What I'd steal (and what I'd avoid)
Opinionated takeaways for my own (Godot) projects.
Steal
1. Give the defender a verb
The single most transferable idea in the game, and it costs almost nothing systemically: whatever your combat structure, make the player do something when they're attacked. E33 proves the payoff is enormous even inside the most turn-based of turn-based frames. The dead time between "my turn" and "my turn again" is where genre fatigue lives. Even a small prototype can test this: one dodge button during enemy actions changes the entire feel of menu combat.
2. Pay for risk in resources, not just survival
Parry and perfect dodge occupy the same timing space, but parry is tighter and generates AP. The reward for choosing the harder input is fuel for your best skills, so the economy itself teaches mastery; no tutorial text needed. The general form: when two defensive options exist, make the harder one an economic engine, not just a better shield.
3. Universal passive layer under unique kits
The two-layer build architecture (shared Luminas under per-character subsystems) gets you multiplicative build variety at additive content cost. Every Lumina I design works for six characters; every kit stays mechanically private. Balatro-scale combinatorics with a party-RPG face.
4. Equipment that teaches (equip-to-learn)
"Win 4 battles wearing it and keep the passive" is a rotation-forcing loop: optimal play is to keep cycling unlearned gear, so players explore the catalog by playing optimally instead of despite it. Any game with equipment and passives can retrofit this with one counter per item.
5. Ship the failure screen in the fiction's voice
"Expedition Failed" and "We Continue" cost nothing and do real thematic work. Naming your failure and continue states in-world is the cheapest ludonarrative move available (see the full reading).
6. Refill consumables at checkpoints
Tints refilling at every flag kills item hoarding, the oldest JRPG pathology, with one rule. Potions become per-attempt tactical resources instead of a museum collection.
Handle with care
7. Deceptive timing is a difficulty dial with an accessibility bill
The delayed, off-beat wind-ups are the skill ceiling and the mastery joy, and they're why some players bounce off entirely: there's no timing indicator, surreal enemy anatomy hides the tells, and defensive inputs can't be automated on any difficulty. If I use adversarial timing, I'd ship what E33 only patched in later (or never): practice/retry tools from day one (Battle Retry arrived in patch 1.4.0), a timing-assist option decoupled from the difficulty label, and telegraph animations authored on readable silhouettes.
8. If defense is mandatory, audit the "or die" threshold
Act 3 enemies hitting for full health bars turned "parry is rewarding" into "parry or die," and the build system stopped mattering for anyone who couldn't execute (TheGamer's criticism is the canonical writeup). One-shot damage converts a skill bonus into a skill gate. Keep at least one honest build-based answer (tank stats, shields, revives) viable at every difficulty, or admit the game is an action game.
9. Decide your break-the-game policy before launch
The 9,999 damage cap removed by an opt-in Lumina is a great idea; the billion-damage Stendhal build that one-shot the final superboss was that idea untested. Sandfall's recovery was actually excellent (pre-announced nerf, then player-set Challenge Modifiers), but the lesson is to choose deliberately: either the ceiling is content (then test it like content) or it's the player's toy (then hand them the dials on day one, not in patch 1.3.0).
10. Friction needs tooling even when it's the point
No minimap is a defensible artistic stance (Broche's memory argument is real; it worked on me in Hollow Knight). But E33 pairs it with invisible walls, samey corridors, and no in-level progress tracking, so the friction reads as jank half the time. If you remove the map, spend the savings on landmark-driven, silhouette-readable levels. Friction is a feature only when the environment can carry the navigation load the UI dropped.
11. Menu-heavy build systems need loadout presets at launch
Lumina Sets (build presets) arrived in patch 1.5.0, months after players were hand-rebuilding 40-point loadouts per character per boss. If the build system is the endgame, its management UI is endgame content too.