Your rebuttal raises several questions. You say "campaigns are about arguments and issues," which candidates "road-test" in their polls. But as you know, the "arguments" presented in campaigns are seldom purely logical, and the "issues" are generated as often by the candidates as by the voters.
Take the "arguments" Bush road-tested in his South Carolina poll. One is that McCain "proposed the largest tax increase in United States history." This description makes both a significant assumption and a significant omission. The bill required tobacco companies to pay the government $500 billion to settle various lawsuits, ostensibly aimed at recouping subsidized medical expenses caused by smoking. Bush's poll question assumes that this payment counts as a "tax" because the companies would have to pass the cost of the settlement along to everyone who buys cigarettes. The question also omits the fact that this "tax" applies only to tobacco (it could have been called a "sin tax," which would have changed the poll numbers considerably) and that smokers can avoid it (at least in principle) by refusing to buy cigarettes at the marked-up price. Many "arguments" in campaigns, if not most, rely on this kind of omission and assumption. They involve at least as much characterization ("spin") as logic.
The same is true of campaign "issues." Yes, sometimes they spring up from the grassroots, as health care and term limits arguably did. But often, they are invented by candidates or by their advisers. We all remember the ACLU and flag "issues" of 1988. My point is that "arguments and issues" aren't objectively valid or important. They're always coated with spin, often shaped by spin, and sometimes spun out of trivial matters such as an offhand comment or an arcane vote in Congress. Saying that campaign polls road-test "arguments and issues" is just a kinder way of saying that they road-test spins.
You also point out that push questions are "widely accepted tools of survey research to determine what works." But why should they be so accepted? Should the fact that push questions are widely accepted among political consultants and journalists reflect well on push questions? Or should it reflect poorly on political consultants and journalists?
In questioning our acceptance of push questions, I don't mean to suggest that they should be banned or declared immoral. As we both know, that would be idiotic, impossible, and incoherent. What I do mean to suggest is that we shouldn't characterize push questions as fair-minded science. This is my objection to Stu Rothenberg's article in Roll Call, which suggests that campaign polls "seek to measure public opinion" rather than "to alter opinion." This distinction is way too simple. Here's how I would amend it, and perhaps you and Stu would agree with me. Push polls seek to alter public opinion. Media polls, on the other hand, generally seek to measure public opinion. Campaign polls, however, do both. By confronting respondents with push questions, they measure the alteration of public opinion.
I agree with you and Stu that push polls are uniquely dishonest in terms of the relationship between caller and respondent, because the caller is lying to the respondent when he says he's taking a poll. But I don't agree that push polls are necessarily worse than push questions in real polls in terms of the content of the questions. In both cases, the question is loaded. In a real campaign poll, the question is loaded for precisely the reason you cite: The campaign is trying "to determine what works." And what works is what you call an issue or an argument, which could just as easily be called an attack or a spin.
You're right that campaign polls often test lines of attack that could be delivered by the opposition as well as those that could be delivered by one's own campaign. And by extension, campaign polls, like media polls, routinely ask elementary and fair-minded questions, such as whether you have a favorable or unfavorable impression of John McCain. But campaign polls, unlike media polls, are instruments for figuring out how to win elections, not how to observe them. And for that reason, many of their questions are designed not to learn what's on your mind but to determine how you will respond to a few narrow lines of attack already selected and crafted by the candidates. The unspoken motto of the campaign polling industry is: Ask not how your candidate can respond to you, but how you will respond to your candidate.