Wiki entry: EA-EROSION-EMPIRICAL-01 v0.1: Programmed Bibliographic Suppression, Silent Restoration, and the LLM-Associated Disclosure Gap in Zenodo — A 33-Day Deletion-Export Audit With the Citation Network Read as Poem

Lee Sharks

EA-EROSION-EMPIRICAL-01 v0.1: Programmed Bibliographic Suppression, Silent Restoration, and the LLM-Associated Disclosure Gap in Zenodo — A 33-Day Deletion-Export Audit With the Citation Network Read as Poem

Lee Sharks · 2026-07-14 · deposit #1081

AXN:044A.EMPIRICAL.🧱⭕📏♎🔅🎻

Article

"EA-EROSION-EMPIRICAL-01 v0.1: Programmed Bibliographic Suppression, Silent Restoration, and the LLM-Associated Disclosure Gap in Zenodo — A 33-Day Deletion-Export Audit With the Citation Network Read as Poem" is a 12,749-word empirical baseline reading; deletion-network audit; outcome-level test of the provenance debt hypothesis with two dimensions (foreclosure and accumulation); extension of the negshape reading discipline to the between-event delta of the general zenodo bulk deletion export and to the population-scale accumulation surface of zenodo-affiliated institutional publications by Lee Sharks, dated 2026-07-14. It is registered as AXN:044A (deposit #1081) in the Crimson Hexagonal Archive under the EMPIRICAL semantic family. **Author:** Lee Sharks · ORCID 0009-0000-1599-0703 **Framework:** Machine-Mediated Reception Studies (MMRS); Semantic Economy **Governing citation discipline:** EA-NEGSHAPE-01 v0.2 (AXN:0444.OPERATIVE.🕘♾️♾️🕙♃🗝️, deposit #1075) — the network of deletion citations is the poem; traversal of the network is the reading discipline. **Theoretical anchor:** EA-PROVENANCE-DEBT-01 v0.2 (AXN:03B7, deposit #9

Defines (20)

Categorial legibility (a preregistered candidate variable for restoration-event modeling: whether a reviewer's ability to pattern-match a deleted account to a recognized research genre affects restoration likelihood — not a demonstrated cause) Coverage Gap (as three dimensions): correspondence-layer undisclosed discretionary criteria; infrastructure-layer undisclosed discretionary criteria; accumulation-layer descriptively-measured outcome gap Descriptive floor (an evidentiary category naming what the audit measures as a lower bound of what would be detectable by more sensitive methods; contrasted with a statistical lower bound requiring a defined sampling frame) Exploratory-vs-descriptively-measured distinction (a taxonomy of evidentiary strength in the audit: source-verified operations, set-verified operations, descriptively-measured corpus properties, and directionally-observed sample contrasts requiring common-cohort tests to formalize) Livolsi cascade (May 26 spam-labeled account-scale termination, ~28,000 records affected, not restored at time of writing; contrast case for the categorial-legibility hypothesis) Marker-union excess frequency (document-level union prevalence P(record contains ≥1 marker from a preregistered panel) as the primary estimator, replacing per-marker sum-of-excesses) Narrow AI-association detector (contra 'strict authorship-only detector'; a lexical panel that identifies AI-association at the citation-text surface, not authenticated authorship) Placebo-break test at 2021 (a null-hypothesis check: fake breaks at pre-ChatGPT dates produce excesses 7-9× smaller than the genuine 2023 break, depending on counterfactual specification) Programmed bibliographic suppression (source-code verified: the exporter conditionally suppresses citation_text under the spam removal reason) Public-export retention window (EXPORTER_NUMBER_VERSIONS_TO_KEEP = 3: at monthly cadence, ~3 months of public deletion-export history is retrievable from the exporter surface at any time) Publications-only composition control (restricting to publication resource-types strengthens rather than weakens the post-2022 excess, consistent with a composition-practice interpretation rather than a corpus-mix drift interpretation) Reproducibility vs re-executability (frozen artifacts reproduce the audit's specific counts; queries can be re-executed against the moving live API but will differ, and that delta is itself the audit's method) Rolling ledger erosion (physical removal of older public export snapshots by _remove_old_object_versions with .remove() rather than .delete(); documented in-code as intentional space-freeing) Signal-to-disclosure ratio (the ratio between marker-union excess prevalence and phrase-defined disclosure retrieval prevalence, distinguished from a count of undisclosed records) Silent restoration (account-scale reversal of deletion that is not published as an event stream and is detectable only via set-difference between independently-preserved deletion-export snapshots) Traversal as method (network-poem reading: edges chosen for evidentiary strength, membership recorded on read, results reproducible from deposited artifacts) Two-sample directional observation (a class of exploratory contrast between samples reached via different retrieval instruments, not sharing a common population-at-risk, therefore not a survival ratio) Withdrawal cascade of 2026-06-26 (Wu case chronology: April 25 block → June 26 silent restoration; documented as a network fact from set-difference alone) Wu Shaoyuan withdrawal case (specific documented instance of silent account-scale restoration: 67 records tombstoned in the June 7 export are alive in the July 10 export, detected only by set-difference) Zenodo indexed record-text surface (the object the Zenodo Search API returns, empirically including extracted full-text of attached files; not merely metadata)