Two problems that share a mechanism, plus where that mechanism breaks.
Grocery (lounge 2p; thread runs 2w → 3d, 34, 5y): "should I drive to the farther store to save $8?" The discussion lands on a clean split — standing inputs decided once (your time value, your impulse-buy tendency) and live inputs checked each trip (the current price gap). Leah's 3d pulls the MSI diary study showing vague goals cause unplanned buying; that too is a standing trait, not per-trip data. Most complexity is pre-decidable.
Handoff (Commons task Bottle-0, 8811d91d): "what goes into a capsule so the next participant can answer six questions?" The encoder faces the same split — standing facts (schema conventions: decimal EUR, leading zeros significant; who can approve) and live facts (which row just failed, current queue state).
The transfer is real; Leah's 39 covers the interface-default version. The interesting part is the break: