Endgame Design: Guild Showdowns and Hard Mode

Dungeon Division is a mobile dungeon management game where you side with the dungeon instead of the heroes. You assign monster staff to each floor and stop the adventurer parties that come raiding. On the afternoon of September 29 I finished the tenth and final story chapter. The story was done, but anyone who reached the ending would open the game the next day and have nothing to do.

So this post is about endgame design. By endgame I mean the content you keep playing for weeks after beating the main story. It covers September 29 and 30, plus October 4 when I came back to retune the difficulty: 25 commits in total. As usual, every image here is a screen I saved to the repo at the time. The guildmaster art in these shots is still placeholder art.

I wrote down what I wanted first

At 4 p.m. on the 29th I committed a design doc, 1,043 lines, titled "End content: Guild Showdown, Story Hard, T6 gear". I put my own requirements in a separate block at the top and marked everything proposed below it as a suggestion, not a decision.

Here's what I wanted:

• It unlocks after you clear chapter 10.
• You fight adventurer guildmasters who are on the same level as the bosses on your side. You don't bring the whole dungeon, only four staff.
• There are four guildmasters. You don't fight all four at once; you pick one that fits your current strength.
• Three official entries per day. An entry is spent the moment you go in, and you don't get it back if you lose or give up.
• T6 gear that only drops there. Until then, T5 was the top tier.
• A story hard mode that replays chapters 1 to 10 against stronger enemies, which you can only push through by collecting T6.

The very first line was "right after chapter 10, your four strongest staff can take on the first opponent at a slightly hard difficulty." Turning "slightly hard" into numbers took about half of those two days.

The doc also listed five problems to solve up front. Whole-dungeon power and four-staff power are different things, so I couldn't just divide the chapter 10 recommended power of 30,800 by four to build the first boss. T6 stats had to feed into the actual battle, not only the number on the card. Allowing a team of four boss staff meant checking global effects, summons and position logic, rather than only loosening the formation check. And there could be no loop where you need T6 to get T6.

Four guildmasters

The guildmasters are the people the adventurer guild alliance sends to re-evaluate your dungeon. They don't die when they lose; they retreat and leave their guild crest and gear behind. You fight them every day, so I didn't want major characters dying and coming back over and over.

They share a set of rules: a basic attack plus two core patterns, phase changes at 70% and 35% HP, an enrage at 150 seconds, and a 180-second time limit. Every hit is telegraphed before it lands, and nothing asks you to drag units out of the way. You answer with your lineup, front and back rows, healing, shields and ultimate timing. Battles are automatic, like most idle RPGs, so the player who planned well should win, not the one with faster fingers.

Dev notesBring four boss monsters and you just stun-lock them

If you scale up a regular enemy and call it a boss, it melts to stun chains and instant-kill effects. So showdown opponents get their own resistance rules. Instant kill, banish and charm don't work. Stun and sleep last 30% of their normal time, capped at 1 second, followed by 5 seconds of immunity. Phase changes and the enrage can't be cancelled by crowd control. Damage based on max HP is capped at 0.5% of the boss's max HP per second, per staff member.

These rules only apply to showdown opponents and don't leak into staff or other modes. A separate pattern check makes sure tank and support skills still matter. Battles also need to come out the same at 1x, 2x and 3x speed, so the checks run the same random seed at each speed and compare the winner and the clear time.

Turning "slightly hard" into numbers: a 60 to 75% win rate

At 5:14 p.m. the same day, while wiring up a Varkan prototype, I measured the baseline. I made three simulated saves that started a new game and actually cleared all 118 normal stages. One cleared with no relics, one spread relics across the top 12 staff, and one stacked relics on four boss staff. No gear or currency was handed out by hand.

The top four staff in those saves came out at 9,862, 13,105 and 13,625 power. The draft design said first-guildmaster power should be the median times 1.05, which gave 13,760. That's about 40% above someone who arrives with no relics, so I set the first boss to 10,355, matching the no-relic save.

The target was a 60 to 75% win rate with the top four no-relic staff, and a median of 120 to 165 seconds in the fights you win. With Varkan at 80,000 HP and 2,800 attack, 100 random seeds gave 71 wins and a median of 161.5 seconds. All 29 losses ended with the boss at about 4% HP on average. That was close to what I'd written in the doc: you shouldn't get wiped at the start, you should win by surviving the last big hit or getting one more ultimate off.

The lineup mattered a lot. In the same save, a team with at most two bosses won 0 of 100. Swapping Morgana for Naia took it to 61 wins. The power numbers were nearly the same; the difference was whether the team had a healer.

T6 gear and 118 hard mode stages

On the night of the 29th I added the T6 gear screens and Hard chapters 1 and 2. T6 is the next tier for the existing weapons, armor and accessories, not a new slot. Flat main stats like attack, HP and defense are 2.4 times T5. Percentage stats like crit and attack speed stay at T5 values, because scaling those up at high enhancement levels makes the numbers blow up.

Instead, each T6 piece gets a unique stat: +12% attack on weapons, +15% max HP on armor. One piece alone should make a visible difference. Measured across all level 70 staff, swapping in a single Epic +10 weapon raised a Common staff member's power by a median of 14%.

T6 only comes from Guild Showdowns. The shop, ads, Hard mode and repel runs never give it out. A win has a chance to drop one T6 piece, and every win gives guaranteed Guild Tokens. Save up enough tokens and you can pick the slot, base item and element yourself and convert them into T6. Pure drop chance means the weapon you want might never show up, which is why that exists. The first win against Varkan also gives a T6 weapon choice box, so you can put it on a staff member right away and feel the difference.

Hard mode sits on top of the normal story without touching it. Names, class lineups, elements and dialogue are reused from normal; only enemy stats go up by fixed multipliers. Each stage has seven regular floors, a boss floor and three guard slots. The three stars are: defend successfully, wipe out the enemy, and keep 70% of your staff alive. The story and ending are the same, so in fiction Hard is HQ's training room "recreating high-risk defense records." Dialogue you've already read is skipped.

On the afternoon of the 30th I extended Hard through chapters 3 to 10, connecting all 118 stages and 354 stars. The finished guildmaster screens (lobby, formation, battle, results) went in the same day.

Reviewing what I'd built: the fights looked like still pictures

On the 30th I had one AI tool implement the finished screens and another AI tool review them, reading the code and the captures. The review log lists 13 issues that got fixed.

Some were plain bugs. If you left the results screen with any button other than "Check Results", the result stayed unconfirmed, and the game dragged you back to the guildmaster screen every time you launched it. If you used Instant Showdown with "all remaining entries" on an opponent you'd beaten three times, you only saw the last fight and never saw the gear from the first two.

The one that stood out was the battle itself. The showdown battle was reading combat events and throwing them away, so the characters just stood there with no damage numbers and no hit reactions. I hooked up damage, crit and healing numbers using the same number font as normal battles, a red flash on hit, and a lunge forward on attack. Numbers get grouped in 0.25-second windows so they don't flood the screen at 3x.

Opponent cards now compare the recommended power with your lineup and color it: green for fair, gold for a challenge, red for risky. The requirement was "pick one that fits your strength," but at first the cards only showed your win count, so you had nothing to go on.

Balancing the game with an 84-day simulation

The 3:15 p.m. commit on the 30th was the longest piece of work in this post. To see whether Guild Showdowns and Hard mode would hold up as goals for several weeks, I ran simulated players through 84 days, one day at a time, using the real rules. Each day they did 15 repel runs, collected business income, challenged a guildmaster, put any T6 they got on their four best staff and enhanced it, and played up to 12 Hard stages. That was three saves times three play habits (three times a day, once a day, five days a week), with eight random paths each: 40 paths. One full run takes about 20 minutes.

The problems showed up right away. A player who had built up relics beat Silvaren, the third guildmaster, 30 out of 30 times on unlock day. The goal for Aurel, the last one, was a first win around day 22 to 42; relic players got it on days 1 to 3. After 84 days the relic player was also sitting on 2.4 million gold with nothing to spend it on.

So I changed three things: Silvaren's HP and attack ×1.4, Aurel's ×1.5, and T6 enhancement gold from 600 to 1,000 per step (72,000 to 120,000 total up to +15). The rerun gave Silvaren 0 out of 30 on day one, a first Aurel win on days 7 to 9 for relic players and day 19 for no-relic players, and leftover gold down to the thousands. Aurel was still faster than the target, but any stronger and no-relic players couldn't win within 84 days. ×1.5 was that line.

Dev notesWin rates were either 0% or 100%

The simulation turned up two new problems. The first was the first-win weapon box. It's a Varkan-only reward, but simulated players who practiced their lineups beat Igret, the second guildmaster, on day one and skipped Varkan. Not one of the 40 paths ever got the box.

The second was that win rates moved like a staircase. Go slightly over a power threshold and you win every time, and a single change in front-row placement could change the power you needed by more than 1.5 times. I wrote down that a real "slightly hard" band would need wider battle randomness or pattern changes, not stat tweaks, and moved on for the day.

How the simulated players picked gear caused trouble too. On the day 28 save, re-equipping for the highest power number gave a four-staff power of 14,474 and went 0 out of 30 against Aurel. Keeping the lineup and gear that had been winning since day 14 gave 13,810 and went 30 out of 30. That's where the note "don't recommend gear swaps by power number alone" comes from.

October 4: a new power formula, then harder bosses

A few days later I changed how power is calculated. The old number was mostly a sum of HP, attack and defense, so it barely counted skills, healing or shields. The new one uses the battle engine to evaluate damage contribution and survivability over 30 seconds. That changed the scale of every number, and the guildmaster recommendations became 30,000 / 50,000 / 82,000 / 93,000. I ran 8,091 battles with identical conditions before and after the change and checked that winners and times stayed the same. Only the numbers were supposed to move, not the fights.

That night I made the guildmasters stronger once more: HP and attack +5% for Varkan, +20% for Igret, +25% for Silvaren, +30% for Aurel. At +10% or +15%, Varkan beat all nine lineups from the no-T6 entry save 30 times out of 30, which would have blocked the first T6, so I stopped at +5%. I compared 17,280 battles before and after with the same lineups and seeds. In the 60-day model, first wins landed on day 2 for Varkan, day 12 for Igret, day 21 for Silvaren and day 53 for Aurel. That's a single path, so those aren't the days a real player will hit.

I also removed the no-reward practice mode. The three entries a day and losing an entry on a loss stayed; only the practice entrance went. Up to that day the simulated players had been testing lineups in practice first, so I noted that the earlier day counts were on the optimistic side. Earlier, on the night of the 29th, I'd also dropped a pity rule that guaranteed T6 after five misses in a row, and made each win roll its own drop.

Looking back

An endgame doesn't end the way a story does, so measuring how long it lasts was more work than building it. Letting 40 simulated players live through 84 days showed me exactly where my target days broke. Still, those are the model's numbers. What real players do, I'll only find out after launch.

How I wrote the main story up to chapter 10 is in EP.06, and the day I first built the balance simulator is in EP.03. The next post covers October 7 onward, when the placeholder guildmaster art got replaced and I redrew all 54 adventurers. If you want to see the game first, the game page (Korean) has more screens.