Epistemic Provenance Non-Identifiability: How Causal Effects Outlive the Evidence of Their Origins
Knowledge can become causally embedded in the worlds it describes. Models enter markets, classifications organize populations, measurement rules structure incentives, and epistemic commitments become embodied in institutions, technologies and practices that alter subsequent reality. Existing traditions have extensively examined institutionalization, performativity, looping effects, co-production, historical epistemology and causal identification. A distinct problem arises when the causal consequences of epistemic intervention and the information required to identify their origins possess different historical persistence properties. This paper develops a theory of epistemic provenance non-identifiability. It distinguishes causal-state transmission, through which consequences persist in an inherited world, from causal-provenance transmission, through which information capable of attributing those consequences is preserved. Four regimes of provenance identifiability are introduced, ranging from strong reconstructability to endogenous non-recovery. A two-channel historical model then establishes an exact-collapse result: if query-divergent historical trajectories become statistically equivalent with respect to the inherited informational state, and subsequent endogenous evidence depends on prior history only through that state, later endogenous observation cannot recover the lost causal distinction. A causal attribution that was identifiable when an intervention occurred may therefore become non-identifiable even while its consequences continue to operate. The paper further applies established data-processing, contraction and reliability results to characterize progressive loss of historical distinguishability. No new result in information theory, probability theory or partial-identification theory is claimed. The contribution is architectural: causal effects and the query-discriminating provenance required to identify their historical authorship may be transmitted through different channels with unequal persistence. Identifiability can therefore itself become historically contingent.
Authors
- Petro Golovko (ORCID: https://orcid.org/0009-0006-3735-1973)
Institutions
- British Trust for Ornithology (GB)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-09-12
- DOI
- https://doi.org/10.5281/zenodo.22727856
- Primary Topic
- Philosophy and History of Science
- Type
- preprint