Answers are grounded in this paper and cite their sources.
IS research evaluates digital architectures by two criteria that pull in different directions: integration-era research valued operational coherence, while research on layered modular architectures and platforms values changeability and in some statements assumes that contracts that decouple change also decouple operation. Research on fit infers misfit from outcomes and user evaluations. The modularity and fit strands both overlook that components that change independently often still depend, transaction by transaction, on agreement conditions that no contract carries: shared codes, identities, states, cut-offs and rules. We call these conditions operational correspondence. Where they fail, people and tools at particular positions in the flow repair the deficit, using what we call positioned compensation capacity. A positioned compensation bound shows that, with repair done as well as possible, the deficit that reaches outcomes equals the largest excess of correspondence deficits over the capacity positioned to reach them. Three consequences follow. The change frequency at which separation pays rises with throughput, and the drift of separately governed components raises it. While capacity covers the deficits, misfit does not register in operational outcomes and appears instead as compensation labor. The labor above a feasible contract baseline is structural waste, the interest an uncarried correspondence charges on digital debt. And because which deficits bind depends on where capacity sits, automating repair roles can move the binding segment to positions that outcomes did not reveal. We derive four propositions and five corollaries, state what would refute them, and outline a field design. This is version 2.6, the first public version of the paper. It rebuilds an earlier formulation of the argument as an information-systems theory paper; the earlier empirical probes are reported in full in Supplement C. The files are the paper (PDF, and its Markdown source with references.bib), Supplement C (PDF and Markdown), and the replication package, whose seeded scripts reproduce every number in Appendix A. The stored data of the earlier probes are not included; they are available from the authors on request. The analysis plans of those probes are at https://osf.io/urh3g. This work was carried out with AI assistance: Claude (Anthropic) and ChatGPT (OpenAI). The paper's Acknowledgements state how each was used.
No takes yet. Share an insight, caveat, or question.
Azarang et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: