diff --git a/README.md b/README.md
index 2ce6b2f..5f489dd 100644
--- a/README.md
+++ b/README.md
@@ -101,7 +101,7 @@ You can always check which modules are installed/available to you by running `ge
 ### Composite / Ensemble Utilities
 
 1. `ZScoreRescaledScorer`
-   - Wraps any per-variant scorer and rescales its log-ratios within an amino-acid group (destination residue or substitution type) — the z-score ("Z") ranking from the [MULTI-evolve](https://www.science.org/doi/10.1126/science.adr8628) ensemble. Behaves as a single scikit-learn transformer (`fit`/`transform`), and `score_table` returns the per-mutation DataFrame.
+   - Wraps any per-variant scorer and rescales its log-ratios within an amino-acid group (destination residue or substitution type) — the z-score ("Z") ranking from the [MULTI-evolve](https://www.science.org/doi/10.1126/science.aea1820) ensemble. Behaves as a single scikit-learn transformer (`fit`/`transform`), and `score_table` returns the per-mutation DataFrame.
    - See the "Ensemble Variant Nomination" user guide for a full structure + sequence first-round scan. Lower-level primitives (`per_variant_mutation_info`, `zscore_by_aa_group`) live in `aide_predict.utils.scoring`.
 
 ### Embeddings for Downstream ML
diff --git a/docs/user_guide/multi_evolve_ensemble.md b/docs/user_guide/multi_evolve_ensemble.md
index c8beff9..63cc793 100644
--- a/docs/user_guide/multi_evolve_ensemble.md
+++ b/docs/user_guide/multi_evolve_ensemble.md
@@ -6,7 +6,7 @@ title: Ensemble Variant Nomination
 
 ## Overview
 
-This guide reproduces the first round of the [MULTI-evolve](https://www.science.org/doi/10.1126/science.adr8628) zero-shot nomination workflow using AIDE primitives. The idea is to score a full saturation-mutagenesis (SSM) library along two complementary tracks and combine them:
+This guide reproduces the first round of the [MULTI-evolve](https://www.science.org/doi/10.1126/science.aea1820) zero-shot nomination workflow using AIDE primitives. The idea is to score a full saturation-mutagenesis (SSM) library along two complementary tracks and combine them:
 
 - a **structure track** — `ESMIFLikelihoodWrapper` (ESM-IF1), conditioned on the backbone;
 - a **sequence track** — an ensemble of sequence PLMs (`ESM2LikelihoodWrapper` loading the ESM-1v / ESM-2 checkpoints), averaged.