Accuracy over cleverness in register search

Question

What should register search optimize for when the cost of a false positive is professional risk — not engagement?

Research

Consumer search teaches users that approximate matches are helpful. Legal register lookup teaches the opposite: operators need to know whether a mark exists under a normalized form, who owns it, and which class — fast enough for a phone call. Fuzzy suggestions without clear “exact / not found” semantics create false confidence.

Observations

• Latency budget is socialIf search feels slow, the desk abandons it mid-conversation.
• Normalization is product workSpaces, case, and punctuation variants matter more than exotic ranking.
• Empty is honestTeaching “no match” as a valid outcome preserves trust.

Hypothesis

Exact-first FTS with aggressive normalization will outperform fuzzy-first UX for IP desks, even if demos look less “AI.”

Experiment

Shipped in NepalIPMS search; tradeoffs recorded in Decisions and engineering notes. Try the illustrative demo in the playground.

Conclusion

Register search is a trust surface. Prefer precision, teach emptiness, measure latency against real calls — not against consumer search benchmarks.