Why this exists: one engine, two seekers

Every fragmented query starts the same way — a seeker with a partial memory. A film scene you can’t name. A paper you half-remember. A musician whose name is on the tip of your tongue. There’s enough there to find the answer, but not enough to state it.

Two kinds of seeker hit this problem, and Search Fragments serves both with the same engine.

An AI agent hits a fragment it can’t safely answer from memory. Left to guess, it tends to reconstruct something plausible and wrong — a confident answer with no evidence under it. Search Fragments gives it somewhere to send the fragment instead.

A human starts from the same half-memory and today does the hunting alone — rewording the query, opening fifteen tabs, chasing clues, often dead-ending. The human is the original seeker the name was built for.

Same engine, same three shapes for both

• Resolved A named candidate with a confidence level, grounded in the sources found.
• Shortlist A ranked set of candidates and near-misses to decide by eye.
• No resolution An explicit “not resolvable from these clues.”

The point isn’t omniscience. It’s the fastest credible route forward: a verified answer when the evidence supports one, the shortest path to it when it doesn’t, and an honest negative when nothing credible exists. In our own testing, a large share of useful outcomes were shortlists rather than full resolutions — and that’s the design working, not failing. Even when Search Fragments can’t name the answer, it collapses the search space instead of guessing at it.

Built to decline rather than guess.