ab astris
a cross-domain method for separating physics-constrained signals from behavioural noise.
Ab Astris began as an evening practice in astronomy and turned, slowly, into a general claim about how periodic signals behave. The pipeline began as a way to look for variable stars in NASA TESS data; it now serves as a small, portable instrument that tells you whether the periodicity you have found is the kind that obeys physics, or the kind that does not.
Public sky surveys produce more candidate signals than the community can responsibly review. Most are spurious: instrument artefacts, stellar variability misread, eclipsing binaries mistaken for planets. The bottleneck has never been data; it has been the labour of looking carefully.
The methodology grew from a question that seemed local at first and then refused to stay there: what distinguishes a real periodic signal from a convincing accident? Six domains in, the answer is a single quiet statistic.
The pipeline combines a Lomb-Scargle periodogram with multi-window validation and bootstrap error estimation. Confidence is scored from frequency stability, signal significance, multi-window agreement, and detection rate. Anything that passes is registered as a candidate for community follow-up.
The result of running the same pipeline across six domains is a working ladder of stability, and a clean separation between systems whose periodicity is enforced by physics and systems whose periodicity is not.
Across six scientific domains, physics-constrained periodic signals sit below 15% coefficient of variation; behavioural and stochastic signals sit above 40%. Bootstrap confidence intervals on fourteen domain means support the gap as a methodological (not merely empirical) fact.
The pipeline has been validated against bearing-fault data (CWRU), tidal constituents at six NOAA stations, volcanic tremor across five volcanoes, civil-structure resonance, and the Z24 bridge progressive damage dataset. Negative controls (cryptocurrency, sunspots, heart-rate variability) sit clearly outside the physics-constrained band.
The work is ongoing. New survey releases are incorporated as they appear; the cross-domain validation register is updated quarterly. We accept correspondence from observers willing to take follow-up time on the shortlist, from engineering teams whose periodic data we have not yet seen, and from students looking for a first publishable reduction.