Skip to content

Scientific Background

Transit photometry

A transiting companion blocks a small fraction of its host star's light when its orbit crosses the line of sight. Repeated, approximately periodic dips can therefore reveal an orbital period, transit duration, epoch, and depth. The signal is difficult to recover when stellar variability, instrument systematics, gaps, dilution, eclipsing binaries, or low signal-to-noise produce similar structure.

SXS operates on public Kepler long-cadence light curves. It is a computational recovery and review system; it does not produce spectroscopy, high-resolution imaging, or new telescope observations.

Why a pipeline needs multiple layers

No single score is sufficient for a discovery claim:

  1. Preprocessing can suppress noise but may also distort a transit.
  2. BLS finds box-like periodic signals but also responds to binaries, systematics, and stellar variability.
  3. Machine learning can prioritize familiar signal patterns but inherits the training distribution and decision threshold.
  4. Independent vetting can reject many failure modes, but missing evidence and observational limits remain.

SXS therefore separates detection, ranking, and independent audit. The RF/CNN scores are excluded from the final validation rules.

Research questions

The research record addresses four bounded questions:

  1. How often does the configured BLS search recover known Kepler planets in its five highest distinct peaks?
  2. Can a target-grouped feature model reduce candidate-level false passes while retaining most BLS-recovered positives?
  3. What happens when the frozen model screens a deterministic workstation-scale target sample with no selected catalog history?
  4. Do the shortlisted signals survive independent statistical, photometric, catalog, Gaia, and TESS checks?

It does not estimate exoplanet occurrence rates or claim a complete Kepler search.

Ground truth and external evidence

  • confirmed-planet parameters: NASA Exoplanet Archive pscomppars;
  • Kepler false-positive labels and flags: cumulative KOI table;
  • time-series products: MAST, acquired with Lightkurve;
  • scene context: Gaia DR3;
  • independent photometric comparison: public TESS products where available;
  • public candidate history: ExoFOP-derived TOI lookup.

Each upstream source retains its own terms and citation requirements. The citation guide lists the project-level expectation.