Weights that never freeze. A continual-learning architecture, and the pre-registered experiment that falsified its central claim. https://arxiv.org/abs/2607.20792

从20个项目中系统提取可复用方法论模式的实验记录。包含10轮正式审查实证（4后端）、58项发现、G5可追溯审计。不成熟框架，诚实的实验记录。

"Publication bias and the canonization of false facts" published in eLife (2016)

Reproducible evaluation harness for hidden coordination variables in multi-agent LLM systems.

[zenodo.20574533] CGAA: Concept Guided Adversarial Attacks

A $0 falsification lab across two markets — crypto (~111 hypotheses) and Polymarket prediction markets (172,830 resolved markets, 1.36M trades). 184+ techniques through one committed anti-overfitting gauntlet. 0 survive to a deployable edge; the reusable validation harness is the asset. Agent-ready. MIT.

Can a model estimate software effort from public data? A pre-registered blind test on nine open datasets — reproducible, and honest about the negative result.

TACT: signed, label-free confidence weighting for self-consistency voting — with the thin-window boundary (2.5–7.5% of items) that explains why six other designs died

A rigorous negative result: no pre-publish feature predicts YouTube Shorts engagement above chance, and the one 95% model is a leakage trap.

Signed attention for transformers — sinh/cosh attention giving weights in [-1,1] instead of softmax's positive-only, made FlashAttention/SDPA-compatible by channel doubling. Includes a from-scratch softmax-vs-SBA comparison and a documented negative result.

Synchronization Resistance: a pre-registered study measuring multi-LLM agreement difficulty. 3 pass / 1 inconclusive / 1 fail — critique wanted.

A global, open-source registry and standardized schema for null findings, non-significant outcomes, and failed trials in medical research. Built to eliminate publication bias and accelerate biomedical discovery.

A negative-result study and a falsification protocol for LLM memory systems.

Diagnostic study of adaptive gating failure in vision-language prompt learning, focusing on gradient imbalance and gate collapse under frozen CLIP backbones.

Bounded negative results for subject-independent EEG affect evaluation under held-out controls.

Graph-Tabular Fusion for Bitcoin Fraud Detection - Demonstrating when Node2Vec embeddings don't improve XGBoost. Scientifically rigorous negative result validating that tabular features encode graph structure.

An autonomous trading agent that has never placed a single trade — and that's the point.

Cryptographic commitment registry for negative-result publication and research integrity.

Is a ternary (2/9)-sigma zero band worth its keep for compressed retrieval? 9 preregistered experiments, honest negatives included.

Code, data, and frozen records for testing whether short-context mechanistic probes predict RULER effective context length beyond model size.