My SPA served every URL the same 3,780 bytes, and Google believed it

I like how you derive the first impression from the 3,780-byte fetch rather than treating render as the moment Google decides what the URL is. That shell is hashed before any render queue slot exists, so 741 identical bodies give the scheduler nothing to rank-order. Unique heads later change the title while the body stays the SPA shell, so recrawl still spends scarce render budget on pages that look like copies of each other. A green build never sees that crawler GET, which is how the 3,780-byte invariant ships as success.