What the score actually is.

The production score is a weighted blend of traditional astrological techniques: partile transit hits (benefic vs challenging), secondary-progression support, planetary-hour alignment, and Hellenistic / Vedic timing flags. Weights are theory-anchored — chosen from the classical literature each technique is drawn from, not fitted to data.

Think of it as an astrological calculator with a structured point system. If you're an astrologer, you can read it as "are today's classical supportive conditions stacking up, or are today's challenging ones?" If you're not, treat it as a reflective timing heuristic — a prompt, not a prediction.

Not an empirically validated predictor. We pre-registered 53 astrological hypotheses and tested them at 10,000 permutations on a cohort of 596 peak events and 450 setback controls, with a subject-level discovery / held-out split and a label-shuffle null. Only H34 replicated in the held-out arm. Its most parsimonious reading is cautionary rather than lucky: Thursday / Friday benefic-weekday lore was lower than null for peak rows and higher in setback rows. Five other discovery survivors failed held-out confirmation.

Cross-validated classification AUC is 0.505 — chance.The scorer as a "lucky vs unlucky day" binary classifier does not outperform a coin flip on held-out data.

We retracted earlier claims.Earlier versions of this page referenced "BY-FDR survivors," a "15.1% held-out lift," and numbered hypotheses as validated signals. Those claims were artifacts of a statistical-null construction error (subject-shuffle where label-shuffle was correct) plus a discovery / held-out subject leakage. When the analysis was rebuilt correctly, none of those claims survived.

The complete rebuild and its corrected pipeline live on the research/phase1-rebuild branch. The honest findings document is docs/research/phase1-rebuild-findings.md.

The separate lottery / windfall research lane now runs in the production model as research-anchored-v4-N593. The current handoff adds Apophis (99942), Lucifer (1930), Mean Black Moon Lilith, and Asteroid Lilith (1181) to the ephemeris engine and promotes two evidence-weighted features into the scoring layer: Jupiter-to-Apophis partile activation and Jupiter-to-Mean-Lilith partile activation.

In the latest event-weighted lottery permutation run, H108 Jupiter-to-Apophis measured 14.4% observed versus 4.5% null (Cohen's h +0.35, p=0.0035). H109 Jupiter-to-Lilith measured 13.3% observed versus 5.3% null (h +0.28, p=0.0185). The active backtest across lottery-winner rows showed a 0.700 mean same-day percentile, 0.708 median, 30.4% top-quartile rate, and 16.1% top-decile rate.

The boundary is unchanged: the report found no Benjamini-Yekutieli survivors at q=0.10, so these are evidence-weighted timing hypotheses, not proof claims, gambling advice, or win probabilities. The full public note is on the research blog.

Because classical astrology has been scored by hand for two thousand years, long before anyone had a way to validate it against outcome data. The vast majority of astrology apps and astrologers use exactly this kind of theory-anchored scoring, just without telling you it's theory-anchored.

We're telling you. The score is a structured way to explore classical timing on your own chart. It's an interface to a tradition, with full disclosure that the tradition didn't survive our rigorous empirical test. Use it if that's useful to you. Skip it if you want predictive performance, because nobody, including us, has demonstrated that.

Cohort lists, permutation outputs, reproducibility scripts, and the full retraction live in the public repository. Email research@astraledge.app for the raw JSON dumps or any follow-up.