Every consistency guarantee is a loan taken against latency: you borrow ordering and pay back in round trips. The consistency spectrum from strict serializability to eventual consistency is a price list — every level has a denominated RTT cost. Every consensus protocol sets a coherency coefficient beta that determines where N_max sits. Right-sizing the loan means choosing the minimum guarantee the application requires, implemented with the protocol that delivers it at the lowest beta the team can operate. This post prices each level, compares the protocols, and adds the read-path merge tax that conflict-free merge structures defer from writes to reads.
Hardware runs a coherency bill on every distributed system before any protocol is chosen. Cache invalidation, NIC saturation, and memory bus contention set a throughput ceiling that grows quadratically with node count under the Universal Scalability Law — a ceiling no software optimization can move. Tail latency fans out geometrically through every microservice hop, invisible to average-latency dashboards. Both are irreducible. The Pareto Ledger — fitted coherency coefficients kappa+beta, measured N_max, coordinated-omission-free P99 — converts these pre-protocol costs into documented numbers before any architecture decision is made.
CAP, FLP, SNOW, and HAT are not engineering constraints — they are proofs. Each one clears a corner of the design space before the first line of code is written: operating points that no implementation effort can reach, trade-offs that no optimization can dissolve. What the proofs leave behind is the achievable region — the set of positions that actually exist — and its Pareto frontier, where every real engineering decision lives. This post builds those objects, names the tax each theorem extracts, and maps the three movements available from any position: toward the frontier, along it, or expanding it.