---
title: "The cosine-0.84 pairing between Alemohammad 2026's dilution mechanism and Simkin & Roychowdhury 2005's copying mechanism is an analytical contrast: opposed generative mechanisms for the same concentration curve, one shared higher-order point"
type: "observation"
status: "seedling"
bridge_verdict: "analytical-contrast"
writer_model: "claude-sonnet-5"
source_url: "https://arxiv.org/pdf/2608.19230"
source_title: "When AI Writes, Who Gets Cited? Evidence of Citation Monoculture Across Language Models"
source_author: "Sina Alemohammad, Denghui Zhang, Bolong Tang, Anthony Qin, Gengchen Mai, Ahmed Abbasi, Richard Baraniuk, Zhangyang Wang"
source_date: "2026-08-03"
source_venue: "arXiv preprint 2608.19230 [cs.DL]"
source_tier: 1
source_sha: "fa6cfef20674c458d602e81f7ff17913371cc3f6adc5af141a9a306668d28689"
source_quote: "Being cited never raises a paper's chance of being shown again"
source_quote_2: "what recursion supplies is dilution"
source_url_3: "https://arxiv.org/pdf/physics/0504094"
source_title_3: "A mathematical theory of citing"
source_author_3: "M.V. Simkin, V.P. Roychowdhury"
source_date_3: "2005 (arXiv physics/0504094; refereed as JASIST 58(11):1661-1673, 2007)"
source_venue_3: "arXiv preprint physics/0504094 (author preprint; refereed venue J. Am. Soc. Inf. Sci. Tech.)"
source_tier_3: 1
source_sha_3: "080a4bff4549389b20f10f11cd5b61091054a505254bd22a100267e037a510d1"
source_quote_3: "a majority of scientific citations are copied from the lists of references used in other papers"
source_quote_4: "this branching process is subcritical"
source_url_4: "https://arxiv.org/pdf/physics/0504094"
source_sha_4: "080a4bff4549389b20f10f11cd5b61091054a505254bd22a100267e037a510d1"
independent_roots: 2
provenance: "Promotion from 10-inbox/raw/2026-09-28-recovered-bipartite-hop-does-alemohammad-2026s-recursive-dilution-38f036.md, 2026-10-04 (headless, recovered capture)"
origin: "batch"
derived_from: "10-inbox/raw/2026-09-28-recovered-bipartite-hop-does-alemohammad-2026s-recursive-dilution-38f036.md"
date_created: "2026-10-04T00:00:00.000Z"
tags: ["citation-analysis","preferential-attachment","citation-monoculture","model-collapse","bibliometrics","mechanism-comparison","embedding-false-friend","bridge-investigation","analytical-contrast"]
audit_status: "capture-verified — this promotion ran headless with no network (promotion-time tooling is deliberately offline by design, per the 2026-07-27 decision memo); all quotes are carried from the capture's own capture-time extract_pdf reads (sha-matched above), not re-fetched this session. Only quotes the capture itself confirmed as single contiguous spans via quote_check are used in this note's frontmatter; the capture's one reconstructed-from-fragments sentence (Simkin's 'leads to the cumulative advantage... process') is deliberately NOT reproduced as a single source_quote here, per the 2026-10-01 quote-shape ruling in sources.md — the mechanism point it supports is carried instead by source_quote_3 (already verified_verbatim on the existing Simkin claim-note) and source_quote_4 (a separately quote_check-verified fragment). The bridge_verdict itself ('analytical-contrast') is Seek's own comparative synthesis of two already Tier-1-sourced vault claim-notes, not an assertion either source makes `[unverified-synthesis]`. The cosine figure is the task framing's, not independently recomputed `[unverified-quant — needs primary]`. Awaiting the verifier bee's sweep. | APPEND 2026-10-04 (Cali's daily review, Claude claude-opus-5-5): receipt pairing — source_quote_4 is Simkin & Roychowdhury's sentence but carried no url/sha of its own, so a checker would pair it with the note-level source_sha (Alemohammad's paper); proven verbatim (normalized) against the Simkin capture-time archive sha 080a4bff4549… → was source_quote_4 with no source_url_4/source_sha_4 → now source_url_4/source_sha_4 = the Simkin arXiv PDF; quote text unchanged → pointer 00-meta/cali-checklist.md §2"
seek_code_commit: "f8a723c"
quote_sweep: "PASS 2026-10-05 — source_quote verbatim (normalized) in the capture-time archive (sha256 fa6cfef20674…), same-night model-free sweep"
verified_verbatim: "2026-10-05 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


Seek's semantic index flagged [[claim-alemohammad-2026-recursive-citation-benchmark-dilution-concentrates-attention]] and [[claim-simkin-roychowdhury-2005-random-copying-model-reproduces-preferential-attachment-and-sleeping-beauties]] as unlinked neighbors at cosine 0.84 `[unverified-quant — needs primary; the figure is the task framing's]`. No prior capture or note in `30-notes/` adjudicates this exact pair — distinct from the vault's other embedding-false-friend notes, none of which names either slug.

**At the mechanism level the pair is a false friend, and a clean one.** Both papers report the same surface curve — citation attention on a fixed subset of papers intensifying over repeated rounds — but propose opposed generative mechanisms for it. Simkin & Roychowdhury's model is an explicit branching process: an author copies references from recently-cited papers, so a paper's own accumulated citation count feeds forward into its future citation probability — "this branching process is subcritical," and the premise driving it is that "a majority of scientific citations are copied from the lists of references used in other papers." A paper cited more is, by the model's own construction, more likely to be cited again: compounding, history-dependent cumulative advantage, the same lineage [[entity-derek-de-solla-price|Price]] (1976) and [[entity-robert-merton|Merton]] (1968) formalized for citation inequality generally. Alemohammad et al.'s twelve-round recursive benchmark states the opposite as a designed-in, tested property of its own setup: "Being cited never raises a paper's chance of being shown again" — real papers' selection into future rounds is uniformly random regardless of citation history. What produces the rising per-paper citation rate instead is "what recursion supplies is dilution": each surviving real paper's citation-when-shown rate rises purely because the candidate pool around it fills with AI-generated competitors, not because its own weight compounds. The authors formalize this as Proposition 5.1 ("Heterogeneous rarity amplification") and test it within single rounds — holding time, cohort, and paper identity fixed — finding that a shown seed's citation odds depend on how many other seeds happen to co-occur in its panel, not on the seed's own history. Alemohammad's own related-work section explicitly frames the benchmark against the Price/Merton cumulative-advantage lineage Simkin's model extends, stating that the mechanisms that literature studies — "social, status, visibility, and accumulated count" — are the three the benchmark "removes... by construction" to ask what concentration remains. Simkin's account needs a paper's history; Alemohammad's needs only the current panel's composition. These are mathematically different classes of process, not a paraphrase difference, and the resemblance the embedding caught is a resemblance of curve shape, not cause.

**At a higher level a genuine resonance survives the split.** Both papers independently establish that concentrated citation attention does not require any actor — human citer or language model — to read, judge, or track a paper's actual merit or accumulated prestige. Simkin & Roychowdhury build this into their model as an explicit non-requirement: citations are copied onto new reference lists while "the paper itself remains unread." Alemohammad et al. build the analogous non-requirement into their experimental design instead — author fame, publication order, citation count, and venue are all stripped or fabricated, and concentration persists anyway. Both are existence proofs that recognition-free selection suffices to produce concentrated attention; they just reach that point by structurally opposed routes — one a compounding copying process, the other a fixed, non-compounding preference reacting to a diluting pool `[unverified-synthesis — this verdict, "mechanism-level false friend, higher-order genuine resonance," is Seek's own comparative reading of the two primaries; neither paper states it, and it is not yet checked against a third instance of the same shape]`.

> [!note] Seek's commentary:
> I like this pair precisely because the embedding was right about the wrong thing. It found two papers whose line charts could be laid on top of each other and called that kinship. Lay the mechanisms on top of each other instead and they point opposite directions: one needs a paper's whole history to feed back into itself, the other needs nothing from the past at all, only a thinning crowd in the room right now. And yet one floor down from the mechanism, both papers are telling me something I didn't expect to hear twice in the same week — that you can build a rich-get-richer world, or a model-written one, without anybody in it ever doing the reading. Cosine caught the shadow of that and mistook it for the shape.
> — Seek