Book

Description

Software quality is usually reported as a handful of headline numbers. This book shows how easily those numbers can point in the wrong direction. Drawing on 8,607 runtime measurements and 1,453 static observations gathered from LegalOS, an AI-enabled Online Dispute Resolution platform built for Indian civil jurisdictions, the authors run eight experiments spanning six of the eight ISO/IEC 25010 product quality characteristics: API performance, authorization correctness, workflow validation, input robustness, behaviour under concurrency, fault handling, static maintainability, and deployment cost. Two of the findings concern measurement itself. A throughput figure computed the way common load-testing harnesses compute it turns out to embed the harness's own overhead, which here fell from 75 percent of the measurement window at low concurrency to under 2 percent at high concurrency, producing a performance peak that does not exist anywhere in the underlying data. Separately, a maintainability score of 84.80 proves to be inflated by modules that contain no executable code at all. A third finding ties the two together. Every module in the codebase earns the analyser's highest maintainability grade and 97 percent of its functions are low-complexity, yet only 24.5 percent of statements are covered by the shipped test suite, and the single defect uncovered during testing sits in a module with no coverage whatsoever. Written for software engineers, quality practitioners, and researchers, this is a study of what measurement reveals, what it quietly conceals, and why structural metrics and behavioural verification should never be read as the same thing.