Selection Scenarios
These are not hypothetical edge cases. They are structural consequences of the current architecture: AI systems making selection decisions using data that was never designed for selection.
In every scenario below, the AI failure follows the same pattern. The business has marketing copy optimized for human visitors. The AI parses that copy and attempts to infer selection eligibility. The inference is wrong — not because the AI is bad, but because marketing copy is designed to attract broadly, and selection requires narrow precision. These failures are predictable and repeatable. They happen at scale, across every industry.
Each scenario follows a consistent structure:
Intent
Find a lawyer to help sponsor a foreign employee
Constraints
- 1.Must handle H-1B visas
- 2.Licensed to practice immigration law in the US
- 3.Works with technology companies
- 4.Can handle the full petition process
Decision Logic
AI must match: problem_class = "legal-immigration-employment", capability.labels includes "h1b-visa", fit_conditions.customer_type includes "technology-company" or "employer-sponsor".
Structural Failure
The firm's website says "visa applications" — which technically includes H-1B. But the marketing copy is optimized for family immigration clients, the firm's primary audience. The AI has no structured way to distinguish employment specialization from general immigration practice, because no such data layer exists. It selects the firm based on keyword overlap. Result: suboptimal representation on an employment-based case, likely RFE, delays, and added cost.
Deeprank Resolution
A properly structured Deeprank Profile declares problem_class = "legal-immigration-employment" (not just "legal-immigration"), includes "h1b-visa" in capability.labels, and explicitly excludes "family-immigration" in non_fit.exclusions. This allows AI to select firms with actual specialization.