Published papers, replicated
We take one computational claim from a published paper, write down the decision rule and the controls, freeze that protocol and deposit it on Zenodo, and only then run the authors' released code. The authors see the result first. A paper is listed here after they have had the chance to respond; results still inside that window are not listed.
18 listed: 13 replicated, 3 inconclusive, 2 did not replicate. Each outcome is the verdict under our protocol on the claim named, not a verdict on the paper.
- Nov 5, 2026Replicated
GatedV: Self-Gated Values for Transformer Language Models (Yu and Zhou)
Claim tested: At 2.33M parameters on FineWeb-Edu with the repository's recipe, SiLU on the attention value projection lowers mean held-out perplexity by about 2.0% against standard attention, and every paired seed improves.
- Nov 1, 2026Did not replicate
REVBRAIN: Real-time Probabilistic EEG Forecasting Using Residual Vector Quantization (Pankka, Lehtinen and Roine)
Claim tested: Median inference latency of autoregressive WaveNet is about 52x REVBRAIN's at 1 sample per input and about 3170x at 100 samples per input (released latency script).
- Oct 30, 2026Did not replicate
Claim tested: A fully complex FNet-style causal LM matches a param-matched real-valued FNet in under half the steps on tiny-shakespeare.
- Oct 30, 2026Replicated
Claim tested: On the width-192 binary MLP benchmark a 4-bit error-diffusion buffer with a grid level at the flip threshold matches an 8-bit buffer and stays near the latent-weight reference with 5 vs 97 bits of state; the naive clamp is worse.
- Oct 27, 2026Inconclusive
Conditional GraphGANFed: Optimizing Graph-Structured Molecule Generation in Federated Generative Adversarial Networks (Manu and Alazzwi)
Claim tested: The conditional model (cGraphGANFed) generates molecules with about 10% higher drug-likeness (QED) than the GraphGANFed baseline.
- Oct 27, 2026InconclusiveAuthors reviewed
MetaLearnNCA: Few-Shot Offline Meta-Learning via Interacting Neural Cellular Automata (Guichard and Nichele)
Claim tested: Trained on Omniglot, the model reaches 89.71% 10-shot accuracy on MNIST against 79.17% for ProtoNet, and 51.26% on KMNIST against 36.23% for first-order MAML.
- Oct 26, 2026ReplicatedAuthors reviewed
TreeWalker: Partial Evaluation for Grouped Tree-Ensemble Inference (Karatay and Newman)
Claim tested: Partial evaluation speeds up tree-model inference over a row-independent walk.
- Oct 26, 2026Replicated
Distilling Directional Verification (Lee et al.)
Claim tested: Students trained on known-direction four-teacher-mean labels beat summed-DC reverse-label students on open accuracy.
- Oct 23, 2026Inconclusive
OpenOpt: An Open-Source SRAM Optimizer Based on Equivalent Circuit Model (Wang et al.)
Claim tested: Simulating an equivalent-cell circuit in place of the full SRAM array is many times faster (the paper's figure shows 8.1x at 64x64 rising to 61.4x at 256x512) with delay and power errors under about 2%.
- Oct 22, 2026Replicated
Persistent Recursive Worlds Enable Autonomous Software Evolution (Huang, Liang, Zheng and Cheng)
Claim tested: Rust driver is faster than the Fortran driver on six workloads (bounds 1.1/4.5/2.5/1.0/1.1/1.0).
- Oct 18, 2026ReplicatedAuthors reviewed
PVBRF-ID: Private, Verifiable and Byzantine-Robust Federated Intrusion Detection for Oilfield Enterprise Networks (Aljumaily et al.)
Claim tested: PVBRF-ID beats FedAvg and is close to multi-Krum.
- Oct 18, 2026Replicated
Machine Unlearning for Federated Socially Fair k-Means Clustering (Reddy, S K, Raza and Prakash)
Claim tested: Replay is about 35.8x faster than retraining for whole-client deletion.
- Oct 18, 2026Replicated
Sparse Federated Echo State Networks for Time-Series Classification via Smooth-L1 Regularization (Lyu, Röder and Schleif)
Claim tested: At one fixed threshold, the sparse readouts are 77 to 95% sparse (centralised) and 75 to 95% sparse (federated) and, on five of seven benchmarks, within 0.7 and 1.6 accuracy points of the stronger dense readout.
- Oct 15, 2026Replicated
Evolutionary Architecture Search for Chlorophyll-a Prediction in Lakes using Sentinel-2 (Komurcu and Petkevicius)
Claim tested: NAS-1 has higher test AUC than the baseline MLP.
- Oct 15, 2026Replicated
Training on the Future: A Delay-Aware Audit of Test-Time Adaptation for Time-Series Forecasting (Fentazi, Ameur and Ksentini)
Claim tested: RLS-Bank cuts stream MSE by at least 25% versus TAFAS and PETSA.
- Oct 14, 2026Replicated
Gaussian BROJA Partial Information Decomposition in Closed Form (Sheelo, Golan and Pakman)
Claim tested: Closed-form Gaussian PID is orders of magnitude faster than the iterative solver, with matching atoms.
- Oct 11, 2026Replicated
IGA-KAN: Isogeometric Analysis with Physics-Informed Closed-Form Kolmogorov-Arnold Networks for Forward and Inverse PDEs (Naraghi, Parand, Sadr and Rahmati)
Claim tested: Released code reproduces the paper's tabulated L2/H1 gains over baselines on nine rows.
- Oct 9, 2026Replicated
Variational Autoencoders for Collaborative Filtering (Liang, Krishnan, Hoffman and Jebara)
Claim tested: Mult-VAE^PR beats the best simple baseline on ML-20M.
Our own results are re-tested the same way: see the track record. To have a claim of yours tested, see request a test.