Campaign One · Status: Running — Phase 1
Dead papers don’t stay buried. They get cited.
The Zombie Literature Audit: every retracted paper in biomedicine, cross-referenced against every paper that cited it afterward. The corpus defines itself — a paper is in because it was retracted, not because we picked it. Nobody can call it cherry-picking. There is no picking.
| Campaign | 001 — The Zombie Literature Audit |
|---|---|
| Corpus | All retracted biomedicine papers (Retraction Watch database via Crossref — 72,790 records4) × every citing paper (OpenAlex citation graph) |
| Data cost | $0. Every source is open. |
| Budget | Under $50,000 all-in — corpus engineering and expert adjudication, not compute |
| Method | Published before the run. Raw outputs published with the results. |
| Output | Open dataset · league tables · notification ledger · plain-language report |
Why biomedicine
Three reasons. The retraction record is densest here — decades of it, machine-readable. The full text is open: the PubMed Central Open Access subset holds more than ten million articles, free to audit. And the stakes are clinical: treatment guidance and evidence reviews are built on this literature, citation by citation.
The expected yield is not a guess. It is pre-measured. About 94% of post-retraction citations show no awareness of the retraction.1 In the most careful case study done to date, researchers traced the papers citing one fraudster’s retracted work: 39 of 88 had conclusions that were likely to be substantially weaker once the retracted papers were removed — and journals had flagged just 4 of them.2 One team ran that experiment for one fraudster. Nobody has run it for the whole corpus. That is the campaign.
The three phases
Phase 1
The Map
Purely mechanical. Every retracted paper × every post-retraction citation. Existence and status only — no judgment calls. Published as an open dataset with league tables by journal and field. The counters on the home page are this phase, live.
Phase 2
The Context
For citing papers with open full text, the checker classifies each citation: relied on it, mentioned it, or cited it knowing it was retracted. A stratified sample gets human adjudication. Output: the reliance rate — and the exposure list of meta-analyses whose conclusions rest partly on retracted inputs.
Phase 3
The Replies
Every journal that published a reliant citation gets notified, with the evidence attached. Then we publish who corrected, who responded, and who ignored it. The ledger of replies is part of the result. Editors have said for years that correcting the record is not a priority. This phase prices that sentence.
What gets published
- The complete open dataset — every retracted paper, every post-retraction citation, every classification. Re-runnable by anyone.
- League tables by journal and field, built as shareable cards: source, date, and denominator baked into the table itself, so a screenshot carries its own citation.
- The notification ledger: which journals were told, when, and what they did.
- A plain-language report. No jargon shield.
Post-retraction citations by journal — top 10
Counting… publishes with the Phase 1 dataset.
Method, in short
- Retraction dates come from Crossref updated-by events — the authoritative record. The OpenAlex retraction flag was independently validated at 95.8% recall against a Retraction Watch gold-standard sample.3
- Context classification runs only where full text is openly available. Coverage is labeled on every output — never silently extrapolated.
- A post-retraction citation is a flag, not an accusation. Authors usually cannot know: the system failed to carry the correction to them. The audit measures that failure — of the system, not the scientist.
- Anyone named in the results gets a right of reply, attached to their record.
Status: Running — Phase 1, the map · Counters: home page · Data: results.json
Sources — every number on this page
- Investigation published in JAMA (1998): 94% of 299 citations to retracted articles did not note the retraction. Schneider (2021): 94% of post-retraction citations in biomedicine showed no awareness of the retraction.
- Avenell et al., research on citations to one disgraced researcher’s retracted trials, reported by Science: of 88 citing papers, 39 had conclusions likely to be substantially weaker without the retracted work; journals flagged 4.
- OpenAlex is_retracted flag validated against a Retraction Watch gold-standard sample drawn from the Crossref-distributed database: 95.8% recall (115 of 120 sampled retractions flagged). Crossref updated-by events supply the dated retraction notices.
- Retraction Watch database, distributed via Crossref: 72,790 retraction records (October 2026), updated daily. Citation graph: OpenAlex (CC0, ~260M works). Full text: PubMed Central Open Access Subset (10M+ articles).
Re-run this: Campaign One’s dataset, code, and full method publish with the run. The live counters are in results.json.