Progress history
Releases processed over time
Loading history…
Results dashboard
AI-driven research requires evidence. Here we compare published final tests of machine-learning models and statistical strategies, including baselines and uncertainty. The ongoing Qwen study does not yet have final predictive results.
Qwen3-4B LLM · 6,493 earnings releases
Final predictive figures will not be disclosed until the scheduled evaluation. Only factual processing information is shown below.
Progress history
Loading history…
Recently processed
This list shows the most recently saved successful analyses and refreshes with each public update.
Published final tests
Earnings releases · replication
Model’s 95% interval: 47.20–52.51%. The model did not outperform the simple baseline.
Oil → Shell
Tested on 163,962 one-minute observations across four forward-looking horizons.
Overnight effect
251 holdout days; uncertainty intervals include zero.
Earnings releases · initial study
Interpretable text-feature model, not the TF-IDF variant (50.00%). 118 holdout releases. The intervals include zero, which is why an independent replication followed.
Insider clusters
289 holdout decision points. Average cluster returns were uncertain and did not survive costs.
Reading guide
Learn how the baseline, holdout, uncertainty interval, costs and replication jointly determine the conclusion.