For a typical deck (spells averaging 3.25 mana), every land count in this range can be built within 1 mana of the best deck over 10 turns.
The best total of mana sources climbs as the average spell cost goes from 2 to 4.5. One number can't fit every deck.
How little swapping lands for cheap ramp changes over 10 turns, once the total number of sources is right.
How much the best total rises if games are decided by turn 8 instead of turn 10. The answer depends on your table's speed.
The myth, tested
For each curve, the range of land counts (and of total sources) whose deck stays within 1 mana of the best one in its worst draw scenario, over 10 turns. 1 mana is about 3% of what a deck spends in that time.
A wide range is not “anything goes”. Outside it, decks either stall for mana or flood. And this model leaves out things that push real decks toward more lands: color requirements, removal that kills mana dorks, and lands that enter tapped. The point is narrower and firmer: a fixed land count, given without knowing the deck's curve and speed, is not backed by the numbers. 24 lands is the lowest we tested, so the bottom of each range may go lower.
Explore every mana base
Each square is one deck: lands going up, ramp going across. Darker means closer to the best deck for that curve. Tap or hover a square for its numbers.
Most resilient bases
The right total depends on your curve
The most resilient deck (smallest worst-case gap across the 8 draw scenarios) for each average spell cost. Cheap decks want few sources and no ramp; heavy decks want nearly 50.
The 8 draw scenarios
Most decks draw extra cards once their engines are online. The scenarios add cards on turns 5 to 8, from none to 7 extra cards.
| Scenario | T5 | T6 | T7 | T8 | Extra cards | Best base, all curves |
|---|
With few extra cards, the deck runs out of spells: more lands and less ramp do best. With lots of extra cards, ramp pays off because there is always something to cast. A base around 42 sources is never far from the best in either case, and that is what “resilient” means here.
Dorks or rocks?
A zoom on 45 bases with 20,000 games each, averaged over all curves and scenarios.
Dorks come out ahead because they cost 1 mana instead of 2. But nothing in this model kills a creature, so in real games with board wipes, rocks are safer. Treat this table as “cheap ramp beats expensive ramp”, not “play elves over Signets”.
A trap we found: the fixed total budget
The first version fixed the total mana cost of the 99 cards (130 to 230). That sounds fair, but every land you add takes no budget, so the remaining spells get more expensive. Since the model counts all mana as equally useful, a 5-drop scores like two 2.5-drops, and the “winner” runs 47 to 55 lands.
That is why the explorer fixes the average spell cost instead: adding a source then removes a spell of the same size, which is the real trade-off when you build a deck.
How the simulation plays
- The deck: 99 cards plus a 4-mana commander. Dorks cost 1 and tap the next turn. Rocks cost 2 and tap right away. Spells all cost the curve's average, rounded up or down so the total is exact.
- The hand: keep 2 to 5 lands (or 1 land and 2 ramp). The first mulligan is free; the second puts a card on the bottom.
- Each turn: draw (everyone draws on turn 1 in multiplayer), play a land, cast the commander as soon as there are 4 mana, cast ramp first through turn 6, then spend as much mana as possible on spells.
- Measured: mana made, mana spent on real spells (the commander counts, ramp doesn't), unused mana, spells cast, and spells stuck in hand, through turn 10 and through turn 8.
- Fair comparison: every deck plays the same shuffles, so small gaps between decks are real, not luck. 5,000 games per deck, curve and scenario.
- Resilience: for each curve and scenario we find the best deck, then measure how far behind every other deck is. The most resilient deck has the smallest gap in its worst scenario.
What this model gets wrong
It is a goldfish: no opponents and no real cards. These gaps matter, biggest first.
- All mana counts the same. Mana on turn 3 matters far more than mana on turn 10, but here they score equally.
- Extra cards are free. Real card draw costs mana, and real decks have mana sinks. Here the hand is nearly empty by turn 10, so “unused mana” is overstated.
- The curve is flat. Every spell costs the same. A real hand stuck with three 6-drops never happens here.
- Averages hide the bad games. How often you miss your third land drop says more than the mean.
- Smaller gaps: no removal on dorks, no Sol Ring or tapped lands, no colors, a simple mulligan rule.
Earlier studies
Across 11 decks in our game engine, win rate against Bracket 4 bots tracked the share of goldfish games won by turn 6 (correlation 0.87), not wasted mana (−0.28) or useful mana (0.39).
miku-deck/b4-research/MANA-STUDY.md44 public Bracket 4 lists average 33.7 lands and 10.6 accelerants. 22 cEDH lists average 28.9 lands and 17.5. Over 10 turns, the average Bracket 4 base is within 1 mana of the best for any curve from 3 to 4.5.
miku-deck/b4-research/MANA-STUDY.mdPaired seeds cut the noise of deck benches about 8 times, which is why every deck here plays the same shuffles. The report also rebuilds Karsten's coloured-source tables for 99 cards.
research/mtg-analysis-methods/REPORT.mdRun it yourself
The simulator is plain Node, no install. A full sweep takes about 3 minutes on 4 cores.
cd tools/sim/manabase node manasim.js sweep --mode avg --games 5000 # every base, fixed curve node manasim.js sweep --mode budget --games 5000 # the fixed-budget version node manasim.js one --mode avg --config 34:5:6 --budget 3.25 --games 20000 node analyze.js results-sweep-avg-tight.json --metric useful8