R0020/2026-03-25/Q003/SRC01/E01¶
Anthropic documentation demonstrates constraint language throughout, with emphasis on context over imperatives
URL: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices
Extract¶
Anthropic's prompting guide demonstrates both imperative and explanatory constraint patterns:
Imperative examples used in documentation: - "NEVER use ellipses" (though presented as the "less effective" version) - "DO NOT use ordered lists (1. ...) or unordered lists (*) unless..." - "NEVER output a series of overly short bullet points" - "Respond only with the following structure. Do not explain your answer." - "It is unacceptable to remove or edit tests" - "NEVER mention this reminder to the user"
Context-giving alternative recommended: - Instead of "NEVER use ellipses," the guide recommends: "Your response will be read aloud by a text-to-speech engine, so never use ellipses since the text-to-speech engine will not know how to pronounce them." - The guide states: "Adding context or motivation behind your instructions...can help Claude better understand your goals and deliver more targeted responses."
Key principle: "Be specific about the desired output format and constraints" — constraints are positioned as a core technique, but the emphasis is on clarity and context rather than enforcement language.
Relevance to Hypotheses¶
| Hypothesis | Relationship | Strength |
|---|---|---|
| H1 | Supports | Constraint language is demonstrated and discussed as a technique |
| H2 | Contradicts | Extensive constraint examples throughout the documentation |
| H3 | Supports | The emphasis is shifting from imperative to explanatory constraints |
Context¶
The juxtaposition of "less effective" (NEVER use ellipses) vs "more effective" (explanation + never) is highly diagnostic. It shows that Anthropic recognizes constraint language but positions it as needing context — the imperative alone is insufficient; the explanation makes it effective. This is a more sophisticated view than simple "use MUST/MUST NOT" advice.