<snip>
Consider the DC to see through a hag's disguise using Insight. This is set high enough (generally, Hard DC of the hag's level +5) to provide a real challenge to task resolution. Which is fine. But won't work if incorporated into a skill challenge (too high! it will muck up the maths).
Bluff DCs are similar - work well for opposed checks in task resolution, too high for skill challenges.
The solution that I use is to use skill challenge appropriate DCs, and require multiple successful checks before the full details (of the lie, of the hag's identity, etc) come out in the course of resolution.
But I can see how this sort of thing, which the rules don't really deal with at all, can cause some confusion and/or dislike of the system.
I have never run a social Skill Challenge with a hag, but I think that is a classic trope that would be well serviced by the Skill Challenge resolution system. If I ever do run one, I would assuredly do the same thing as you describe as it is standard operating procedure for a Skill Challenge.
I have run a ruthless, Jack Bowersesque, interrogation Skill Challenge whereby two PCs attempt to secure mandatory information regarding an imminent terrorist plot (poisoning the nobility district's water supply with a deadly toxin) from an otherwise unwilling target. Naturally Bluff and Insight were heavily involved and I used Skill Challenge appropriate DCs to resolve it. Its a very intuitive process once you understand the (relatively) outcome-based simulation of Skill Challenges versus the (relatively) process-based simulation of mundane task resolution.
Its very relevant to Hussar's topic. Although the 4e rules texts are quite thorough in many respects and, as a whole, the product is exceedingly coherent, there are "holes in its game" whereby the authors and editors did not conceptualize how their lack of bridging of logic (just expecting it to be intuitive and therefore no need to be made explicit) would affect the polish of their final product. In editions predating 3e, it was expected that players would infill the authors/editors lack of explicit bridging of logic with their conceptualized understanding of the rules texts (the marriage of RAI with RAW). However, we no longer grok this responsibility as the mental fog of a subtle paradigm shift (which Hussar outlines in his initial post) - "fidelity to the orthodox of the anointed texts" - has become cultural mainstay over the course of the last two editions. That being said, I hope I've made it abundantly clear at this point that I favor coherent, transparent, thorough rules texts over incoherent, relatively opaque, shallow rules texts. However, what I do hold as true is that if rules texts conveyed a schematic of each subtle (intuitive?) "logical transition" (Skill Challenge metagame, level-appropriate, DC design over-rules mundane, non-level-centric, task resolution DC standards), then word count and page count would increase dramatically...and I suspect that rules texts would start reading like my posts with a thousand and one caveats and clarifiers embedded within (driving the gaming community at-large to madness).
With that caveat on top of another caveat...I have one more caveat. I do have issues with the coherency of some of the designers' rules text (and its likely, poor manipulation by its editors). I've stated as such in a few posts here and there. If there is one thing I hope for in the next edition (if I choose to partake) is that the designers' intentions are coherent/clear and that, after editing, they are allowed to go back through the books again to make sure their message (to make sure important "logical transitions" are still well-represented, and are not outright omitted, in the edited version), was not pruned too much to the point that the rules text move from coherent to relatively incoherent (with respect to their iteration prior to editing). Too many cooks in the kitchen spoil the soup. If players must bear to much of the burden of conceptualization of RAI from RAW, if long term gamers end up with too many exchanges over RAW, it means that RAI (logical transitions) was not clear enough in concept.