The following model is my new hypothesis for generating better OKCupid profiles for myself while remaining honest.

• I brainstorm what I want to include in my profile in a positive way without lying. This may include goal-factoring on what honest signals I'm trying to send. Then, I see how what I brainstormed fits into the different prompts on OKCupid profiles.

• I generate multiple clause-like chunks for each item/object/quality of myself I'm trying to express in my profile. I then A/B test the options for each item across a cross-section of individuals similar to the ones I would want to attract on OKCupid. This may include random assignment to conditions to participants to some extent. I would still need to think of metrics or ratings for this to best suit my goals.

• Construct complete paragraphs for the various sections of my profile using whichever were the most successful Caveats: I would want enough experimental control to ensure the test participants were people I could trust to respond honestly, and without trolling me. However, this would decrease random selection. How much should I care about random selection, and thus external validity, in this case?

Otherwise, what do you think of the model? What's wrong with it? If it's not completely awful, I'll play-test it with an OKCupid profile just for the value of information, and see if we can't learn something.