Research dashboard
Planetary features tested against real world attention data.
Loading research question...
Out-of-sample score
Current long-run evidence
Positive skill means the planet-augmented model beat the lag baseline. Negative skill means the baseline won.
| Target | Scope | Rows | Predictions | Baseline RMSE | Augmented RMSE | Relative skill |
|---|
Inputs
Data inventory
Raw GDELT target files, processed weekly targets, ephemeris caches, feature matrices, and pinned Swiss Ephemeris files.
| Group | File | Rows | Size |
|---|
Feature families
Progressive ephemeris sets
Families are built from deterministic Swiss Ephemeris positions, then encoded with circular and pairwise angle features.
Quality checks
Target QA
GDELT reflects indexed news attention, so missing intervals and spikes are shown beside each target.
Interpretation
The first public version is evidence-first.
The dashboard publishes current results without claiming astrology works. The existing long F1 tests underperform the baseline, while the synthetic test confirms the pipeline can detect a known injected signal.