Compare two routes to the same useful finish.

An uncached computation and a retrieved value can be compared meaningfully only when the intended work is clear.

If one route includes validation and the other stops at retrieval, their costs describe different finishes. If one serves a stable request population and the other is observed during unusual change, the conditions also differ.

A baseline makes those differences visible before the comparison becomes a claim.

Write the finish in ordinary language.

“Produce a response” is often too broad. What must the response enable? Which correctness, freshness, permission, and delivery conditions matter to the caller?

Then define the population: the requests, period, operating conditions, and relevant exceptions included in the account.

The baseline is a chosen reference, not a universal truth about the system. State why it suits the question.

Keep both accounts honest.

Fresh work can involve source acquisition, computation, checking, retries, and delivery. Reuse can involve retrieval, validation, adaptation, misses, renewal, and maintenance.

Some work belongs to the system; some may be pushed onto a person or another team. The comparison should make that transfer visible when it affects the intended result.

See the total-cost account.

Choose a comparison that can change your action.

You might compare reuse for stable versus changing inputs; broad versus narrow equivalence conditions; or one approach to validation against another.

Each comparison needs a useful purpose. An elaborate measurement that cannot inform any design or operating choice may consume more attention than it returns.

The result can favor different approaches for different request classes. A single system-wide average should not erase a useful distinction.

Record the remaining uncertainty.

An omitted consequential cost is not zero. A short observation window does not establish every long-term condition. An illustrative model does not become a measured result because it has precise arithmetic.

Use the worksheet to make those limits explicit. Read the critical path before treating a faster component as a faster finish.

A good baseline gives disagreement somewhere concrete to begin.