Uh, no. This only follows logically if B lies all the time.

There's no "all the time" in the example, only person U who assumes that B lies on TV. Your ojbection has merit of course, but only if you extend the scope of the original example hugely.

It turned out in retrospect that he lied maybe 95% of the time. But that fact wasn't available in 2003.

Actually, it was pretty obvious at the time that he lied, which is why the whole thing was such a hard sell. B's statements were repeatedly contradicted by the IAEA investigators on the ground in UN reports, his claims were not confirmed by any major powers (France, Russia, China) except for the US, his dossier was immediately shown to be lifted (and edited) from internet sources, his claims agreed with Powell's UN lies, which were also contradicted by the IAEA at the time, and of course most of his claims were contradicted by the Iraqi government at the time.

The only way anybody at the time could claim that it wasn't obvious that the UK government was lying was by weighting the USUK statements 99%, and all other statements 1%, say. That's actually quite reasonable for British people in general to do, on the grounds that they'd have to become paranoid otherwise, but non-anglophones had no such conflict of interest.

Which nicely again illustrates my point about unstated assumptions leading to divergence. Anybody who placed even 50% weight on USUK statements and 50% weight on statements from other sources essentially had to consider B a liar, simply due to the large number of contradicting claims of fact by other independent sources.

There's a vast uncrossable gulf between that kind of narrative logic, in which anything goes as long as it sounds vaguely plausible, and evidence-based argument, which requires a decent data set to argue implications from.

You seem to think that evidence-based argument (as opposed to calling something narrative logic?) requires a specific set of common initial assumptions (such as the famous I think therefore I am, which alone is obviously insufficient). Please list them.

--
$E(X_t|F_s) = X_s,\quad t > s$