The Three Rules of Homebrew Balance
1. Compare to existing content. 2. Test at the tier you are designing for. 3. Make it situational, not universally optimal.
Rule 1: always find the closest official equivalent and compare. Designing a magic sword? Compare to a Flame Tongue (rare, +2d6 fire). Designing a feat? Compare to Great Weapon Master or Sharpshooter (the ceiling for martial damage feats). If your homebrew is strictly better than the closest equivalent, it is probably too strong. If it is strictly worse, nobody will use it.
Rule 2: test at the tier you are designing for. A magic item balanced for level 15 will break a level 3 campaign. A subclass feature balanced for early game will feel irrelevant at level 14. The community designs for four tiers: 1-4 (local heroes), 5-10 (regional heroes), 11-16 (realm-shaking), 17-20 (world-ending). Know which tier your homebrew targets.
Rule 3: the most elegant homebrew is situational. A sword that deals +1d6 damage to undead is more interesting than a sword that deals +1d6 damage. It is also less likely to be universally optimal, which means the player has to think about when to use it. Situational power is the designer's best tool for creating content that feels powerful without breaking every encounter.
The Three Most Common Homebrew Mistakes
Giving extra actions. Removing concentration. Allowing stacking bonuses from multiple sources.
The action economy is the most fragile part of 5e. Any homebrew that grants extra attacks, bonus action spells, or additional reactions should be treated as potentially game-breaking. A feature that lets a Wizard cast two leveled spells per turn (normally impossible without Action Surge) is not "strong" — it fundamentally breaks the encounter math. Limit action-granting features to once per short rest, or tie them to a resource cost (spell slots, hit dice).
Concentration exists for a reason. Removing it from any spell that currently requires it (Haste, Polymorph, Greater Invisibility) is the fastest way to break your campaign. Homebrew items that "let you concentrate on two spells at once" or "remove concentration from X spell" should not exist at any rarity.
Stacking bonuses: bounded accuracy means small numbers matter. A +1 weapon, +2 from Archery fighting style, +1d4 from Bless, and advantage from being hidden is +3.5 + 1d4 with advantage — that turns a 50% hit chance into roughly 85%. Homebrew should not add static bonuses that multiply with existing ones. Prefer features that grant new options (you can use your reaction to X) over features that grant numerical bonuses (you get +1 to Y).