Tower-defense numbers are mostly already public

A 30-minute field guide · six principles from public sources, one from our own parameter sweeps · every number checkable

中文 →

Written for people who have never balanced a tower-defense game. By the end you will know how much to send in a wave, what a tower should cost, how to write "heavily armoured" without breaking the game, what a difficulty curve actually looks like — and why "just tune it until it feels right" is much harder than it sounds.

One rule throughout: every specific number here either comes from a source you can open, or from a script you can re-run. Where we could not reach the original, we quote no numbers at all.

1"How hard is this wave" can become a single dial

Source: the Defender's Quest team · Building a Better Wave

Their finding: "difficulty" is not one number in your game. It is four things arriving at once — how many enemies you face, how strong they are, how densely they are packed, and how strong you are by comparison.

So they stopped hand-writing waves. A designer says two things: how hard this level starts and how hard it ends, plus a stress multiplier per wave (1.0 = baseline). The tool solves backwards for how many enemies of what strength that implies.

The win isn't automation. It's that difficulty becomes a curve you can shift and stretch as a whole, instead of two hundred numbers scattered across a spreadsheet.

Actual wave weight for the first 100 rounds of Bloons TD 6 (log scale). Red dots = this round is lighter than the one before it.

Round 1 is 20; round 100 is 67,200 — an average of +8.55% per round. Yet 62 of the 139 round-to-round steps go down (44.6%).

The sawtooth isn't a figure of speech — it's in the data. Roughly every other round is lighter than the one before it. The lightest mid-game round is round 24, at just 43 weight. The deepest breather of all is round 90, 93.6% lighter than round 89 (55,712 → 3,556) — immediately before round 100 releases the largest enemy in the game. You let the player breathe first, or the spike doesn't land.

There's a blunter fact too: BTD6 only hand-authors 140 rounds. After that, waves are generated from a budget, and the wiki spells the formula out: RBE budget = round × 4,000 − 217,000. Not even Ninja Kiwi kept hand-writing.

How to use it: express "how hard is this level" as a curve and let a tool solve for the enemies. Tuning difficulty means editing the curve, not two hundred numbers. And schedule the breathers deliberately — without them the player can't feel you ramping.

2Range is worth far more than damage

Source: the same team · Optimizing Tower Defense for FOCUS and THINKING

+1 tile of range does two jobs at once

It covers more path (more places you can shoot).
Each enemy stays in range longer (more total damage landed).

+10% damage does part of job 2 only. The article's word for it: range is overpowered.

Consequence: short-range towers are born unused. The only way to save one is to pay it in a different currency — bigger hits, crowd control, armour shred, splash. "The tower nobody ever builds" is the number-one sign of an unbalanced tower defense.

Kingdom Rush, actual numbers (melee vs ranged, same price point)

Tower	Tier	Total cost	DPS	Gold per DPS
Archer Tower	1	70	6.88	10.18
Marksmen Tower	2	180	16.5	10.9
Sharpshooter Tower	3	340	28.6	11.89
Militia Barracks	1	70	6	11.67
Footmen Barracks	2	180	7.72	23.31
Knights Barracks	3	340	17.65	19.27

For the same 70 gold at tier 1, ranged buys DPS at 10.18 gold each; melee pays 11.67. By tier 3 the gap widens to 11.89 vs 19.27 — melee DPS costs 62% more.

Melee loses the DPS column at every single tier — and is still a pillar of that game. Because DPS isn't what it sells. It sells holding enemies in place, and that isn't in the table.

How to use it: price towers by role, not by damage. Every short-range tower must carry a non-damage ability, or it shouldn't ship.

3Eight hard rules that let the player think

Same article. It defines the fun of tower defense as: a test of thinking, never of reflexes

Rule	Why
No scrolling	The player never worries about what's off-screen
All information visible	Every stat readable; nothing memorised
Pause and speed-up any time	Thinking time belongs to the player
Melee towers must be viable	See principle 2
Legible state	You can see who is shooting what at a glance
No single-answer enemies	That makes players recall an answer, not find one
No flying enemies	The article calls this the worst lock-and-key design: without anti-air you simply cannot touch them
No reflex tests	Tower defense shouldn't reward clicking fast

How to use it: treat it as an acceptance checklist. "No flying" is the one teams skip — if you want the effect of "some towers can't reach this", build it out of pathing (a side route, a shortcut, a mid-path insertion), not out of immunity.

4Armour should subtract — and must keep a floor

Source: the Arknights damage formula (the most carefully modelled resistance system in the genre)

Physical damage = ATK − DEF
Arts damage = ATK × (1 − RES%)

but never less than
5% of ATK

Verbatim: The minimum value of Dmg_phys is 5% of Atk_final.

A 100-ATK tower against various DEF values:

DEF 0 → 100 damage

DEF 60 → 40 damage

DEF 95 → 5 damage

DEF 999 → still 5 damage (the floor holds)

Why the floor matters: without it, an armoured enemy becomes "bring the right tower or you literally cannot hurt it" — the worst lock-and-key from rule 3. With it, bringing the wrong tower makes you slow, not dead.

How to use it: make every resistance a multiplier, never an immunity, and give it a percentage floor. Same for weakness systems — a weakness should make the right answer faster, not make the wrong answer useless.

5A whole wave should collapse into one number

Source: RBE (Red Bloon Equivalent), from the Bloons TD 6 community

Bloons split into smaller bloons when popped, so "how heavy is this wave" has no obvious answer. The community invented a unit: express the wave as a count of the most basic red bloons — i.e. total hits needed to clear it.

wave weight = one number

With that number, previously unanswerable questions become arithmetic: how much heavier is wave 40 than wave 30? Does this level ramp smoothly or spike? Is the player's firepower enough right now?

This is the other face of principle 1: stress is the feeling, equivalent weight is the ruler. You can only tune the feeling once you have the ruler.

The BTD6 enemy ladder (excerpt; full table in the sources below)

Enemy	Own HP	Speed	Weight
Red	1	1.0×	1
Blue	1	1.4×	2
Pink	1	3.5×	5
Black	1	1.8×	11
Lead	1	1.0×	23
Rainbow	1	2.2×	47
Ceramic	10	2.5×	104
M.O.A.B.	200	1.0×	616
Z.O.M.G.	4000	0.18×	16,656
B.A.D.	28000	0.18×	67,200

4.8 orders of magnitude from lightest (1) to heaviest (67,200) — inside one game.
The first 11 tiers all have 1 HP. Difficulty isn't built by inflating health; it's built by splitting into more layers. The first enemy with health of its own (Ceramic) has only 10.
Heavier means slower. Pink 3.5× vs Z.O.M.G. 0.18× — a 19× spread. Weight and speed run opposite to each other; the two threat axes are sold separately.
Every bloon pays 1 cash regardless of tier. Income tracks pops, not enemy strength — so the economy curve isn't dragged along by the difficulty curve.

How to use it: give every enemy a weight, plot each wave's total weight as a curve, and treat that curve as your difficulty design. Editing it is far safer than editing individual enemy HP. Steal the inverse relationship too: fast must be fragile, tanky must be slow.

6Don't invent numbers from scratch — start from someone's table

The full Kingdom Rush tower table (community-compiled, recomputed here)

Gold per point of DPS, by line and tier

You aren't buying DPS. You're buying a role.

Gold per point of DPS at tier 1: cheapest is Mage Tower at 8.71, dearest is Dwarven Bombard at 19.76 — a 2.27× spread.

The expensive one isn't mistuned. It is expensive because it splashes, and splash isn't in the DPS column. Inside a single table, the same gold buys more than twice the DPS depending on role.

Upgrading is never a DPS bargain

Across all four lines, going tier 1 → 3 makes every point of DPS more expensive:
ranged +16.8% · melee +65.1% · magic +17.3% · artillery +37.6%.

What that premium buys is concentration onto one tile.
Which is to say: the scarce resource is the tile and its range — not the gold. This is principle 2 seen from the other side.

(In the chart the middle melee tier costs more per DPS than the tier above it. That is what the source table says, not a plotting error — that tier is famously the worst buy in the game.)

Corollary: if upgrading is cheaper than building in your game, players will carpet the map in tier-1 towers — a sign you've priced the wrong thing as scarce.

How to use it: let someone else's table set your starting point and orders of magnitude (how much stronger and dearer a tier-3 should be), then sweep for the final values yourself. Get the ruler first (principles 1 and 5); argue about numbers second.

7Balance is not a point. It's a narrow slit.

This one isn't from the literature — it's from our own parameter sweeps

The first six principles are about how to choose numbers. This one is about something more sobering: the region where the numbers are right is far smaller than intuition suggests.

We wrote a minimal tower-defense economy simulator — 10 waves, spatial placement abstracted away, keeping exactly one decision axis: defend now or invest in economy first. Then we gave it seven gates. Note what the gates ask: not "is this winnable", but "is this actually a game with a decision in it". The reference strategy must win; pure defense must lose; pure economy must lose; winning strategies must be a minority and not at the edges of the space; the reference must win cleanly but not luxuriously; a clumsy player must take damage yet survive; and the winner's production must not have been burnt to zero.

Then we swept two axes that pull against each other: defense unit price (dearer forces you into economy) × wave growth rate (steeper leaves less time for economy to pay off).

143 parameter sets, roughly 49,049 simulated runs. 2 passed all seven gates — 1.4%.

The passing region is a vertical sliver, not an area. Of the 11 growth rates swept, exactly one works; the defense price tolerates two adjacent steps. So the two axes are nowhere near equivalent — the pacing axis has almost no slack, the price axis has a little.

If you only ever tune prices, everything will feel about the same — because you are not touching the axis that is actually binding.

129 of 143 sets died on the same gate — the one requiring that winning strategies be a minority and not at the edges.

Translated: the vast majority of parameter sets are not unwinnable. They are winnable but decisionless — either everything wins or nothing does.

"Is there a decision here"
is much harder to satisfy than "can this be won"

Honest limits: that 1.4% belongs to this model and this gate set. It is not a universal constant of the genre. The model runs 10 waves, abstracts placement away, and assumes perfect information; change the gates or the model and the number moves. What doesn't move is the shape: the passing region is a sliver, and what usually blocks you is "is there a decision", not "can it be won".

How to use it: three things. ① Write the gates before you touch the numbers — turn "what would count as balanced" into assertions a machine can check, or you're just oscillating on feel. ② Sweep instead of guessing: tens of thousands of simulated runs are faster than dozens of manual attempts, and more honest. ③ At least one gate must test for a decision; a gate set that only checks winnability will happily pass a dead flat game.

The 30-minute crash course

Read this first (10 min): Optimizing Tower Defense for FOCUS and THINKING — the 8 rules and the range argument
gamedeveloper.com · Defender's Quest
Then this (10 min): Building a Better Wave — generating waves from a stress curve
fortressofdoors.com
Then skim the tables (10 min): Kingdom Rush tower stats · Bloons enemy stats · Arknights physical damage formula

If you want to go deeper: Balance in TD games · Tower Defense Game Rules · BTD6 Rounds (includes the wave budget formula)

In one sentence: tower-defense numbers don't come from inspiration. They come from three things — get a ruler for difficulty first (stress / equivalent weight), use someone else's table to set the orders of magnitude, and write your gates down so a machine can search for the values. Save the inspiration for what makes each tower special.