Title: Engage’s balance is broken at the formula level. I rebuilt it from scratch and I’m tuning enemies with an LLM reading playtest logs — tell me where this falls apart.
Engage’s lategame isn’t “undertuned,” it’s structurally broken. Avoid stacks with no ceiling — terrain +30, doubled to +60 on covert, plus support — until enemy hit hits ~0%. And their “fix” is the same band-aid slapped over every kind of collapse at once: enemies just skip any unit they can’t meaningfully hurt — 0% to hit, 0 damage, doesn’t matter, same patch. It doesn’t fix anything, it hides every failure mode under one rule. It’s a band-aid on a corpse that’s already been through a meat grinder. The avoid stacking is still there, the damage cliff is still there — the game just looks away from both.
I built a from-scratch SRPG (not a GBAFE hack) so I can change the formulas. My fixes:
・Hit floor at 30%, no ceiling. I clamp the floor only — permanent sources (base avoid, terrain, supports) can’t push enemy hit below 30%. Only temporary effects (active skills, consumables) can go under, down to a second floor of 10%. I’m deliberately not capping the top: I kill avoid-stacking by capping the avoid inputs, not by hiding hit behind a ceiling. A displayed-hit ceiling is the same symptom-hiding as Engage’s skip hack — it leaves the real cause (avoid with no ceiling) untouched.
・No subtractive-damage cliff. Instead of letting DEF cross ATK and pin damage at 0/1, I cap final DEF (all bonuses included) so the wall-unit-takes-0 case can’t exist.
・HP scales to a per-turn survival line, not a flat curve. Enemy burst grows compound; HP follows it for exactly one turn of hits, and AoE healing covers the cross-turn accumulation. HP has a hard cap at ~1.3–1.5× the survival line so the rare triple-rolled tank can’t go invincible.
Here’s the part some of you will hate: I’m not hand-tuning enemy stats. I auto-playtest each map, dump the logs (hit cdistribution, hits-to-kill, clear turns), and feed them to an LLM that adjusts enemy stats and re-runs. The human writes the rules (the floor, the survival-line target); the model just chases the target from data instead of me eyeballing it. Build is fully playable, ch01 generated end to end. Maps are rough because the log corpus is still thin.
■Questions for people who’ve actually shipped/playtested balance:
１：Does a hard 30% floor feel blunt in play, or is Thracia’s 1–99 hated for the inputs swinging too wide rather than the clamp itself?
２：Final-DEF cap vs. minimum-guaranteed-damage vs. sub-1 DEF coefficient — which holds up best lategame in your experience?
３：For anyone who’s tried data-driven / automated tuning: where does “tune to a target metric” break down? My fear is it converges to “technically survivable” maps that are boring — that the metric can’t see fun.
４：The big one. Everything FE does to patch lategame collapse is symptom-hiding: “enemies skip 0-damage units,” guaranteed-hit weapons to dodge the avoid problem, etc. — they hide the broken number instead of fixing why it broke. My clamps are arguably one tier up (capping inputs instead of masking outputs), but I don’t think they’re the root answer either. So: has anyone found a genuinely structural fix? Something that stops player power and enemy threat from diverging in the first place — not a clamp that catches them after they’ve already split. I’m thinking about the player-scales-multiplicatively / enemies-scale-additively asymmetry as the actual disease. How do you treat that, short of just rubber-banding enemies to player stats (which kills the reward of getting stronger)?
(English isn’t my first language — this is partly machine-translated, so apologies if anything reads oddly.)
