A Five-Stage Framework for Slitless Grism Spectra: Demonstrated on Euclid Q1 Strong-Lens Candidates and Ported to CEERS NIRCam

Abstract

Wide-field slitless grism spectroscopy is difficult in the co-spatial regime, where multiple sources share one cross-dispersion PSF element and catalogue-level decontamination (aXe/LINEAR/Grizli) does not apply -- the regime of strong lenses, binary AGN, and mergers in HST, JWST, Euclid, and Roman grism data. This paper presents a five-stage framework for per-source 1-D spectra and redshifts: distortion-calibrated extraction, Fourier/Sersic/NNLS deblending, robust-PCA cleaning, weighted friend-of-friend with Hungarian cross-orientation matching, and a Bayesian posterior layer. Of 579 published Euclid Q1 lens candidates, 473 fall in the NISP footprint. A 3" DESI DR1 Iron match gives 57 counterparts (51 ZWARN=0, 46 clean v6d anchors). Extending to 9200 A adds features below 12500 A in 312 source and 324 deflector spectra. Source/deflector features are detected in 467/471 of 473; 371 source redshifts are secure, 75 probable, 21 tentative. Adopted z_def is filled for 440/473 (55 DESI + 385 PHZ MODE_1). Catalogue: 342 complete, 29 source-only, 75 Silver, 21 Bronze, 6 none. Posteriors succeed for 398/473; DESI deflector coverage is 0.67+/-0.07 (46 anchors). Tiers platinum/gold/silver/bronze/none = 62/230/82/24/75. Median source 68% CI is dz=1.30; DESI |dz|<0.1 recovery is 35%. Source-side support is synthetic-only; the full deblender's real-data performance is unmeasured here. A ported v6c on 10 CEERS DR0.7 JWST/NIRCam targets improves median |dz|/(1+z) from 0.43 to 0.18, recovers 4/10 within 0.10, and contains truth in the 68% CI for 8/10 -- portability across instruments, not independent calibration. The release is a 473-row catalogue with deflector redshifts, posterior intervals, and quality tiers.