Chunk, embed, retrieve

Sentence packed — coherent, variable size · 7 chunks, avg 50 tokens each

1. rank 1score 0.502 · dense 0.456 · keyword 0.57

   Enhanced due diligence must be refreshed every twelve months. Transaction monitoring runs continuously.

2. rank 2score 0.152 · dense 0.254 · keyword 0.00

   The report is drafted by the investigator, reviewed by the compliance officer, and filed with the financial intelligence unit. No report is filed without a named human approver.

3. rank 3score 0.126 · dense -0.076 · keyword 0.43

   Enhanced due diligence applies whenever a customer is classified as high risk. High risk classifications include politically exposed persons, customers in sanctioned jurisdictions, and customers whose source of funds cannot be evidenced.

• fixed6 chunks · 0.196

  The onboarding policy requires every corporate customer to complete identity verification before the first transfer is released. Verification covers the legal entity, its registered address, and the ultimate beneficial owners. Enhanced due

• overlap8 chunks · 0.367

  persons, customers in sanctioned jurisdictions, and customers whose source of funds cannot be evidenced. Enhanced due diligence must be refreshed every twelve months. Transaction monitoring runs continuously. Alerts are generated for tran

• sentence7 chunks · 0.502

  Enhanced due diligence must be refreshed every twelve months. Transaction monitoring runs continuously.

• paragraph6 chunks · 0.345

  Enhanced due diligence applies whenever a customer is classified as high risk. High risk classifications include politically exposed persons, customers in sanctioned jurisdictions, and customers whose source of funds cannot be evidenced. Enhanced due diligence must be refreshed every twelve months.

• pos 1
  1
• pos 2
  0.55
• pos 3
  1

Put the strongest chunk first or last. Anything buried in the middle of a long context is the least likely to be used, which is why more retrieved chunks is not the same as a better answer.