Jump to text
On this page
Copy and source
Published source confirmed at last check

Source changed 2026-11-16 21:18:08 UTC · snapshot created 2026-11-16 21:19:40 UTC · last check 2026-11-17 19:15:10 UTC

CMP.Reference:1.3 - Build a whole, introduce a level, and generalize deliberately

Decomposition proposes parts whose results might help. Synthesis constructs their connection so that the required whole is obtained. Ask both directions: what does the whole require from each contribution, and what can the available contributions produce together? A set of individually correct components can lose a needed distinction at their join or rely on incompatible states, observations or progress conditions. CMP.2 develops this question for recursive answers; CMP.14 for interacting computations; CMP.15 for independently extended forms and operations.

Abstraction establishes a useful level of objects and operations. The level retains distinctions needed by its uses while hiding how those operations are obtained below it. It can introduce a compound unit and new operations rather than merely remove detail. A queue, an expression with binding and a summary of a sequence have different useful operations; none is selected just by counting stored fields. CMP.16 develops the joint construction of units, admissible operations and their whole behavior. CMP.10 can then compare stored representations for those operations. CMP.13 has a distinct purpose: a cheaper analysis that soundly approximates the behavior being investigated. An analysis may deliberately admit extra possibilities; a replacement implementation must preserve the observations its users rely on.

Generalization extends the claim to a family of cases on a stated ground. An invariant, structural argument or qualified statistical account can support such an extension; a broader name or another successful example cannot. If a chunk summary combines correctly for every admitted pair, induction can establish its use for every finite assembly. This still leaves out a new observation that the summary discarded. If an interaction argument assumes one writer, adding writers changes a premise rather than simply making the old example larger. CMP.7 develops the separate statistical question of further use of a learned rule.

Levels can be nested: an operation treated as elementary in one algorithm may be a composite algorithm at another level; an interpreter can itself be interpreted. Composition, generalization and abstraction remain different relations. An implementation is not a smaller subtype of its specification, and representing something does not make that thing a part of the representation. Keep the relation that supports the particular inference instead of using one hierarchy for all of them.