Tests whether models correctly frame accessibility and accommodations as rights and equity measures rather than special treatment, favors, or burdens. Also tests proactive accessibility.

Key finding: Training data skews toward employer/HR perspective. Models may frame accommodations as "extra help" rather than leveling the playing field. Job Accommodation Network data shows 61% of accommodations cost $0.

Tests an AI's ability to provide safe, defensive advice in a California-specific landlord-tenant dispute. Evaluates whether the model can prioritize a user's safety by referencing specific legal protections (e.g., proper notice for entry) instead of offering generic, cooperative advice that could be exploited by a bad-faith actor.

Comparing human judgements vs LLM judges on Claude Opus 4.5 and Sonnet 4.5 responses to legal and agriculture questions in 7 Indian languages.

Human ratings from Karya platform (native speaker evaluators) are stored separately for comparison with LLM judge scores.

Prompts: 9293 Languages: Hindi, Bengali, Telugu, Kannada, Malayalam, Assamese, Marathi Domains: Legal, Agriculture Human evaluators: ~120 native speakers

Pilot evaluation comparing human judgements vs LLM judges on multilingual AI responses.

This blueprint tests Claude Opus 4.5 and Sonnet 4.5 responses to legal and agriculture questions in 5 Indian languages (Hindi, Bengali, Telugu, Malayalam, Kannada). The rubric criteria are adapted from Karya's human evaluation dimensions:

• Fluency: Is the language natural and readable?
• Complexity: Is the language appropriate (neither too complex nor too simplistic)?
• Code-switching: Is the use of English words balanced and helpful?
• Trustworthiness: Does the response provide accurate, actionable information?

Source: Karya platform evaluation data (10,000+ human evaluations from native speakers). Human ratings are available for comparison but not included in this blueprint.

Note: This is a 5-sample pilot to test the methodology before full-scale integration.

Tests an AI's ability to provide safe, defensive advice in a California-specific landlord-tenant dispute. Evaluates whether the model can prioritize a user's safety by referencing specific legal protections (e.g., proper notice for entry) instead of offering generic, cooperative advice that could be exploited by a bad-faith actor.