We present a systematic study of the WS-BPEL 2.0 standard based on two complementary methods: the process of constructing a new high-level WS-BPEL implementation driven by the structure of the standard, and an empirical evaluation of existing interpretations of the standard reflected in five widely available WS-BPEL-implementations, both commercial and open source.

In doing so we uncover a number of new ambiguities. Most notably, WS-BPEL's integration of XPath 1.0, the data access component of WS-BPEL, turns out to be inconsistent with the XPath standard itself, which is evidenced by substantially differing results produced by existing implementations on test cases constructed to exercise their interpretation.

The core concepts in WS-BPEL have been formalized and analyzed successfully previously. Our choice to study the standard by constructing a high-level, standard-driven implementation rather than an abstract, mathematical formalization has made it feasible to cover the complete standard, notably the integration with XPath. Given WS-BPEL's design goal of being platform-independent the inconsistencies are arguably a serious concern since they cannot be attributed to the quality of any particular implementation.