By documenting the company's own systems and the specialist's role in them: internal training records, authorship of procedures, escalation histories, and a statement of how many employees hold comparable knowledge and how long training a replacement would take.
Evidence beats adjectives
Petitions fail on this point when they describe the employee as outstanding rather than showing what the employee knows that others do not. Useful material is usually already inside the business. Version histories show who wrote a calibration procedure. Service tickets show which faults were escalated and to whom. Training curricula and completion records show what internal instruction exists and who has completed it. A short statement from the engineering lead can say how many staff can commission the system unaided and how many months of supervised work a competent new hire would need. Where the company licenses some underlying technology from a third party, the file should separate what is licensed from what the company built on top of it, because the knowledge claimed has to be the organization's own rather than the licensor's.
Hypothetical example: A Canadian water-treatment membrane producer wants to transfer the person who developed its internal failure-analysis method. The best file starts with the qualifying relationship and one full year of foreign payroll in the preceding three years, then connects the person's work to records the business already uses: test protocols, restricted training modules, design revisions, and incident reports. A comparison should explain why a normally qualified U.S. hire could not immediately perform the same function. Generic praise letters fail because they do not identify the organization's knowledge or its operational value. The company should redact confidential technical details only after preserving enough context to show what is distinctive. It should also count all prior L status, because specialized-knowledge admission has a five-year aggregate limit.