The Four Token Lanes
Four lanes, a fifty-to-one spread, the context threshold cliff, and why output is where the money goes.
Prefix Caching, Mechanically
Caching in one sentence, what is stored, where the breakpoint goes, minimum cacheable length, TTLs, and how to measure hit rate.
Invalidation: Everything That Breaks a Cache
The invalidation hierarchy, the six habits that break a cache, the silent ones nobody instruments, and a diagnostic procedure.
Recycling Context
The case for discarding context, four operations, what compaction silently costs, when to reset, and how to reset well.
Budgeting and Instrumentation
Cost is stochastic rather than linear. The minimum instrumentation, alarms worth having, runaway detection, and a defensible budget.
Compression and Token-Reduction Tools
Four different things called compression, the research base, a worked example, what the tools cost you, and how to evaluate a savings claim.
Model Routing
The routing premise, four strategies, what the numbers actually say, why agentic workloads are the hard case, and where to start.