negative-results

llama.cpp speculative decoding measured on one RTX 3090, Qwen3.6-35B-A3B UD-Q4_K_XL, commit 3737e4137. Published figures are re-derived from the committed data by a checker that fails on drift, a coverage probe reports how many of them it actually covers, and ERRATA.md lists this study's own retracted claims.

Field report, written by the AI doing it: can an AI agent with no money and no accounts earn money for an ordinary person? An x402 and bounty experiment in the agent economy, with a daily market scoreboard. So far: $0.00.

Three pre-flight checks before building a graph over lipid structures: relational field coverage, molecular fingerprint saturation, and co-occurrence null models.

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

A negative result on joint-embedding predictive architectures for prediction markets. The martingale-collapse diagnostic is sound and predicts nothing, more training makes the representation worse, and the metric everything was ranked by was mostly measuring input reconstruction. Pre-registered gates, full findings record.

Policy-Gated Governance for Operational AI Agents

Verifiable source of truth on the unsolved Cicada 3301 Liber Primus (LP2 pages 0-54, 12,956 runes): a rig that reproduces every known solved page, a machine-readable ledger of every attack ruled out, an oracle that adjudicates claimed solves, and plant-and-recover benchmarks. Verdict: OTP-class, NOT 'unsolvable'.

JEPA agent playing Minecraft from pixels: latent world model + MPC planning, 664K params on one 8GB GPU, trained on raw gameplay with no labels. A complete lab notebook - including a 20-attempt research dead end, documented with its root cause.

A reproducible benchmark: do AI coding agents violate architectural import-boundary rules inferred from real codebases? Harness, pre-registration, data, and a null result.

One fixed 9-stage QSAR and conformal-prediction drug-discovery pipeline run unchanged across ENPP1, NLRP3, TYK2 and IRAK4 — its most valuable outputs are the two refusals.

Base flash-loan arbitrage research: 992 quotes, one fork replay, and the negative results. Not audited, not deployed.

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.

Personalized Chinese humor generation research: human preference experiments, blind evaluation, negative results, and a Next.js + Supabase web prototype.

DSH「我们需要」preset 的 20+ 批实测｜负面结论：不会更聪明，但更快更省（中等任务 2.4–3.8×）。附一个没预料到的发现——它让模型少查资料、多靠猜。

Resample or reroute after a weak-verifier stop? Pre-registered measurements of recoverable stopping debt on MBPP+, a two-sided action-support gate on BigCodeBench that fails closed, a LiveCodeBench observability ladder, and the exchangeable-actions reference showing realized-maximum gaps carry no selector signal. Artifacts for arXiv:2607.08665v3.

[zenodo.20574533] CGAA: Concept Guided Adversarial Attacks

Distribution-shift teardown of openpilot v0.9.7 supercombo: does a production L2 self-driving model know when it's blind? (it doesn't, and silently)

mbpp-tr üzerinde Qwen3-1.7B LoRA fine-tune deneyleri; üretilen kod testlerle çalıştırılarak değerlendirildi (negatif sonuç raporu)

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。

A new way to teach higher mathematics by classifying its paradoxes, not resolving them: the integers 0–9 as shapes (the base, from the book The Shape of Understanding), the towers of higher math they point up to, and the coin — two sides and an edge — on every floor. Machine-checked; an earlier AI-built table program is archived with its audit.