A photo-z cautionary tale: Redshift confirmation of COSBO-7 at z=2.625

Photometric redshifts are widely used in studies of dusty star-forming galaxies (DSFGs), but catastrophic photo-$z$ failure can undermine all redshift-dependent results. Here we report the spectroscopic redshift confirmation of COSBO-7, a strongly lensed DSFG in the COSMOS-PRIMER field. Recently, a photometric redshift solution of $z\gtrsim7.0$ was reported for COSBO-7 based on ten bands of {\it James Webb} Space Telescope (JWST) NIRCam and MIRI imaging data. This $z$ value was favored by four independent spectral energy distribution (SED) fitting codes, and the result provided an appealing candidate for the most distant massive DSFG known to date. This photo-$z$ solution was also supported by a single line detection in Atacama Large Millimeter Array (ALMA) Band 3 consistent with CO(7-6) at $z=7.46$. However, our new ALMA observations robustly detect two lines in Band 6 identified as CO(7-6) and [CI](2-1) at $z_{\rm spec}=2.625$, and thus the Band 3 line as CO(3-2). These three robust line detections decidedly place COSBO-7 at $z=2.625$, refuting the photo-$z$ solution. We derive physical parameters by fitting near-infrared(NIR)-to-millimeter(mm) photometry and lens modeling, revealing that COSBO-7 is a main sequence galaxy. We examine possible reasons for this photo-$z$ failure and attribute it to (1) the likely underestimation of photometric uncertainties at 0.9\,$\mu$m and 1.15 \,$\mu$m; and (2) the lack of photometry at wavelengths beyond 20\,$\mu$m. Notably, we recover a bona fide $z_{\rm phot}\sim 2.3$ by including the existing MIPS $24\,\mu$m photometry, demonstrating the critical importance of mid-infrared (MIR) data in bolstering photo-$z$ measurements. This work highlights a common challenge in modeling the SEDs of DSFGs, and provides a cautionary tale regarding the reliability of photometric redshifts as well as pseudo-spectroscopic redshifts based on single line detection.