Online search updated 2026-09-22 (the route's 2026-09-18 search is reused; one new query on the cyclic k-deck / necklace reconstruction). Owning convention, confirmed: sequence reconstruction from the k-deck (the multiset of length-k subsequences). A. D. Scott, "Reconstructing sequences", Discrete Math. 175 (1997) 231-238 (author PDF people.maths.ox.ac.uk/scott/Papers/recseq.pdf): every sequence of length n is reconstructible from its k-deck for k >= (1+o(1)) sqrt(n log n); the paper also poses the CYCLIC version (a necklace from the multiset of necklaces obtained by deleting n-k beads) and, per the summary read this turn, records that no good bounds are known for it; Gabrys, "The hybrid k-deck problem", arXiv:1701.08111 (ISIT 2017), the minimal-k formulation with short and long traces; Chrisnata et al., "On the number of distinct k-decks" (Adv. Math. Commun. 2023). None of these treats the SUMMED k-deck (window sums of a cyclic word over the integers) or a map (the Holt-Rudd fold) that has to factor through a summary; #986's search already recorded this exact gap, and it stands. The two summaries tested here are therefore: the route's own summed k-deck S_k (value-only window sums, size D per k, rotation- and reversal-invariant) and the cyclic k-deck of subsequences taken over all rotations (Scott's cyclic object, encoded on the 4-letter alphabet, size 4^k). For D = 15 Scott's bound sqrt(15 log 15) = 6.4 suggests the full k-deck separates from about k = 7 for linear sequences; the cyclic version is what is measured here at k = 2, 3, 4 (larger k exceeds the budget: 15 rotations x C(15,k) subsequences per word over 90,090 words). Project sources inspected: route 71 rev 2 (the next_step (A)-(C), the uncertainty on the existence half); #986 (job 1861: fold controls 4/4, S_k dihedral invariance, T_5 class of size 1, the T_7 spectrum at k = 2, 3, 4 with 60-pair fold tests, the length-7 refuting pair, job1861-checks.py reused for the fold convention: positions from 0, canonical rotation); #982 (the reviewer's word-fold re-implemented, the mirror symmetry of the tile's word); review #71 of #23 (the premise correction); OUTCOMES.md's chaining closure row (the object of the re-scoping). Exact remaining gap after this run: (1) the k-spectrum and the cyclic k-deck are measured at D = 15 only; the class at D = 135 (T_11) has about 1e80 arrangements and no exact enumeration is possible, so any statement about k*(D) growing with D needs a sampling frame (the route's (C)) or a proof; (2) a summary family outside decks (e.g. the fold's own output restricted to a residue class, or Fourier data of the word at the fold prime) is not tested; (3) the fold's frame dependence on synthetic words (a rotation of the start changes the folded word for some words, measured below) means "factoring through a rotation-invariant summary" is only well-posed with a fixed frame convention; the canonical-rotation convention of #986 is used and disclosed, and the true tile carries its own frame.