Citations are necessary, not sufficient
A citation count tells you that other researchers have read a paper. It does not tell you that the result is correct, that it has been independently rebuilt, or that the field has even tried. The lab tracks both numbers — the citation timeline and the replication ratio — and publishes them side by side.
A claim crosses the 100-citation threshold → replication audit opens.
Once a claim has accumulated 100 citations, the lab automatically opens a replication audit and posts the results in the public journal. We do this whether the claim is one of ours or not.
What we count, and what we do not
We count appearances in the OpenAlex and arXiv indexes, weighted by venue type, with self-citations excluded by default. We do not count blog mentions or social-media posts in the headline number, though those are tracked in the secondary feed for context.
| Source | Weight | Self-cite filter |
|---|---|---|
| Peer-reviewed venue | 1.00 | yes |
| Workshop | 0.65 | yes |
| Pre-print | 0.40 | yes |
| Thesis | 0.50 | yes |
| Book chapter | 0.85 | yes |
| Patent | 0.20 | no |
Replication tracker
Every cited paper from the lab carries a status flag in the public tracker:
- Replicated — at least one independent group has rebuilt the central figure to within reported error bands.
- Pending — replication is open but not yet complete; reviewer assigned.
- Adjusted — replication revealed a bound that needs updating; the paper carries an addendum.
- Withdrawn — replication failed and the claim has been withdrawn. We publish the negative result with the same prominence as the original.
Top venues
The headline citation count masks a long tail. The five venues below account for 62% of the indexed citations.
NeurIPS
Reinforcement learning and agentic-systems track. Largest single venue contribution.
Nature Methods
Methodology pieces on reproducibility pipelines and content-addressed artefact management.
IEEE S&P
Cryptographic protocol verification, including the 18-protocol audit series.
ICML
Adversarial robustness benchmarks and counterfactual cohort analysis.
How to read the impact page
If you have arrived here from a citation list and want to know whether a particular paper has been replicated: the public tracker is queryable by DOI. If the tracker says Pending, the lab welcomes your replication — and will cite it.