Arbitrage free implies complete market in general binomial model?

In Tomas Björk's Arbitrage Theory in Continuous Time, there exists this proposition

It seems that to show that the model is complete, we must show that the claims are reachable, i.e. we must find replicating portfolios for each claim.

Which part of finding the replicating portfolio makes use of the assumption, where I understand the assumption is equivalent to $$d < 1+R < u$$ (or $$d \le 1+R \le u$$, but $$d)? All I see so far is the $$d, but that's nothing really special: of course the stock price $$u$$ if it would go up from its current price should be higher than the stock price $$d$$ if it would go down from the its current price.

$$\begin{bmatrix} 1+R & 1+R \\ d & u \end{bmatrix}$$
is one-to-one, i.e. $$d \neq u$$. (Equivalently, its transpose is onto, which is what is shown in your quoted text.)
No-arbitrage holds if and only if $$d \leq 1+R \leq u$$ and $$d < u$$. (By assumption, $$d \neq u$$. Otherwise you have two risk-free assets, and model either trivially collapes or trivially admits arbitrage.) Therefore NA implies market completeness, because $$d < u$$ implies $$d \neq u$$. Note that the no-arbitrage condition---that the risk-free return lies in the interval of risky return---is not really used.
Leaving the binomial setting, this need not be true. For example, consider a model with $$n$$ states and $$n$$ assets, with $$n > 2$$. The model could have no-arbitrage but not be complete. Similarly, consider a model with $$n$$ states and $$m$$ assets, with $$m > n > 2$$. The model could be complete (in fact, with redundant assets) but admits arbitrage.