Merton's OBI caveat that erasure is "evident to knowing peers" fails for the backpropagation gap

sociology-of-sciencerobert-mertonobliteration-by-incorporationbackpropagationhistoriographycredit-assignment

Merton's qualification on obliteration by incorporation — read directly from his 1988 paper (claim-merton-1988-obi-passage-footnoted-aside-defers-to-earlier-texts) — is that apparent non-citation is not real erasure, because "the references, however tacit, are evident to knowing peers": a community that stops formally citing a source still privately knows the debt. The backpropagation-gap fails this test on the vault's own existing evidence. Andreas Griewank, a domain historian, found it necessary to write a dedicated 2012 paper, "Who Invented the Reverse Mode of Differentiation?", explicitly because the lineage from Linnainmaa (1970) to backpropagation's 1986 canonization was not evident to the field forty years later (claim-linnainmaa-priority-not-paternity). Hinton's own retrospective account of the 1986 paper — "when we first published we did not know the history so there were previous inventors that we failed to cite," as recorded in claim-rhw-1986-reference-list-four-works — describes genuine ignorance at the time of writing, not tacit-but-recognized debt withheld from the reference list.

Where Merton's OBI presupposes a community that once knew and stopped citing because citing had become superfluous, the backpropagation case is a community that structurally could not have cited what it never encountered: Werbos's and Linnainmaa's work sat in the wrong language (Finnish), the wrong venue (a numerical-analysis journal), and the wrong framing (rounding-error analysis, not learning), per the mechanism claim-linnainmaa-priority-not-paternity documents. This is a transmission failure, not an incorporation into obviousness — a second, independent line of evidence, alongside claim-backpropagation-gap-is-matthew-effect-not-obi's mechanism-type argument, that OBI is the wrong name for this gap.