That actually doesn't fix the problem as far as registrars are concerned, because the fees only apply if you get refunds above a threshold of 10% of the domains registered. That puts registrars in a uniquely advantageous position for tasting because they register so many domains.
Here's a strategy a registrar could use. For a given 10 minute period, you take the number of domains actually bought by customers, and call that X. Then you multiply that by 10% and call that Y. Now, you take all of the domain searches for unregistered domains that have come through you within that 10 minute period, and run them through a machine learning algorithm that ranks them by the likelihood that they will be purchased. You take the top Y of that list, and taste only those. Now you are guaranteed not to exceed the threshold where you start getting charged for deleted domains.
I am not sure how this strategy would work if the end registrant gets a refund. It may be that that ends up adding to the registrars own tally of refunds, but it would be trivial to tweak the above strategy to decrease Y to adjust for customer refunds, with a controllable amount of uncertainty which can be optimized to maximize the profitability of the strategy over the long run.
Here's a strategy a registrar could use. For a given 10 minute period, you take the number of domains actually bought by customers, and call that X. Then you multiply that by 10% and call that Y. Now, you take all of the domain searches for unregistered domains that have come through you within that 10 minute period, and run them through a machine learning algorithm that ranks them by the likelihood that they will be purchased. You take the top Y of that list, and taste only those. Now you are guaranteed not to exceed the threshold where you start getting charged for deleted domains.
I am not sure how this strategy would work if the end registrant gets a refund. It may be that that ends up adding to the registrars own tally of refunds, but it would be trivial to tweak the above strategy to decrease Y to adjust for customer refunds, with a controllable amount of uncertainty which can be optimized to maximize the profitability of the strategy over the long run.