[mooism2]

These attacks are probably automated, and it would be trivial for the spamming software to be updated to get round these restrictions. I don't think it would take long.

The cost would be that HN would be less welcoming to newcomers.

That's a permanent loss for temporary gain, and I don't think it's worth it.

I think your idea can be improved on:

(a) instead of hardcoding conditions that we think indicate spamminess, the server should be keeping track of what characteristics spammers tend to display (sprinkle Bayesian magic pixie fairy dust here)

(b) new threads that are neither clearly spammy (banned) nor clearly unspammy (allowed on as now) should be provisionally allowed on the new page, but then banned should a single reader flag it, without the intervention of a moderator

(c) moderators should be able to unban threads if need be

The important thing is: although we have to play whack-a-mole with the spammers, we must take care to play whack-a-mole at as high a level as possible and not have to worry about the details.

[aj]

<i>(b) new threads that are neither clearly spammy (banned) nor clearly unspammy (allowed on as now) should be provisionally allowed on the new page, but then banned should a single reader flag it, without the intervention of a moderator</i>

Don't you think that will start flagging wars or egregious flagging without a just cause? I think a score should be maintained by tracking how many users flag an article and based on a combination of Bayesian filtering and user reputation (karma) and perhaps some other factors, the post be killed or let remain.

[mooism2]

Maybe.

I was thinking that if the server marginally decides a thread isn't spam, it should be easier for readers to overrule it, and so classifying threads into 3 threads is better than into 2.

I think your idea is better.