http://mathoverflow.net/questions/117432/splitting-of-the-weight-filtration (there are also other MO questions on the weight filtration)
In the letter to Beilinson, Voevodsky formulates axioms for a homology theory. He considers as an (n-1)-dim sphere, write also S for the 1-dim sphere in this sense. Let Sch/k be the cat of separated schemes of finite type over a base k. Then a homological theory is a functor from Sch/k together with a family of natural isos . This functor should satisfy some conditions: Morally, homotopy invariance, MV exact triangle, an exact triangle for blowups, and transfer for flat finite morphisms. Get a 2-cat of homological theories over . Examples: Algebraic K-th with rational coeffs, l-adic homology, Hodge homology ass to a complex embedding. Thm: There is an initial object in this cat, which we call the triang cat of eff mixed motives over k. Notion of reduced homological theory, and reduced motive of a scheme. Any motive in the above sense is of the form , where we may assume affine and . Tate object and comparison with K-theory. Bigger cat which contains the previous as a full triang subcat, but admits a more explicit description rather than just the universal property. Can also be viewed as the closure of the previous, wrt direct sums and inductive limits. Need the h-topology, in particular coverings including surjective blowups, finite surjetive maps, etale coverings. Various filtrations on (homotopy canonical, geometrical, motivic canonical, weight). The weight filtr should be related to pure numerical motives.
nLab page on Weight filtration