In my comment to my first answer, I noted that I didn't address Greg's question (1). This answer aims to address that question. I am indebted to Bill Richter for explaining this argument to me. We are trying to understand whether or not the square $$ \require{AMScd} \begin{CD} O(n+1) @>>> S^n \\ @VVV @VVV \\ F_{n+1} @>> H > \Omega^{n+1} S^{2n+1} \end{CD} $$ commutes. I mentioned in my comment that James proved it commutes after a loop. I will sketch a proof of a more general result. Let $\lambda = \Sigma H$ denote the Boardman and Steer Hopf invariant. This is a natural transformation of based function spaces $$ F_\ast(A,B) \to F_\ast(\Sigma^2 A,\Sigma B\wedge \Sigma B) $$ given by suspending the James-Hopf invariant. Let's consider the related problem of whether or not the diagram $$ \begin{CD} G_{n+1} @> \pi >> S^n \\ @V\cap VV @VV E V \\ F_{n+1} @>>\lambda > \Omega^{n+2} S^{2n+2} \end{CD} $$ commutes. If it does then Greg's question (1) will commute if we replace $H$ by $\lambda$, since the map $O(n+1) \to S^n$ factors through $G_{n+1}$. **Claim:** The last diagram above homotopy commutes after taking a single loop. Reformulate the claim as follows: let $B = \Sigma A$ be a suspension and let $B\to G_{n+1}$ be any map. Then the claim is equivalent to the assertion that the diagram $$ \begin{CD} B @> \pi >> S^n \\ @V\cap VV @VV E V \\ F_{n+1} @>>\lambda > \Omega^{n+2} S^{2n+2} \end{CD} $$ commutes (by taking $B = \Sigma \Omega G_{n+1}$). To verify the reformulation of the claim, we will use the Boardman and Steer Cartan formula. Let $X = S^n$, then the action of $G_{n+1}$ on $X$ induces a map $$ F: B \times X \to X\, . $$ The Hopf construction of $F$ is a map $h_F: \Sigma B\wedge X \to \Sigma X$ whose adjoint is the left vertical (inclusion) map of the square. Thus we need to compute $\lambda(h_F): \Sigma B \wedge \Sigma X \to \Sigma X \wedge \Sigma X$. Take the suspension of this map $\Sigma F: \Sigma (B \times X) \to \Sigma X$. Then we have a factorization up to homotopy $$ \begin{CD} \Sigma (B\times X) @>\Sigma F >> \Sigma X \\ @V p_1 + p_2 + p_{12} VV @| \\ \Sigma B \vee \Sigma X \vee \Sigma B\wedge X @>>(f,g,h_F) > \Sigma X \end{CD} $$ where $p_1$ is the projection onto $\Sigma B$, $p_2$ the projection onto $\Sigma X$ and $p_{12}$ is the quotient map. The map $f$ is the suspension of the restriction of $F$ to $B \times \ast$, the map $g$ is the suspension of the restriction of $F$ to $\ast \times X$. Since $\Sigma F$ is a suspension, its Hopf invariant $\lambda(\Sigma F)$ is trivial. So, if we take the Hopf invariant of the composite $$ \Sigma F = (f,g,h_F) \circ (p_1 + p_2 + p_{12}) $$ and use the composition formula and the cartan formula, we obtain, after some rewriting, the formula $$ \lambda(h_F) = (f \wedge g) \circ \Sigma p_{12} $$ Implicit in this calculation, which I've omitted, is the fact that the reduced diagonal maps of both $B$ and $X$ are trivial since these spaces are suspensions. I refer the reader to Boardman and Steer's paper for the details. Returning to our original notation, note that the map $f$ is the suspension of composition $$ \begin{CD} B @>>> G_{n+1} @>\pi >> S^n \end{CD} $$ and the map $\lambda(h_F)$ is adjoint to the composite $$ \begin{CD} B\to G_{n+1} \to F_{n+1} @>\lambda >> \Omega^{n+2} S^{2n+2}\, , \end{CD} $$ whereas the map $(f \wedge g) \circ \Sigma p_{12}$ is the composite $$ \begin{CD} \Sigma^2 (B \times S^n) @>>> \Sigma^2 (G_{n+1} \times S^n) @>q >> \Sigma^2 G_{n+1} \wedge S^n @>\Sigma \pi \wedge 1 >> \Sigma S^n \wedge \Sigma S^n \end{CD} $$ where $q$ is the quotient map. By definition last composition is adjoint to the composite $$ \begin{CD} B @>>> G_{n+1} @> \pi >> S^n @> E >>> \Omega^{n+2} S^{2n+2} \end{CD} $$ and the claim is proved! **Comment:** The above argument shows that the diagram of Greg's question (1) commutes after looping. The equation we derived $$ \lambda(h_F) = (f \wedge g) \circ \Sigma p_{12} $$ isn't valid when $B$ isn't a suspension. In this instance there is a correction term in the formula involving the reduced diagonal $B\to B \wedge B$. This gives evidence that Greg's diagram will fail to commute before taking loops.