The dual space and its basis
The dual space $V^* = \operatorname{Hom}(V, R)$ is the module of linear functionals on $V$ — its elements eat a vector and return a scalar. If $V$ is finite-dimensional with basis $(e_1, \ldots, e_n)$, the dual basis $(e^1, \ldots, e^n)$ is defined by $e^i(e_j) = \delta^i_j$ (the Kronecker delta: $1$ if $i = j$, else $0$). So $e^i$ ‘reads off the $i$-th coordinate.’ Then $\dim V^* = \dim V$, and there is a canonical isomorphism $V \cong V^{}$ — but not** a canonical $V \cong V^*$ without extra structure such as a bilinear form.
Tensors of type $(p,q)$ and contraction
A tensor of type $(p,q)$ on $V$ is an element of $\;\underbrace{V \otimes \cdots \otimes V}_{p} \otimes \underbrace{V^* \otimes \cdots \otimes V^*}_{q}$ — $p$ contravariant (upper) slots and $q$ covariant (lower) slots. Type $(1,0)$ is a vector, $(0,1)$ a covector, $(1,1)$ a linear endomorphism (a matrix), and $(0,2)$ a bilinear form. Contraction is the linear map that pairs one upper slot with one lower slot via the evaluation $V \otimes V^* \to R$, $v \otimes \alpha \mapsto \alpha(v)$, dropping the type from $(p,q)$ to $(p-1, q-1)$.
Contraction = trace, made concrete. A type-(1,1) tensor is
T = sum_{i,j} T^i_j (e_i (x) e^j) in V (x) V^*.
Contracting the upper slot against the lower slot:
C(T) = sum_{i,j} T^i_j . e^j(e_i)
= sum_{i,j} T^i_j . delta^j_i
= sum_i T^i_i = trace(T). <- the diagonal sum!
So 'contract the two indices of a (1,1) tensor' IS 'take the trace.'
Basis-independence of trace is now obvious: contraction was defined
without reference to any basis (it is the canonical map V (x) V* -> R).
Einstein summation, decoded. Writing alpha_i v^i with an implied sum
over the repeated index i is exactly the contraction
V* (x) V -> R, alpha (x) v |-> alpha(v) = sum_i alpha_i v^i.
A repeated upper/lower index pair always means 'contract here.'What you can now read
With dual basis, type, and contraction in hand, index notation stops being mysterious. A matrix product $C^i_k = \sum_j A^i_j B^j_k$ is contraction of a $(1,1)$ with a $(1,1)$ along one index pair; the transpose swaps which slot is which; the determinant is the totally antisymmetric contraction $\det A = \frac{1}{n!}\,\varepsilon_{i_1\cdots i_n}\varepsilon^{j_1\cdots j_n} A^{i_1}_{j_1}\cdots A^{i_n}_{j_n}$ — exactly the top-exterior-power story of guide 4, written in coordinates.