The Epistemology of Maintenance: Reconceptualizing Progress Through the Sustainability of the Existing

Dominant narratives of technological progress privilege novelty over the disciplined, knowledge intensive work that keeps existing systems reliable. This paper develops the epistemology of maintenance, the systematic study of how knowledge is generated, validated, and applied to sustain system persistence, as a framework that complements, rather than simply opposes, an innovation centred view of progress. We position the framework explicitly within the growing philosophy of technology and science and technology studies (STS) literature on maintenance and repair, and we articulate precisely what a quantitative, reliability engineering grounded treatment adds to that literature. The framework integrates reliability engineering, engineering economics, life cycle sustainability accounting, and the tacit, experiential knowledge of practitioners. We formalize these components through an exponential degradation model, a Weibull based hazard model, a net present value (NPV) maintenance interval optimization, an illustrative carbon accounting model, and a Bayesian updating mechanism, all calibrated against the American Society of Civil Engineers (ASCE) 2025 Infrastructure Report Card. A fully reproducible Python simulation, with a fixed random seed, multi parameter sensitivity analysis, two way parameter interaction analysis, and a 2,000 iteration Monte Carlo uncertainty propagation, is used to characterize the optimal preventive maintenance interval for the population of poor condition U.S. bridges, together with honest reporting of where the model is fragile, in particular a diagnosed degeneracy in the Bayesian exposure specification and a boundary solution in one region of the two way sensitivity grid. We quantify, rather than assert, the comparative advantage of an optimized maintenance policy over a reactive (run to failure) baseline: a 31.1% reduction in discounted cost and a 58.9% reduction in illustrative carbon equivalent emissions over a ten year horizon, both of which are conditional on the stated calibration and are not claims of unconditional superiority over innovation driven strategies. We close with an explicit statement of the model's scope of validity, a scientific and technical risk assessment, and a falsifiability roadmap specifying the retrospective and prospective empirical tests that could refute the framework's central claims. All numerical results in the text are traceable to the single, self contained Python script reproduced in the appendix.