Recovering Pipeline-Suppressed Stellar Rotation Signals in TESS Using Delta Flux Autoencoders

We present a novel anomaly detection pipeline that recovers stellar variability signals systematically suppressed by the TESS Pre-search Data Conditioning (PDC) pipeline. The central contribution is training a 1D convolutional autoencoder directly on the delta flux. The difference SAP − PDCSAP, which is what the pipeline removed as the primary anomaly detection input channel. Training a reconstruction-based anomaly detector on this channel to identify outliers does not appear in any prior published work on TESS anomaly detection or rotation period recovery. Applied to 33, 135 non-variable stars from TESS Sectors 1–5, the pipeline identifies 88 clean anomaly candidates. Of these, 55 are slow-rotator candidates with periods P > 13. 7 days all absent from any existing rotation catalog and all representing new period measurements. The population splits into a long-period group (n = 25, P > 27 d) and a mid-period group (n = 30, P = 14–27 d), with a median gyrochronological age of 3. 4 Gyr. Multi-sector repeatability validation confirms 57% of the primary anomaly cluster show persistent suppressed signals in independent TESS sectors the model never trained on. Two standout systems are characterised in detail: TIC 167812509 (M dwarf, P = 54. 37 ± 2 d, 100% multi-sector repeatability, confirmed at stellar centroid by per-pixel period analysis) and TIC 278614418 (solar analog, P ≈ 48–60 d, independently consistent with a 2016 spectroscopic vsini measurement). A third system, TIC 261202918, is a GAIA-confirmed binary (Pₒrb = 204. 4 d, 67. 5σ) showing long-period rotational suppression in all 24 observed sectors plus anomalous 15–21% dimming events in 2018–2019 absent from 2020–2025 data, consistent with a transient circumstellar debris feature. Code: https: //github. com/AliBaig-xD/tess-quiet Dataset: https: //doi. org/10. 5281/zenodo. 19638863