and deeper without ever testing the alternatives. Analysts apply it to open-source problems, such as the dam failure above, where the maintenance hypothesis won because specific prior warnings and ignored technical alerts contradicted drought and sabotage. Others have turned it into a prompt for checking AI output: list alternative explanations (including "the source is simply wrong"), score the evidence against each, and refuse to treat non-diagnostic facts as proof. The recurring theme is the one Heuer emphasized: most "supporting" evidence is consistent with more than one story, so the real work is hunting for what knocks hypotheses down.
It remains one of the most cited CIA analytic techniques precisely because it forces the work onto a visible matrix instead of leaving it inside one analyst's head.
How this site will apply ACH to the Charlie Kirk assassination
Each thesis page in the Hypotheses section will follow the eight steps, with the matrix printed on the page:
- The official account is one column, not the baseline. For cause of death, for example, the columns include the government's narrative of a single .30-06 round from the Losee Center rooftop, the explosive-at-the-microphone conclusion that most citizen investigators on X have reached (Microphone), and at least one further alternative plus a null hypothesis.
- Evidence rows carry a source and a credibility weight. Court filings, hearing transcripts, video, flight records, and witness accounts are each weighted for reliability, and evidence controlled by an interested party is flagged for the sensitivity step.
- Absent evidence gets its own rows — the paved courtyard, the sealed autopsy, the empty camera systems, the erased flight data (Chain 8: the scene was destroyed).
- Every page reports all hypotheses, not only the leader, with the indicators that would change the ranking.
No person is accused of a crime by any matrix on this site. Tyler Robinson is charged, not convicted, and is presumed innocent. A matrix ranks explanations of events; it does not establish anyone's guilt.