Bought NLY in January, held through both cuts, and the second cut was actually the easier decision

What is your current book value discount or premium, because that single number tells you whether the market has already priced in the spread risk you now understand?

The reframe you landed on is the right one. Dividend yield on an agency mREIT is an output of the spread, the hedge book, and the funding structure. Treating it as an input to the thesis is where most people start and where they stay until something forces the correction.

The assumption doing the most work in your original position was that payout consistency was a signal about future spread conditions rather than a record of past ones. That is a category error that is easy to make because the yield is visible and the repo ladder is not.

One thing you have not named: the basis risk inside the hedge book. A short hedge book protects against parallel rate moves, but agency mortgage spreads can widen against Treasuries even when rates are flat or falling. If the hedges are indexed to rates and the P&L hit came partly from spread widening rather than rate movement, the hedge did less work than its notional size suggests it should have. Worth confirming which spread widening episodes in your holding period were rate-driven versus basis-driven, because those require different hedge structures and the 10-K will show you the notional but not always the decomposition.

The risk you did not mention is prepayment optionality on the asset side. Agency pools in a falling rate environment accelerate. That compresses the asset yield just as repo costs may be resetting, which is a different spread compression mechanism from the funding side. If NLY's specified pool composition shifted in either holding period, the reported yield tells you less than you think.

You revised the mental model under pressure, which is the harder version of the lesson. Most people revise the position instead.

What does the current hedge duration look like relative to the asset duration, and have you stress-tested the basis risk separately from the rate risk?