Skip to content

R0053/2026-03-31-02/C002 — ACH Matrix

Matrix

H1: Negative constraints necessary H2: Enforcement needed, mechanism wrong H3: AI follows all clear requirements
SRC01-E01: Negative instructions less effective -- ++ N/A
SRC01-E02: Anthropic recommends positive framing -- ++ N/A
SRC02-E01: Instruction hierarchies fail + ++ --

Legend: - ++ Strongly supports - + Supports - -- Strongly contradicts - - Contradicts - N/A Not applicable to this hypothesis

Diagnosticity Analysis

Most Diagnostic Evidence

Evidence Why Diagnostic
SRC01-E01 Directly tests the mechanism proposed in the claim — negative instructions shown to be counterproductive

Least Diagnostic Evidence

Evidence Why Non-Diagnostic
SRC02-E01 Shows instruction hierarchies fail but doesn't discriminate between positive and negative framing

Outcome

Hypothesis supported: H2 — Enforcement is needed (the problem is real) but negative constraints are not the right mechanism.

Hypotheses eliminated: H1 — Evidence shows negative instructions often backfire. H3 — Evidence shows AI does not reliably follow all clear requirements.

Hypotheses inconclusive: None.