Format Interoperability

Definition (confidence: high)

How envelope format choice interacts with instruction following, token efficiency, human editability, and machine parseability. The format selection reflects a tool’s priority between LLM compliance and human readability.

Trade-offs by Format (confidence: high)

FormatToken EfficiencyLLM ComplianceHuman EditabilityMachine Parseability
XML EnvelopeMediumHighMediumHigh
Markdown HeadersHighHighHighLow
YAML ConfigHighLowMediumHigh
JSON-EmbeddedMediumMediumLowHigh
Hybrid XML/MarkdownLowMediumLowHigh
JSON-OnlyHighLowLowHigh
Template VariablesVery HighN/AHighN/A

Key Findings (confidence: high)

  • XML envelopes estimated 15-20% better LLM compliance than markdown, likely because LLMs are fine-tuned on web markup (HTML/XML) and tag names serve as semantic anchors
  • Markdown dominates for developer-facing tools where humans will read or edit prompts
  • Template variables offer the highest token efficiency (minimal skeleton + runtime substitution)
  • JSON and YAML experiments have not gained traction — they lack narrative flow that LLMs respond to

xml-envelope-format — The dominant format (~35%) with detailed structure patterns and variants.

markdown-header-format — The developer-favored format (~29%) with flat and deeply nested variants.

xml-vs-markdown-envelopes — Direct comparison between the two dominant formats across all evaluation dimensions.

template-variable-prompts — The edge case: minimal skeleton prompts with very high token efficiency.

Implications (confidence: high)

The choice of envelope format is not arbitrary — it reflects the tool’s priority between LLM compliance and human readability. XML envelopes dominate because they optimize for the LLM’s parsing ability. Markdown dominates for developer-facing tools where humans interact with the prompt. The hybrid and pure-data formats (YAML, JSON) remain experimental. Controlled empirical validation of format performance is a high-value research opportunity.

Sources

  • [raw/prompts/articles/taxonomy-synthesis-2026-07-16.md]