> A push poll doesn’t provide information to authors. They don’t care about the results.

That's not necessarily true, there are polls in the middle, where the authors attempt to both push agenda, and also collect information on whether the push is working. I'm saying that any amount of pushing constitutes a push poll; the defining characteristic is not whether the authors care about the results, but whether they're attempting to influence the results.

Examples of what I'm talking about are even listed in the Wikipedia article on push polling: https://en.wikipedia.org/wiki/Push_poll

If a poll is really asking whether a certain line of negative attack is going to work better than another, and not attempting to influence the answer, not attempting to sway the respondent, then yes that would be an "honest" survey in some sense. However, there aren't actually many polls related to negative campaigns that do this, and it's also very difficult to do without influencing the respondent. Negative attack campaigns are rarely if ever interesting in doing careful neutral statistics. You'd have to be scientific about it and include controls as well as positive and neutral statements in addition to the negative ones.

So, I still don't buy the original claim that most polls on negative lines of attack aren't really push polls. They are push polls, but they might indeed be softer pushes than other more extreme push polls.

The difference, as so often, is scale. If you poll 500 people in a congressional district and ask them what their reaction to Candidate X's drinking problem is, that's an information gathering poll, not one intended to sway a significant number of voters. It might not be "scientific", but it is intended to gather information about broader voter sentiment to guide your campaign strategy lines, because even changing 500 voters minds isn't going to have much effect on the race. If you did asked the same questions, and even made the same decisions based on the data, but did it to 100,000 voters then it does start to look more like a hybrid or closer to the push-poll side of things.

You're talking about the attack, not the poll.

I disagree strongly that scale has anything to do with it whether you call a poll a push poll. Scale is irrelevant. Asking what you think about a candidate's drinking problem is a push intended to communicate the idea that the candidate has a drinking problem, without regard to context and without discussion. You're right that it's not scientific. You're wrong that it's neutral information gathering. It doesn't matter how many people you ask that question. You can ask the same question to a single person and it's still a push, or more commonly called, gossip or shade.

Your mistake here is assuming that a push poll needs to be influencing a "significant" number of voters. It does not. Again, the defining characteristic of a push poll is the push. Not the number of people surveyed, not whether the pollers are collecting results, not whether they're looking for an angle of attack versus attacking. If the poll is pushing an agenda, if it's trying to cast doubt on a candidate, or trying to "inform" them of something bad or good, then it's a push poll, and I think the WP page made this pretty clear.

Several small scale push polls done in search of or in preparation for a large scale attack are exactly the kind of poll that count as both push and poll. Your example is demonstrating exactly the case where the pollers care about the result, and they're pushing at the same time. The intent is to influence the voter, and also measure how influenced the voter was. That's more of less the definition of a push poll.