Design a service that computes a location- and time-varying price multiplier from real-time supply/demand imbalance across geographic cells, for an on-demand marketplace, and serves that multiplier to the quote path.

Operating context. Demand signals (open requests) and supply signals (available providers) flow in continuously and are geotagged. Every few seconds the engine derives a multiplier per cell reflecting how starved that cell is. The quote/estimate path reads the current multiplier for a coordinate and must be fast; once a user is shown a price and accepts, that multiplier is locked for the quote's validity window even if the cell's multiplier moves. Rapid oscillation between neighboring cells and across time is undesirable.

Out of scope. The pricing policy and fairness rules that decide the exact multiplier formula, the payment and quote-charging system, the internals of any demand-forecasting model, and provider incentive payouts. Assume a formula and a payments system exist and consume your outputs.

What to produce. A high-level architecture covering: ingestion of demand and supply signals per cell; the spatial cell scheme and its resolution trade-off; the windowed computation that turns supply/demand into a multiplier on a few-second cadence, with smoothing/hysteresis to damp oscillation; publishing multipliers to a low-latency read store and the coordinate→cell lookup on the quote path; quote locking so an accepted price is stable; and degradation when the compute path stalls. Sketch the components and the flow from signal to a locked quote; we will probe specifics in checkpoints.