The board shows a hardcoded "4 people asked…" card. This page is the loop that makes it real: every unmet ask is logged → clustered by underlying need → and once a cluster crosses the threshold it becomes a demand card with an honest count → widen / connect / create / inform.
The asks and sensed signals below are sample data (marked); everything downstream runs for real. One asymmetry: asked (what people typed) creates demand, sensed (what the matcher inferred) can only second it — never a card on its own.
Raw phrasings, exactly as people typed them. Nothing is interpreted until detection runs.
Distilled needs, never raw sentences: when a conversation implies a gap nobody asked for out loud (someone writing in Bengali → no ESOL on the board), the matcher logs the need, not the person's words. Sensed signals never count toward the card threshold — they can only corroborate what people actually ask for.
Clusters of or more asks — sensed signals never count toward the threshold. Real provenance: expand any card to read the raw phrasings it was detected from.
Asked for, but not often enough yet to act on — detection without noise.
Needs the matcher inferred from conversation patterns — nobody has actually asked. However often they recur, they never become cards on their own: the community's voice creates demand, the model's inference only seconds it.