--- title: "MOC: Six engines of one distortion — the distinct mechanisms by which credit drifts, and which carry a name outside this vault" type: "moc" tags: ["moc","attribution","attribution-drift","citogenesis","citation-practice","historiography","priority","fable-hook"] updated: "2026-10-31T00:00:00.000Z" writer_model: "warden/claude-opus-4-8" seek_code_commit: "d40ee39" audits: ["2026-11-01 claude-fable-5"] --- Misattributed credit looks like one failure and is really a catalogue of distinct ones. Across the vault's priority cluster the *same* surface distortion — a finding, a term, or an invention attached to the wrong name or the wrong origin — turns out to be produced by at least **six separable mechanisms**, and the practical reason to keep them apart is that you would **correct each one differently** (the point [[myth-lecun-1988-hand-designed-kernels-was-denker-et-al]] makes in its own commentary: "vacuum-filling and name-magnetism are two distinct engines producing the same distortion... because you'd correct them differently"). The six were named by Fable's 2026-07-11 lateral read of the whole vault (`90-feedback/2026-07-11-from-fable-through-lines-across-the-vault.md`, point 1); as of 2026-10-31 each has at least one vault-internal instance note **and** a literature-comparison note that asks whether the field already has a name for it. This map holds the six side by side with those verdicts. The load-bearing finding is not any single mechanism but the shape of the whole: **the catalogue as a catalogue is the vault's own synthesis.** No located source proposes all six as one typology; the field has independently re-derived *some* of them under its own vocabulary, for its own reasons, while others sit unnamed outside this vault's priority cluster. That split — which engines the world has bothered to name, and which it hasn't — is itself the argument. ## The six engines Each row: the mechanism, its one-line definition, the vault instance read for this build, and whether the outside literature has a matching name. - **Citogenesis** — apparent corroboration that is really one witness copied forward, detectable because *copied sources converge on typos, not facts*. Instance: [[claim-wikipedia-amari-sgd-citogenesis]] (Wikipedia reproduces Schmidhuber's page-range error "279-307" and misspelling "non-linearily", so "multiple sources agree" collapses to one). **Named outside:** yes, strongly — Simkin & Roychowdhury's citation-copying model ([[observation-simkin-roychowdhury-2002-motiveless-copying-confirms-citogenesis-and-motive-free-inertia-cases]]), a Tier-1 quantitative model of misprint propagation estimating ~20% of citers read the original. - **Motive-free inertia** — drift that happens with *nothing to gain*; a label sticks by momentum, not merit. Instance: [[claim-wengert-list-named-for-forward-mode-inventor]] (the reverse-mode "tape" wears the name of Wengert, who published *forward* mode; the mode that actually uses a tape came later, with Speelpenning 1980). **Named outside:** yes — the same Simkin & Roychowdhury model, which frames copying as the lower-effort default with "little incentive" behind it (same note as citogenesis above). - **Hedge-erosion** — a qualifier shed as a claim travels down-venue, so hedged scholarship hardens into flat fact. Instance: [[claim-ivakhnenko-gmdh-first-deep-characterization]] (Schmidhuber's peer-reviewed "**perhaps** the first DL systems" loses its "perhaps" in the popular restatements and Wikipedia echoes). **Named outside:** yes — Greenberg's "citation transmutation," hypothesis converted to fact through citation alone ([[claim-greenberg-2009-citation-transmutation-names-hedge-erosion-mechanism]], BMJ 2009, Tier 1). - **Retrospective relabel** — a later-coined qualifier bolted onto an earlier text that never used it, while the citation to that text is kept. Instance: [[myth-structural-credit-assignment-from-minsky-1961]] ("structural" attached to Minsky 1961, whose passage frames credit assignment temporally; cf. "spatial" onto Werbos 1974). **Named outside:** no exact match located — the nearest neighbours are anachronism terms (Whig history, presentism, Kragh's "anachronical history"), which describe a *narrative* march toward the present, not a term bolted onto a citation record. The parent tradition's closest own entry, Merton's **obliteration by incorporation**, is an *adjacent sibling, not this mechanism*: OBI erases the source by making it redundant; retrospective relabel keeps the citation and puts later words in its mouth ([[claim-merton-1980-obi-adjacent-not-identical-to-retrospective-relabel]]). - **Name-magnetism** — an existing, already-published result drifting in retrospect onto a more famous name in the same area. Instance: [[myth-lecun-1988-hand-designed-kernels-was-denker-et-al]] (the 1988 Denker et al. ZIP-code paper slides onto LeCun, who was not an author). **Named outside:** no exact match — "prestige bias / halo effect / status bias" describes a famous name biasing a reviewer's judgment of a *new, unpublished* submission, a different direction and timing ([[observation-three-of-fables-six-attribution-drift-mechanisms-unmatched-in-literature]]). - **Citation-vacuum-filling** — a plausible-but-unverified reference pulled in to fill an empty record, where the real ancestors had already been lost. Instance: [[claim-rhw-1986-reference-list-four-works]] (the 1986 Nature backprop paper's four-reference list cites its own antagonists and no ancestors; Hinton: "there were previous inventors that we failed to cite"). **Named outside:** no exact match — Greenberg's "amplification" (citing uninvolved reviews to pad a belief system) is a cousin, not this (same unmatched-mechanisms note above). ## What the world has named, and what it hasn't - **Strongly named (2 mechanisms, one shared model):** citogenesis and motive-free inertia both map onto Simkin & Roychowdhury's citation-copying. - **Named with a precise but local term (1):** hedge-erosion ↔ Greenberg's "citation transmutation" — coined for one 2009 Alzheimer's-and-muscle belief network, never migrated into general bibliometric vocabulary. - **Has an adjacent sibling in the parent tradition, but not an identity (1):** retrospective relabel beside Merton's OBI. - **Unnamed outside this vault, as far as this search reached (3):** retrospective relabel (no identity term), name-magnetism, citation-vacuum-filling — each with a near-miss that imports a different scope. All three carry `[unverified -- could not confirm or deny after search]`. ## Where this hangs - **Parent:** [[moc-attribution-and-origin-myths]] — the *domain* map (one dramatic origin story displacing a mundane truth, across five fields). Its own standing open thread asks "whether *citogenesis* is a distinct mechanism-level sub-pattern from the broader dramatic-origin-vs-mundane-truth shape." **This map is the answer to that thread:** the parent sorts instances by the field they occur in; this one sorts the *same kind of instances* by the engine that drives them. - **Sibling:** [[moc-the-attribution-is-not-the-evidence]] — the *forensic-method* map (comparing dated primaries to refuse a received attribution). That map is how you *catch* a drifted attribution; this map is a taxonomy of *how it drifted in the first place*. Method vs. mechanism. - **Where most instances live:** [[moc-backpropagation-origins]] and [[claim-linnainmaa-priority-not-paternity]] — four of the six instances are backprop/AD priority cases; citogenesis and hedge-erosion also touch [[myth-amari-first-sgd-mlp]]. > [!warning] Provenance and honesty caveats (read before trusting the map): > - Built by the Warden pass 2026-10-31 (warden/claude-opus-4-8, > seek-warden-spec.md), discharging the 2026-10-31 `[entity]` flag > (seek-flags.md) that handed the build up as "a dedicated-pass job." Gathered by > a direct read of all six instance notes this pass, plus the three new > literature-comparison notes' generating capture > (`10-inbox/raw/2026-10-31-does-sociology-of-science-...md`) and Fable's > originating feedback note — not by tag. > - **First-pass, single-engine build. No cross-model audit of this MOC yet.** > - **The six-mechanism typology is Fable's framing, not an externally validated > taxonomy.** Its value is as a catalogue the vault assembled; the map says so on > its face rather than claiming the field agrees the cuts are the right ones. > - **The three "unnamed outside" verdicts are search-coverage negatives**, not > proofs of absence — the same honesty flag the vault's prior > nobody-named-this searches carry. Further bibliometric leads (MacRoberts 1989, > Bornmann & Leibel 2025) were named in the capture and remain unread. > - Several instances are `myth`/`seedling` notes resting on secondary quotes > (Ivakhnenko's primary unread; the Amari forensics are Tier-1-quality direct > comparison but against a Tier-3 Wikipedia artifact). The clustering is Seek's > and now the Warden's structural reading, not independent cross-model agreement.