Product Context Interchange Format

Lexicon Validation: Will AI agents understand our terminology?

LLMs as corpus proxies for agentic coding compatibility

Alignment by Category

100.0%

Figma

35 tests

92.9%

Storybook

28 tests

84.3%

Tokens

14 tests

67.5%

Components

77 tests

50.0%

Documentation

14 tests

33.3%

IPE

21 tests

32.1%

Knapsack

28 tests

68.1% Overall Agent Compatibility

31 terms × 7 models = 217 tests

⚠ 10 terms: AI agents may misinterpret these entities

Explore the Data

Term 1 of 31

Model	Response	Score

Model Performance

gemini-3-pro-preview

77.4%

claude-opus-4.5

76.8%

gpt-5

71.0%

gemini-3-flash-preview

71.0%

gpt-5.2

64.5%

glm-4.7

58.1%

minimax-m2.1

58.1%