Today b2 similar ranks candidates by semantic max-sim alone (graph distance is exclusion-only). Possible knob: boost graph-close (triadic closure) or graph-distant (serendipity/bridging) candidates.
With no automated accept-precision signal (the human is the gate), this is a dogfooding-judged experiment — add the knob only if it visibly improves the surfaced list. From planning/tasks.md Backlog.
Today
b2 similarranks candidates by semantic max-sim alone (graph distance is exclusion-only). Possible knob: boost graph-close (triadic closure) or graph-distant (serendipity/bridging) candidates.With no automated accept-precision signal (the human is the gate), this is a dogfooding-judged experiment — add the knob only if it visibly improves the surfaced list. From
planning/tasks.mdBacklog.