Modern pricing and risk stacks often behave as if everything changed on every tick: graphs are rebuilt, objects are rebound, and broad caches invalidated. In reality, most portfolios and market states change only marginally between ticks (and especially between days). This talk presents a practical architecture that exploits that property: snapshot barriers to guarantee “as-of” coherence, delta sets to represent change explicitly, and layered precomputations enabling a fast path for steady-state positions and a slow path for new/amended trades. We’ll show how deterministic compiled valuation plans remove runtime traversal overhead, and how minimal run/snapshot lineage makes results replayable and debuggable.