PvE vs PvP
There's a definite difference in how Player-vs-Environment and Player-vs-Player sandbox design tends to be approached, because the player experience is intrinsically different. PvE balance is relatively easy to accomplish, as the experience on only one side of these interactions actually matters (AI don't need to have a good time). However the point of balance in PvP isn't at a broad picture, since the victim's experience must be accounted for. This can be hard to achieve when a game shares the same sandbox for both PvE and PvP, because they're generally first conceptualized for PvE, as that's often where the bulk of development is spent. PvE balance decisions are often heavily based on trade-off of usage, easily breeding power weapons and abilities, but while frustration is a natural element of competition, it's only inherent due to decision. Making a mechanic's function share the blame is uncompetitive, so the best approach is to balance your sandbox exclusively in PvP, as there's no conflict with then implementing it in PvE.
It's good practice to write out your concepts in detail while they're still fresh, before building the content in-engine. This makes it easier to quantify and make changes to your sandbox, as well as keep a focus on the bigger picture, rather than tunnel-vision on specific content. An emphasis on diverse content will breed dynamic gameplay, while the extraneous only gets in the way, so don't justify anything for the sake of quantity. It's also worth noting that while realistic functions may be an easy way to add depth (complexity) to your gameplay, it's very easy to take that too far. Realism does not necessarily equate to fun, or balance, so don't be afraid to be a little "arcadey". If your goal is to create skillful gameplay or a particularly skillful item, then make its function more predictable by taking most (if not all) of the dice out of the equation, however this doesn't necessarily mean trading off difficulty of use. Also, fast-paced gameplay can be fun, but allowing (not forcing) players to think and outwit their enemies is a great deal moreso; reaction times shouldn't be everything. Lastly, you shouldn't allow your content to become too complicated or boring, so that it maintains the frequency of use best suited to its role. It's never simply enough to balance effectiveness; people aren't computers.
Creating a diverse sandbox is a tough task, and when your pieces are most often judged by how fun they are to use, rather than how original they are, it can also be a discouraging one. However artistic styling can have great impact on the frequency of use for an item, as both initial and continued utilization is based on perceived effectiveness, which can significantly vary from actual effectiveness. So before considering buffing an item to improve how commonly it's used, first examine how it appeals both visually and audibly. The function is your product, but art is what sells it, so they should always have a close relationship with each other. The roles of your content should also be apparent to the user without an explanation, so you must employ aesthetic cues to help give them an idea of what to expect.
Balance should always begin at the written concept, and progress in passes until nearly the end of development, so get your foundation down early and keep polishing. The best way to finalize your content's individual balance is by asking yourself whether you could stand to make it a little more powerful, or a little weaker, which tells you a lot about its standing in your sandbox, and which way you should adjust it. If you feel you could acceptably adjust your piece's effectiveness in either direction, then you generally have it where you need it. Though whenever introducing a new item to an existing sandbox, try to initially under-power it and work up from there. This can keep from upsetting your sandbox during testing, and it's always much easier to buff an item than it is to nerf it. In any sandbox containing weapons bundled with differing mobility and health modifiers, such as when vehicles are involved with infantry, weaponry cannot be independently balanced against each other. Incidentally, one should also be very careful in allowing a weapon's use across different modifiers, which is a tendency when they're not vehicles but instead classes. Most importantly, a role's weakness should be no more than disruptive, unlike the decisive elements of a Tic-Tac-Toe match; a good test of this is through playing out 1v1 encounters without respawns. Any game can be fun if not taken seriously, so much of your job is to make it retain that value at high-level gameplay, where the are no holds barred and the objective is simply to win. Lastly, the polishing of balance in the end should be left to just one person, to maintain a consistent perspective.
PvP Level Design
The balance of a diverse sandbox is dependent in part on its context, which is the job of your maps to create varyingly. The best place to start is with the flow rather than actual geometry of your level, by drawing player trajectories with creative intersections that you only then build your features around. This helps in setting gameplay as the priority and avoiding extraneous complexities. Flow habits will be based heavily on where players feel in control, meaning they'll seek predictive engagements, such as narrower fields of play where the variables are fewer. This can sometimes be counter-intuitive, as such paths tend to generate the highest traffic, but predictability goes a long way in making one feel comfortable, regardless of effectiveness. So to maintain relevant flow, less-linear alternatives need to feature strong incentive. Additionally while the occasional choke-point can be a great addition, they should never be made unavoidable, or your combat here will quickly stagnate as tactical diversity is lost. Regularly running through your level will be an important habit to adopt, mentally playing out various encounters as you go, and paying particular attention to sightlines as well as each route's incentives. Sightlines are your most important features, so they should be more than just reactively considered. Your map should also incorporate a variety of unique landmarks not only to make key locations more enticing, but to greatly simplify callouts.
If your playtests aren't embarrassing, you're doing them far too late; playtesting early in the development process allows you to nip problems in the bud, before you've invested too much in them. Rebuilding or axing things only gets tougher as time goes on. Whenever possible, PvP playtests should be recorded from opposing perspectives for review, which offers a huge advantage in tuning the player experience. Your testers should also be made aware of any significant changes once implemented, and be sure to personally partake in the gameplay following such change. Quite often a failing in balance judgement is just inexperience with the content, so allow your testers to get comfortable with your sandbox before taking their balance concerns too seriously. However, listening to a fresh perspective on your game can help you to improve that acclimation process for future players. Playtesting of course isn't only for tuning balance, but also for finding exploits and bugs, which you're in a far better position to find than your testers will be. You know better than anyone else where and what to look for because you know exactly what went into your content, so always be actively testing in an attempt to break it.