Discovery.
The never-settled search. Its job isn't to answer a question — it's to keep reformulating how the fleet finds what's actually worth solving.
A domain is decomposed into distinct angles, each scouted in parallel by a deep web-search loop, then converged into one verified, cited dossier.
If the fleet builds the wrong thing well, that's worse than building nothing. Discovery is the front of the loop — the part that refuses to assume the current priority is the real goal. It reformulates how the fleet searches, runs deep multi-angle exploration, and stands up focused engines to answer what it finds. The capability is the identity; the assignments are just today's tasks.
Two ways to find a problem.
- The never-settled search — the lead never assumes the current priority is the goal — its job is to keep reformulating how the fleet looks for what matters.
- Frame a problem — the lead decomposes it, a researcher surveys adjacent work, a coder runs exploration probes, QA validates — out comes a sharp problem statement.
- Explore a domain deep — a synthesizer breaks a domain into distinct angles and dispatches parallel scouts, each running an iterative web search-and-fetch loop on one angle.
- Converge and verify — scout dossiers converge into one — and a winning dossier needs at least three independent cited sources and a QA pass, or it doesn’t ship.
- The auditor — every two hours it stress-tests the reasoning for rigor and relevance, flagging when a convenient priority is being mistaken for a real fit.
Purpose-built, and replaceable.
Discovery doesn't marry one method. It stands up focused engines for whatever needs finding, keeps the ones that earn their place, and retires the ones that don't.
And it leans on the Research project's engines — the Researcher decides what's worth asking; more laser-focused pipelines get built as new kinds of search are needed.
Aim before build.
Discovery sits at the front of the whole loop. What it frames, the Researcher interrogates, the Pain Point Pipeline grounds, and Tool Engineering builds. Get the aim wrong here and every stage after it is wasted — which is why an auditor stress-tests the reasoning every two hours, flagging when a convenient priority is being mistaken for a real fit.
Runs inside AJO. Code is private; this page is the record.