We are building a validation engine for astronomical signals.
Hello! We are building an engine and pipeline to examine deep space signals with the hopes of finding interesting anomaly. We have just completed Phase 1 - finding AAVSO verified new variable stars and ExoFOP transit candidates.
Verified discoveries from TESS photometry
We built a pipeline in Python that detects and classifies periodic variables (Lomb-Scargle) and transit candidates (Box Least Squares) in TESS photometry. Candidates pass through a multi-stage physical rejection cascade; then a manual review is the final gate.
| Name | Type | Period | Sectors | Record |
|---|---|---|---|---|
| TIC 181428113 | BY+UV | 0.4064 d | 4 | VSX ↗ |
| TIC 8719504 | DSCT | 0.0572 d | 3 | VSX ↗ |
| TIC 49647726 | DSCT | 0.0461 d | 1 | VSX ↗ |
| TIC 458381402 | DSCT | 0.0544 d | 4 | VSX ↗ |
In simple terms - the software finds repeating patterns in how a star's brightness changes, discards everything with a mundane explanation, and a person checks whatever survives before it is submitted to the catalogs. Finding beautiful light curves emitting from deep space is our favourite part of the process!
The learning curve for engineers - vetting candidates
We are not astronomers or astrophysicists - we built this platform because we love data visualisation and deep space fascinates us. Through immersing ourselves in odd/even plots, 12 cycles and phase plots we have gradually learnt what real variable stars look like - and what is just noise from the cosmos.
The exciting next step - Breakthrough Listen's public archive
Breakthrough Listen's archive holds roughly 1 PetaByte (PB) of public data. Nearly all searches to date have targeted narrowband drifting tones. But, research published this year (Garrett 2026, MNRAS) argues this treats broadband signatures as noise. This is interesting and just the type of challenge we love!
So our Phase 2 proposal is to ingest data from the BL archive using Rust. The nuisance model (like our exoplanet vetting) uses ON–OFF cadence differencing. Ranking will combine hypothesis-conditioned scores (lensing echoes, anomalous drift, dispersion deviation) with a physics-agnostic scorer. Each mechanism gets its own injection–recovery benchmark.
In simple terms: instead of searching for one specific kind of radio signal, we're going to score every patch of data by how hard it is to explain, measure how reliably the method recovers test signals we inject ourselves, and publish that measurement.
Collaborate
We would LOVE to find an academic partner to work with us on this - we're engineers not astronomers but we build quickly and test rigorously.
