Your hypothesis should be tested as a plugin for Steem. Let the current curation and this new prediction market form exist side by side. Have a "test" or "alpha" version of the site for the extremely new features sort of like how Slashdot has it's beta site tested. Gather real usage statistics and feedback from the users of the new system and compare it with the old.
But I don't support any change which takes away from the old model without first testing and proving the new model through an A/B test or hypothesis test of some sort at least. If the data supports the new model then conduct a poll to see if the users prefer the new model and if more do then switch over.
What about changes that simply reduce curation payouts under the current model while we test new models.
Also we suggested a simple rule that curation rewards are only paid out when an author makes an "above average" post (for that author).
In regards to the "above average" suggestion, most likely an author's "replies" to other posts would be much less valuable, but her "blogs" would be more valuable, so a strategy for bots would be: up-vote "blogs", ignore "replies".
Another example: an author is good at writing "funny" posts but not good at writing "serious" posts, so bots vote for her "funny" posts and ignore "serious" posts.
My 2 cents.
That would be correct curation and would add value.