Capacity sensitivity: expose the assumptions behind a crossover

Define demand D in reference resource-hours that deliver a specified useful workload, rather than assume every device-hour has equal throughput. Different hardware, model precision, memory capacity, and scheduling policies can change performance and output quality. Include the required service latency, recovery behavior, and available capacity. An option that cannot complete the work within the horizon is infeasible even if its algebraic cost is lower. A pricing comparison without this normalization can mistake slower execution or reduced service for an efficiency gain.

List the commitments and avoidable terms at the same horizon. Setup, maintenance, operator effort, data movement, and unused capacity can sit outside an attractive variable rate. A fixed fee can remain payable when demand falls; a flexible pool may have admission limits when demand rises. Do not infer guaranteed capacity from a price concession, or a universal right to cancel from a usage-based label. Our fictional models deliberately strip those complexities down to documented parameters so the sensitivity is visible rather than hidden in an asserted market recommendation.

Equate the two costs to find a candidate demand threshold. If variable rates are equal, there is no isolated crossover unless fixed terms also match; then the simplified costs coincide throughout their shared range. A negative solution lies outside nonnegative demand. Even a positive solution can be irrelevant if it exceeds an option’s deliverable capacity. Check the costs on both sides to identify the ordering rather than memorize that a fixed-cost arrangement is always preferable beyond some unspecified volume.

Sensitivity makes the conditional nature concrete. A change in the variable rate matters in proportion to demand, while changing the fixed term shifts the whole curve. Demand itself is uncertain and can vary with season, quality changes, and failed work. Report a range of plausible workload inputs and demonstrate where the ordering reverses. Multi-resource constraints from capacity-allocation research further limit the feasible domain: spare arithmetic throughput does not help a workload that cannot fit its memory or network needs.

Store the workload trace, rate assumptions, unit conversions, excluded costs, and model version with the analysis. Separate measured facts from scenario inputs; changing a fictional price does not become evidence about a real supplier. Evaluate whether the ranking survives reasonable demand and failure assumptions before presenting a stable conclusion. If forecast demand straddles a threshold, report that uncertainty directly. An operational plan can retain headroom or delay an irreversible commitment, but the lesson’s arithmetic alone does not authorize any particular business decision.

The function lab illustrates changing sensitivity in a nonlinear toy. Reset, explicitly set a=1 and x=0.5, predict, then Play or Step. Repeat from a fresh Reset with a=1 and x=1 explicitly set; y values are 0.25 and one, and slopes are one and two. This quadratic is not either linear capacity-cost model, so it cannot solve their crossover or forecast demand. It demonstrates why a derivative belongs to a specified model and operating point, rather than being a portable constant that can be attached to every procurement comparison.