Search & verification

Two engines, one rule: the text is never guessed. Search returns verbatim lines with their Ang or nothing; verification returns a verdict with the canonical line or NOT_FOUND. Both fold Roman input with the same function the database was built with, and both are pinned by golden vectors that every deploy and the iOS app replay.

• Search waterfallThe cascade as do_search codes it (Poster 05 · a simulator over the real API)
• Modes and tiersEach mode you can ask for, each tier’s entry condition, what it queries, what it reports (The same simulator, focused on modes)
• The Roman foldThe five steps of roman_norm and why it lives in three repositories (Poster 06 · the function read from the source)
• Verification engineThe verdict ladder, the thresholds, the Ang modifier (Poster 07 · a playground over /api/verify)
• Harnesses and golden vectorsWhat roundtrip, casual-quote and chaos measure; what make contract pins (The contract files and their counts · today’s numbers)
• Variants and the seeker lexiconHow Roman spellings reach the right word: the variant index built in sggs-data and the hand-kept lexicon (How each is built, used and changed)
• Debugging a search resultA playbook: reproduce, read the tier, find the cause, fix it in the right place, prove it (The debugging loop)

Search is the search bounded context (/api/search, /api/word); verification is the verify context (/api/verify). Both are read-only over the immutable database — see the API routes.