I am trying to understand the statement of the conjectures of Deligne on special values of certain $L$-functions, from his article titled, "Valuers de Fonctions L et periodes d'integrales" which appeared in Proc. of Symposia in Pure Math., 33, (1979), Part 2, 313-346.

In section 1 he makes the conjectures for a motive $M$ defined over $\mathbb{Q}$. He begins by assuming that the $\ell$-adic realizations of $M$, $H_l(M)$ form a strictly compatible system of $\ell$-adic representations. The fact that he assumes this is the reason I am asking this question, since I would have thought that this would always be the case.

If I understand correctly, the assumption that, the $\ell$-adic realizations of $M$, $H_l(M)$ form a strictly compatible system of $\ell$-adic representations, means the following: There is a finite set of primes in $\mathbb{Q}$, say $S$, such that

- for each $p\notin S$ and $\ell\neq p$ the representation $\rho_\ell:G_\mathbb{Q}\to GL(H_{\ell}(M))$ is unramified at $p$,
- Because of (1) we can consider the characteristic polynomial of the geometric Frobenius. This polynomial should have coefficients in $\mathbb{Q}$ (apriori, this is a polynomial with coefficients in $\mathbb{Q}_\ell$),
- For $p\notin S$, the polynomial we get in (2) should be independent of $\ell$.

Now suppose $M$ were the motive $h^i(A)=(A,\pi_i,0)$, where $A$ is an abelian variety defined over $\mathbb{Q}$, and the $\ell$-adic realization of $h^i(A)$ is $H^i_\ell(A;\mathbb{Q}_\ell)$. Let $S$ be the set of primes where $A$ does not have good reduction. Then the $\ell$-adic realizations of $M$ form a strictly compatible system of $\ell$-adic representations (condition (1) holds since good reduction implies unramified, condition (2) and (3) would hold because of the Weil conjectures, I hope I'm correct). I would have thought that for pure motives (as in Scholl's article titled, "Classical motives") the same conclusion would hold. Is this true?

Having written the above question I realize that Deligne uses the notion of absolute Hodge cycles to define motives in this article. Perhaps this is what is causing the problem. Can someone shed some light on this matter. Thanks in advance.

**EDIT**: From pondering over the comment by Eric and the answer by Will, I realize that it is not clear to me why the Frobenius should act on the cohomology $H_\ell(M)$. Let $M=(X,\pi,0)$ be a pure motive as in Scholl's article, and for simplicity let us also assume that it is homogeneous; I mean that the Hodge realization has only one weight. The situation I have in mind is one in which $M$ cuts out a piece from the cohomology $H^i_\ell(\bar{X},\mathbb{Q}_\ell)$. If $p$ is a prime at which $X$ has good reduction then the Galois representation $H^i_\ell(\bar{X},\mathbb{Q}_\ell)$ is unramified and so we can choose 'a' Frobenius endomorphism
$$Frob_p:H^i_\ell(\bar{X},\mathbb{Q}_\ell)\to H^i_\ell(\bar{X},\mathbb{Q}_\ell)$$
If $H^i_\ell(M)$ denotes the image of the action of the projector $\pi$ on the cohomology, then I see no reason why our choice of $Frob_p$ should preserve this subspace. Is this clear? Yes it is, as pointed out by Will in the comments to his answer.