A 1.9% blended reserve against a 12% watch list can be defensible, and the test is what the reserve looks like on the risk rated 4 and 5 bucket alone rather than blended across the whole book. Most of these disclose the allowance split between the general reserve and the asset-specific reserve. If the specific reserve on the impaired loans implies a 15% to 20% loss severity on those loans and the rest of the book carries 30 to 50 basis points, that's a coherent position. If the specific reserve implies 4% severity on an office loan the sponsor has stopped funding, it isn't.
The assumption carrying the whole thing is the appraised value behind each rated 5 loan and how old that appraisal is. CECL on collateral dependent loans keys off fair value of collateral, so a stale appraisal on a half-empty office building produces a reserve that's arithmetically correct and economically meaningless. Look for the date of the most recent valuation, and whether it came from a third party.
On the extensions, a paydown plus a sponsor-funded interest reserve is real money and real subordination, and that's a different animal from a maturity extension granted for nothing. Size the paydown against the loan balance. Also check whether the modification was accounted for as a troubled debt restructuring or treated as a normal amendment, because that classification changes whether the loan keeps accruing.
The thing I'd chase harder than the reserve is the financing side. Ask how much of the liability stack is mark to market warehouse or repo versus non-mark-to-market CLO or term debt. A credit book funded on marked facilities can get margin called on downgrades of its own assets, which forces sales into the exact market that caused the downgrade. That mechanism, rather than the reserve level, is what usually turns a slow credit problem into a dividend cut.
