The Watch
Analytical Intelligence

Analytical Intelligence.

Evidence-based analytic reports for questions that do not require a specialized engine. Tests claims and counterclaims, weighs evidence, and produces defensible judgments.

TYPICAL CADENCE
Runs across other engines; standalone on-request
TYPICAL SOURCE COUNT
Inherited from the engine it pressure-tests
TYPICAL OUTPUT LENGTH
Embedded; 400–1,500 words standalone
TYPICAL CONFIDENCE REGISTER
Meta-analytic; calibrates the underlying report

// QUESTION CLASS

Is this claim defensible, and under which alternative hypothesis does it break?

Analytical is the only engine that is not a report product. It is a method | Analysis of Competing Hypotheses (ACH), plus allied Structured Analytic Techniques | that runs inside every other engine's pipeline. When a Baseline report asserts a structural fact, Analytical tests the strength of the evidence behind it. When a Forecast report assigns a confidence level, Analytical interrogates the reasoning. When a Decision report identifies a critical node, Analytical stress-tests whether the claim survives the most damaging alternative interpretation. Analytical can also be invoked standalone: here is a claim, here is the evidence, tell me whether the claim survives pressure. In that mode, it produces a short report showing the hypothesis matrix, the evidence weighting, the inconsistency count per hypothesis, and the surviving hypothesis (if any). This is the engine that exists to answer the question: are you sure?

// ANATOMY

A standalone Analytical product contains:

An Analytical product looks like a worksheet, not a memo. The visual centerpiece is a matrix: hypotheses on the top row, evidence down the left column, consistency ratings in every cell. The method, developed by Richards Heuer at CIA in the 1970s and refined since, is designed specifically to counteract confirmation bias. It works by focusing on disproving alternatives rather than proving the preferred hypothesis — the hypothesis with the fewest pieces of inconsistent evidence is the one that survives, not the one with the most supporting evidence.

§ 01
Claim Under Test
The specific claim being adjudicated
§ 02
Hypotheses
The set of competing explanations (≥3, preferably 5+)
§ 03
Evidence Inventory
Every relevant piece of evidence, classified for credibility
§ 04
Diagnostic Matrix
The hypothesis-by-evidence grid, with consistency markers
§ 05
Inconsistency Analysis
Per-hypothesis count of inconsistent evidence
§ 06
Surviving Hypothesis
The one with the least inconsistent evidence
§ 07
Confidence Assessment
How much of a margin the surviving hypothesis has
§ 08
Sensitivity Check
What single piece of evidence, if wrong, would flip the result

// TRADECRAFT

ACH is designed to produce the answer you didn't want.

The method is Heuer's.

Analysis of Competing Hypotheses was developed by Richards J. Heuer, Jr., a CIA veteran, in the 1970s, and formalized in his book Psychology of Intelligence Analysis. It is the IC's canonical response to confirmation bias. The matrix structure — hypotheses across, evidence down, consistency markers at each intersection — is his specification, not our invention. We automate the evidence inventory and the cell population; the reasoning is his.

Heuer · Psychology of Intelligence Analysis (CIA, 1999)

The winning hypothesis is the least-disproven, not the most-supported.

ACH's defining move is to focus on evidence that refutes each hypothesis, not evidence that supports it. The hypothesis with the fewest inconsistent pieces of evidence is the one that survives. This is counterintuitive and it is the whole point. "Most supporting evidence" is a confirmation-bias trap; ACH closes that trap by design.

Structured Analytic Techniques (Heuer & Pherson, 2014)

Every cell cites its source.

No "holistically consistent"; no "generally supports." The consistency rating in each cell of the matrix is a specific analytic judgment about a specific piece of evidence, and it carries a source citation. A reader can disagree with a cell; they cannot disagree with the matrix as a whole without naming the cell that is wrong.

ICD 206 · Cell-level sourcing

Runs inside every other engine's pipeline. Also available as a standalone workflow.

// OTHER ENGINES

See this engine run a real question.