Publication and author-overlap sensitivity
This is a 2026-08-29 publication-status snapshot for the 114 included papers. It does not reconstruct status at inclusion: per-paper inclusion/acceptance dates are unavailable. Coding is filtered from the included copies rather than fully re-extracted from every version of record.
Released results
| File | Rows | Content |
|---|---|---|
| publication_status.csv | 114 | Verified status, public evidence URLs, overlap identities, and adjudication notes |
| claim_dependency.csv | 44 | Review-level and local claim changes under removal scenarios |
| method_leaf_sensitivity.csv | 32 | Method-leaf support under reviewed-only and archival-only definitions |
| peer_reviewed_automation_replacements.csv | 10 | Targeted replacement exemplars with direct source links and scope limits |
| summary.json | — | Compact frozen counts, definitions, and limitations |
The status classification contains 74 formal archival publications, three accepted non-archival workshop papers, 22 preprint-only papers, 14 technical/model reports, and one unconfirmed formal-successor link. Of the workshops, two have verified peer review; the third has an unclear review process. S074’s possible formal successor remains unconfirmed.
Main comparisons
The primary reviewed-only scenario retains 76 papers: 74 verified archival publications and two reviewed workshops. The archival-only boundary retains 74. Both retain the eight RQ1 dimensions, 13 RQ2 headings, 12 RQ3 families, Correctness dominance, and all high-level detection/mitigation families. Fine-grained support changes: only 28 of 32 detailed method leaves remain. A targeted full-text audit supplies reviewed automation exemplars for affected broad headings.
An author-overlapping paper has at least one author who is also a manuscript author after identity disambiguation. 18/114 papers (15.8%) meet the definition: 11 formal publications and seven currently preprint-only papers. Removing the seven leaves 94 RQ3 assignments from 62 papers, all 12 families, Correctness at 58/94 (61.7%), and the same four leading families at 64/94 (68.1%).
Local findings are sensitive: one redundancy-to-derailment branch loses all support, and the small Clarity family loses its primary experiment. The original one-exemplar generated-code automation crosswalk drops from 8/8 to 6/8 under that removal; targeted non-overlapping reviewed replacements restore narrow broad-heading existence. The replacement table does not establish exhaustive subtype coverage or deployment maturity.
A separate upper-bound check removes all 18 overlapping papers, including 11 currently formally published papers. It is a deliberately broader removal scenario. “Mainly dependent” means strictly more than 50% of direct support is lost; exactly 50% is a boundary. “Pivotal” identifies exact, ranking, or evidence-type statements that fail even if a majority of support remains. Consult the claim-level rows rather than interpreting preserved high-level structures as unchanged fine-grained evidence.