Many systems have two or more skills/traits/other numeric values that can be pitted against each other in situations where side A tries to assess side B's credibility, where side B may or may not be lying. Among many systems, these skills/traits/values may carry such names as Empathy/Kinesics/Body Language/Detect Lies/etc. and Subterfuge/Acting/Deception/etc. respectively.
Most of the RP-immersion-oriented/associative/character-stance systems I've seen use those two values in an opposed roll of some sort. Usually, if A wins, the referee tells A's player whether B appears to be lying or not. If B wins, no such information is given. For the purposes of the question, how the 'win' is determined is of little concern: some systems count the number of successes scored, some compare margins of success and failure, some have other methods. The point is that in the end of a roll-off, one of the participating characters is deemed the winner. (Also, for the sake of simplicity, let's not consider ties and critical victories/losses/successes/failures.)
This works OK even with open rolls during some sort of hostile negotiation, where B is already assumed to be interested in concealing some information, and it's more a matter of where B tries to mislead A.
However, the above framework breaks down if B is telling the truth and wants to convince A, since in that case suddenly B is interested in having a low trait (or foregoing the roll entirely, if permitted), thus allowing A's lie-detection ability to inform A of the truthfulness involved.
Not only does this produce perverse incentives, but if foregoing a roll is permitted (including by deliberately failing, making A the automatic or near-guaranteed winner), it also results meta hints: a target that doesn't resist lie detection is immediately more trustworthy, while one which does is immediately suspicious to the player even if the character doesn't know the difference. These factors mean that the mechanic is hostile to attempts to build/play an honest-looking good liar.
I'm looking for an alternative approach to using such skills that can be either used when making a system from scratch, or for houseruling the procedure for making such skills (or similar traits) in systems that use them. These are the improvements I'm seeking and the pitfalls I'm trying to avoid:
- Minimise perverse incentives (essential), even if one cannot actually follow them after character creation.
- Minimise possibilities and temptations for metagame ways of figuring out whether a character is lying (essential).
- Avoid increasing requirements for the amount of secret rolls (if possible). In general, making B's roll secret is more acceptable than A's roll, but keep in mind that in the default interpretation above, secrecy of B by itself doesn't solve the prior two issues.
- Avoid excessive complexity (if possible), such as having too many rolls for obfuscation purposes.
Does a design pattern exist for resolving lie detection roll-offs in a way that addresses the above concerns?