Quality measures are extremely difficult to quantify because they depend on many parameters and factors, some of which cannot be identified or measured readily. Thus, a general quality measure that encompasses a wide variety of systems might not be able to provide the information leading to quality increase, as quality control graphs need a numerical assessment of quality. This paper concentrates on a subset of this problem that is of particular interest in the current web application domain. In this domain, Java is the language of choice for interoperable code segments that constitute an effective interface layer between servers (e.g., database or transaction servers) and the user. Realizing those code segments, however, is a challenge. Reusability criteria do not apply as the interface code is highly volatile and subject to changes very often. On the other hand, in order to obtain code that enables concept transfer and that promotes efficiency and reliability, one must maintain a high quality standard. This paper describes a quality model that can be used directly on code, and thus during light development and in rapid development cycles. The model is based on non-quantifiable attributes of quality that then are related to specific measures found using a structured method (Goal Question Metric). The measures identify statistical clusters that can be used to categorize the quality of each Java class file. The relation between quality factors and measures is proven at the mathematical level, using the representational theory of measurement as a guideline, and then at the empirical level, using an independent assessment. The preliminary results collected seem to indicate that the quality model is effective in classifying Java programs. An important indication can then be obtained by the quality analysis, as it is possible to use cluster distances to give advice to programmers on how to improve their programs.
