bearing-performancelisted

# Performance work with GitNexus <!-- BEGIN GENERATED: graph-uncertainty — bearing regenerates this block; edits here are replaced on update --> ## The graph can be wrong A zero is not absence; a near-0.5 `r.confidence` edge is a lead, not proof (~92% of `USES`); a count can be a floor — `impact` says which in `epistemic`. Before a conclusion that matters, confirm with a scoped `Grep` (allowed here, not a gate violation) and say which check you ran. <!-- END GENERATED: graph-uncertainty --> GitNexus does **not** profile runtime — it exposes the *structure* that makes code expensive: deep call chains, high fan-in hubs, work repeated across a flow, and values recomputed instead of reused. Use it to **localize** the cost, then confirm with a profiler/benchmark. ## Workflow ``` 1. query({search_query: "<slow concept>", goal: "hot path"}) → orient on the flow 2. READ gitnexus://repo/{name}/process/<flow> → see the chain + step order 3. trace({from: "<entry>", to: "<expensive sink>"}) → exact call path (depth = cost proxy) 4. cypher (CALLS variable-length / fan-in) → deep chains + high-fan-in hubs 5. pdg_query({mode: "flows", target}) → values recomputed vs reused 6. impact({target, direction: "upstream"}) BEFORE optimizing → don't break callers 7. confirm with a profiler/benchmark, then detect_changes → verify the win + scope ``` > Stale index → `npm run bearing:agent-refresh` (autonomous). PD