physicalreasoning

physicalreasoning/causal-jepa

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.

4 good first / help-wanted issues · Python · last activity Oct 11, 2026

9 stars 0 forks 9 watchers Python Apache License 2.0
jepa machine-learning negative-results prediction-markets pytorch representation-learning reproducible-research research self-supervised-learning time-series
4 Open Issues Need Help Last updated: Oct 11, 2026

Open Issues Need Help

View All on GitHub
physicalreasoning/causal-jepa
9

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.

Python
#jepa#machine-learning#negative-results#prediction-markets#pytorch#representation-learning#reproducible-research#research#self-supervised-learning#time-series
physicalreasoning/causal-jepa
9

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.

Python
#jepa#machine-learning#negative-results#prediction-markets#pytorch#representation-learning#reproducible-research#research#self-supervised-learning#time-series
good first issue research
physicalreasoning/causal-jepa
9

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.

Python
#jepa#machine-learning#negative-results#prediction-markets#pytorch#representation-learning#reproducible-research#research#self-supervised-learning#time-series
physicalreasoning/causal-jepa
9

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.

Python
#jepa#machine-learning#negative-results#prediction-markets#pytorch#representation-learning#reproducible-research#research#self-supervised-learning#time-series