Stars is a testbed for one question: where do independent symbolic systems, built from the same sky, land on the same conclusion? The dataset, the raw scores, and the analytics are open to work against — including the places where an earlier claim had to be narrowed.
the whole engine, behind every category
93 traditional D0–D2 systems converge at cosine 0.515–0.523 across 500 charts (511 comparable pairs), engine hypotheses excluded. Engine 2.7.0. Verified 2026-09-08.
AA-rated timed charts with full convergence scores, as CSV or Parquet. Cite "Stars Engine Research Dataset" in publications.
Pairwise system-to-system cosine values, per dimension, per chart — not just the headline range. Bring your own aggregation.
DuckDB over Parquet: 45-feature registry, correlation queries, template detection, tier-decomposed convergence (D0 / D1 / D2).
Measurement roots and transmission lineages carried as a pair, with the sourced culture-contact relations that justify calling two systems independent.
The actual scripts behind the published figures —
engine/validation/bottom_up_convergence.py, the tier-decomposed
and provenance-filtered variants, the reach computation — with docs. Same
data, same code, same result, or tell us what differed.
5,132 nodes with derivation types, mechanisms, and source tiers — co-occurrence, doctrinal, empirical, convergence edges, each annotated with how it was established.
Every interpretive claim carries a citation to a held primary text — Agrippa for ceremonial, Parashara for Vedic, Ptolemy for Western. Where we hold no source, we say nothing. 113 of 183 claims are warranted today; the rest are marked unwarranted, not dressed up.
We never collapse corroboration to one number. Measurement independence asks how many distinct observables agree; transmission independence asks how many uninfluenced lineages arrived there. Ten systems reading one ecliptic longitude are one measurement, not ten — and we render both axes.
Findings are provenance-audited before they ship. The convergence claim was narrowed from an early over-broad figure to 93 traditional D0–D2 systems at cosine 0.515–0.523 once engine hypotheses were excluded. The correction is on the record.
No advertisers, no data brokers, no training on client charts. Chart computation runs on owned hardware in Texas. Export or delete your data at any time.
An earlier internal run included fabricated / engine-hypothesis scores in the "converging" set and reported a broader figure. The provenance audit caught it, the scores were removed, the claim was narrowed to the 93 traditional D0–D2 systems you see now, and the audit trail stayed in version control. That episode is the strongest evidence we can offer that the pipeline polices itself.
The claim is deliberately conservative and the data is open so it can be attacked. If the convergence signal is an artefact, we want to know. Studio tier carries batch compute and the full raw-score export; researchers with a replication or falsification plan get access at no cost.
Invite-only beta. Studio tier $199/mo; waived for a credible replication or falsification plan.