← All posts

Simulating a Homeworlds Prerelease

Building a library of sealed decks

In Prerelease and sealed events, deckbuilding has a very different flavor than in Premier. Each player opens six packs and builds the best 30-card deck they can from whatever they get. It can be less about finding the best leader and deck in the set, and more about identifying which aspects from your specific card pool are strongest, and pairing them with a solid leader and base.

In this experiment, we simulated building limited-format decks to get a rough idea of which decks to expect at Prerelease events.

We began with 1,000 simulated entrants. Each one opened their own six packs of Homeworlds and built a deck from the cards they got. A separate deckbuilding model is in progress, so for now decks were built at random and the games decided which were good.

There were four steps.

1. Open the packs. Every entrant opens six Homeworlds packs, generated the same way as the packs on the Sealed page. Every entrant also gets the two Prerelease kit leaders, Grand Moff Tarkin and Chewbacca.
2. Build every reasonable deck. For each leader in the pool, and for each base aspect that doesn't overlap the leader's, we build one candidate deck. Its 30 cards are drawn at random from the in-aspect cards in that pool, and decks that would need more than two off-aspect cards are dropped. This gave 19,578 candidate decks, about 20 per entrant.
3. Play an opening round. Every candidate plays five pairs of games against candidates from other entrants. In a pair, the decks play twice and swap who goes first. Each entrant then keeps only their best deck for each leader.
4. Narrow down against the meta. In each of nine rounds, the remaining decks played five more pairs of games against the current best deck of randomly chosen entrants. Then each entrant dropped their weakest deck. At the end, every entrant was left with one deck: the one they would play.

All together, building the library took 367,870 games. The AI played both sides of every game using its policy alone, with no lookahead search. The 1,000 registered decks are the library the opponents on the Sealed page draw from.

Which leaders get played

Each bar is the number of the 1,000 registered decks led by that leader, split by the color of the base it was paired with.

Tarfful is by far the most common leader. Of the 478 entrants who opened Tarfful (out of 1,000), 292 selected him. No other leader was used by more than a quarter of the entrants who had it. He also wins with any of three base colors, which the archetype results below bear out.

The two kit leaders, Chewbacca and Grand Moff Tarkin, show how differently two "free" leaders can turn out. Every entrant had both. Chewbacca was used by about 1 in 6 entrants, usually when the packs offered nothing better. Grand Moff Tarkin was played by no one, and neither were Omega or Maz Kanata.

Among the rare leaders, Darth Vader stands out: few pools have him (123 of 1,000), but 28 of those entrants played him.

Archetype win rates

We group the registered decks into archetypes: a leader paired with a base aspect, such as Tarfful with a Vigilance base. We took the 15 archetypes with the best record in the elimination rounds, and measured each one against all the decks in the library.

Each archetype played 11,000 games, each one a random deck of that archetype against a random deck from the library. The bars are 95% confidence intervals around the win rates; the dashed line is 50%.

All three Tarfful archetypes finish in the top three, and Tarfful + Vigilance wins 59% of its games, a long way clear of everything else. Doctor Hemlock with a Command base is the best non-Tarfful archetype, at 53%.

The base color matters almost as much as the leader. Doctor Hemlock spans 53% to 50% across his three bases, and Bossk spans 50% to 47%. Two rare-leader archetypes, Darth Sidious + Cunning and Darth Vader + Vigilance, sit at 51% each. They are strong when you open them, but too rare to shape the meta.

The cards that matter most

To measure a single card's value we used a swap test. For example, take a random Hemlock + Command deck. Simulate a game between that deck and a random deck from the library. Then, replay it with one card in the Hemlock deck swapped for the card being tested, keeping the same shuffles at the start of the game.

Any difference in the result is down to the swap. We did this 2,000 times for every card in each archetype's aspects that can be opened in a Homeworlds pack, swapping it into a random slot of a random deck each time. This required a total of about 2.65 million simulated games.

The numbers below show how much the swaps changed the archetypes' win rates, in percentage points. Hover over or tap a card name to see the card.

Most swaps make a deck slightly worse. These decks survived nine rounds of selection, so the card being swapped out is usually at least decent. A card near the top of its list is one the archetype should play wherever it can.

Tarfful + Vigilance59.1%

#	Card	Win rate change	95% interval
1	Bestial Bond	+5.1 pp	+3.4 to +6.9
2	Outer Rim Garrison	+2.8 pp	+1.0 to +4.5
3	Hunter — Everyone Get to Cover!	+2.5 pp	+0.8 to +4.3
4	Ryyk Blademaster	+2.5 pp	+0.7 to +4.2
5	Kelnacca — Solitary Master	+2.1 pp	+0.3 to +4.0
6	Banking Clan Warship	+1.6 pp	-0.1 to +3.3
7	Rex — Outserved His Purpose	+1.4 pp	-0.3 to +3.0
8	Emerie Karr — For Your Own Good	+1.4 pp	-0.4 to +3.1
9	Pelta Relief Frigate	+1.2 pp	-0.4 to +2.9
10	Wookiee Protector	+1.2 pp	-0.6 to +2.9

Tarfful + Cunning55.4%

#	Card	Win rate change	95% interval
1	Ryyk Blademaster	+2.8 pp	+1.0 to +4.5
2	Phee Genoa — Liberator of Ancient Wonders	+1.5 pp	-0.2 to +3.2
3	Outer Rim Garrison	+1.1 pp	-0.6 to +2.8
4	Pack Guardian	+0.8 pp	-0.8 to +2.5
5	Tech — I Thought It Was Obvious	+0.8 pp	-0.9 to +2.5
6	Emerie Karr — For Your Own Good	+0.7 pp	-1.0 to +2.4
7	Luminara Unduli — Besieged General	+0.6 pp	-1.1 to +2.3
8	Auzituck Avenger	+0.6 pp	-1.1 to +2.2
9	Rex — Outserved His Purpose	+0.4 pp	-1.1 to +1.9
10	Banking Clan Warship	+0.3 pp	-1.3 to +2.0

Tarfful + Aggression54.2%

#	Card	Win rate change	95% interval
1	Ezra Bridger — What Are You Afraid Of?	+4.4 pp	+2.8 to +6.0
2	Krrsantan — Santo	+3.2 pp	+1.5 to +4.9
3	Ryyk Blademaster	+3.1 pp	+1.4 to +4.8
4	Wookiee Rangers	+2.9 pp	+1.2 to +4.6
5	Emerie Karr — For Your Own Good	+2.8 pp	+1.2 to +4.4
6	Arena Reek — Fixture of Ceremony	+2.5 pp	+1.0 to +4.1
7	Mining Guild Trespasser	+2.4 pp	+0.7 to +4.0
8	Dune Sea Nomads	+2.0 pp	+0.4 to +3.7
9	Ima-Gun Di — Holding Out For Resupply	+1.7 pp	+0.1 to +3.4
10	Crosshair — I've Changed	+1.5 pp	-0.1 to +3.1

Doctor Hemlock + Command53.3%

#	Card	Win rate change	95% interval
1	Stormtrooper Patrol	+5.9 pp	+4.1 to +7.7
2	Kelnacca — Solitary Master	+4.2 pp	+2.4 to +6.1
3	Nuvo Vindi — Blue Shadow Perfected	+3.7 pp	+1.9 to +5.4
4	Wookiee Rangers	+3.7 pp	+1.9 to +5.4
5	Scorch — Imperial Commando	+3.5 pp	+1.7 to +5.3
6	Arena Reek — Fixture of Ceremony	+2.7 pp	+1.0 to +4.4
7	Outer Rim Garrison	+2.7 pp	+0.9 to +4.6
8	Villainous Ambition	+2.7 pp	+0.9 to +4.5
9	Secessionist Convert	+2.7 pp	+0.9 to +4.4
10	Emerie Karr — For Your Own Good	+2.5 pp	+0.8 to +4.3

Darth Sidious + Cunning51.0%

#	Card	Win rate change	95% interval
1	Third Sister — Cycle of Vengeance	+0.8 pp	-0.9 to +2.5
2	Filthy Dianoga	+0.4 pp	-1.2 to +2.1
3	Sandtrooper Squad	+0.3 pp	-1.4 to +2.0
4	Inferno Squad — We Can Grieve Later	+0.3 pp	-1.4 to +2.0
5	Howl	+0.1 pp	-1.6 to +1.8
6	General Grievous — Scourge of Dathomir	-0.1 pp	-1.8 to +1.5
7	A'Koba — Restless Raider	-0.3 pp	-1.9 to +1.4
8	Tusken Raider	-0.3 pp	-2.0 to +1.4
9	Arena Acklay — Screaming Predator	-0.5 pp	-2.1 to +1.2
10	Teeka — You're In Luck	-0.6 pp	-2.3 to +1.1

Darth Vader + Vigilance51.0%

#	Card	Win rate change	95% interval
1	Emperor Palpatine — Consolidating Power	+3.2 pp	+1.5 to +5.0
2	Stormtrooper Patrol	+3.0 pp	+1.3 to +4.7
3	Kelnacca — Solitary Master	+2.6 pp	+0.8 to +4.4
4	Nuvo Vindi — Blue Shadow Perfected	+2.5 pp	+0.8 to +4.2
5	Outer Rim Garrison	+2.2 pp	+0.5 to +3.9
6	Emerie Karr — For Your Own Good	+2.1 pp	+0.4 to +3.8
7	Nute Gunray — Perfectly Legal	+1.4 pp	-0.3 to +3.1
8	Numa — Still Fighting	+1.3 pp	-0.4 to +3.0
9	Batcher — Loyal Hound	+1.2 pp	-0.4 to +2.9
10	Territorial Mudhorn	+1.1 pp	-0.6 to +2.8

Doctor Hemlock + Cunning50.2%

#	Card	Win rate change	95% interval
1	Mae — Kill the Dream	+9.2 pp	+7.4 to +11.1
2	Nuvo Vindi — Blue Shadow Perfected	+5.3 pp	+3.5 to +7.1
3	Territorial Mudhorn	+4.6 pp	+2.8 to +6.4
4	Scorch — Imperial Commando	+4.5 pp	+2.8 to +6.3
5	Inferno Squad — We Can Grieve Later	+3.6 pp	+1.9 to +5.4
6	Numa — Still Fighting	+3.4 pp	+1.6 to +5.2
7	Treacherous Pyke	+3.2 pp	+1.4 to +5.0
8	Qimir — Everyone Has a Weakness	+3.0 pp	+1.1 to +4.8
9	Rho Medical Shuttle	+2.6 pp	+0.9 to +4.4
10	Occupation Officer	+2.5 pp	+0.7 to +4.3

Bossk + Command50.0%

#	Card	Win rate change	95% interval
1	Inferno Squad — We Can Grieve Later	+4.0 pp	+2.2 to +5.7
2	Nute Gunray — Perfectly Legal	+3.6 pp	+1.9 to +5.3
3	Stormtrooper Patrol	+3.5 pp	+1.8 to +5.2
4	Qimir — Everyone Has a Weakness	+3.4 pp	+1.6 to +5.1
5	Emerie Karr — For Your Own Good	+3.1 pp	+1.5 to +4.8
6	The First Legion — Vader's Fist	+2.8 pp	+1.1 to +4.4
7	Arena Reek — Fixture of Ceremony	+2.6 pp	+0.9 to +4.4
8	Treacherous Pyke	+2.4 pp	+0.7 to +4.1
9	Occupation Officer	+2.4 pp	+0.7 to +4.0
10	Surveillance Cruiser	+2.2 pp	+0.6 to +3.9

Doctor Hemlock + Aggression49.9%

#	Card	Win rate change	95% interval
1	Scorch — Imperial Commando	+2.7 pp	+0.8 to +4.5
2	Mother Talzin — Pledged to the Sisterhood	+2.6 pp	+0.7 to +4.5
3	Territorial Mudhorn	+2.1 pp	+0.3 to +3.8
4	Nuvo Vindi — Blue Shadow Perfected	+1.5 pp	-0.3 to +3.3
5	Imperial Venator	+1.4 pp	-0.4 to +3.1
6	Numa — Still Fighting	+1.2 pp	-0.6 to +3.0
7	Filthy Dianoga	+0.8 pp	-1.0 to +2.6
8	Dooku — Corruption Must Be Eradicated	+0.8 pp	-1.0 to +2.5
9	Mining Guild Trespasser	+0.5 pp	-1.2 to +2.3
10	Devotion	+0.1 pp	-1.6 to +1.8

Bossk + Vigilance48.6%

#	Card	Win rate change	95% interval
1	Mae — Kill the Dream	+6.4 pp	+4.5 to +8.2
2	Nuvo Vindi — Blue Shadow Perfected	+3.8 pp	+2.0 to +5.5
3	Inferno Squad — We Can Grieve Later	+3.7 pp	+1.9 to +5.6
4	Dragonboat Freighter	+3.5 pp	+1.7 to +5.4
5	Qimir — Everyone Has a Weakness	+2.5 pp	+0.7 to +4.3
6	Villainous Ambition	+1.9 pp	+0.2 to +3.7
7	Rho Medical Shuttle	+1.7 pp	+0.0 to +3.5
8	Cid Scaleback — Can't Be Trusted	+1.5 pp	-0.1 to +3.2
9	Surveillance Cruiser	+1.5 pp	-0.3 to +3.2
10	Occupation Officer	+1.3 pp	-0.5 to +3.0

Chewbacca + Cunning47.4%

#	Card	Win rate change	95% interval
1	Greater Sarlacc	+3.2 pp	+1.4 to +5.1
2	Ryyk Blademaster	+2.6 pp	+0.9 to +4.3
3	Wookiee Rangers	+2.5 pp	+0.8 to +4.2
4	Arena Reek — Fixture of Ceremony	+2.4 pp	+0.8 to +4.1
5	Commander Gree — Of the 41st Elite Corps	+2.4 pp	+0.7 to +4.1
6	Phee Genoa — Liberator of Ancient Wonders	+1.8 pp	+0.1 to +3.5
7	Tech — I Thought It Was Obvious	+1.5 pp	-0.2 to +3.2
8	Dragonboat Freighter	+1.3 pp	-0.5 to +3.1
9	Outer Rim Garrison	+1.2 pp	-0.5 to +2.9
10	Hijacked AT-ST	+1.0 pp	-0.5 to +2.6

General Grievous + Cunning47.3%

#	Card	Win rate change	95% interval
1	Qimir — Everyone Has a Weakness	+3.4 pp	+1.7 to +5.1
2	Emperor Palpatine — Consolidating Power	+2.7 pp	+1.0 to +4.4
3	Arena Reek — Fixture of Ceremony	+2.5 pp	+0.8 to +4.3
4	Nute Gunray — Perfectly Legal	+2.4 pp	+0.6 to +4.1
5	Stormtrooper Patrol	+2.1 pp	+0.4 to +3.8
6	Emerie Karr — For Your Own Good	+1.5 pp	-0.2 to +3.1
7	Inferno Squad — We Can Grieve Later	+1.4 pp	-0.3 to +3.1
8	Geonosian Picador	+1.4 pp	-0.3 to +3.0
9	Sandtrooper Squad	+1.3 pp	-0.3 to +3.0
10	Disposable B1	+1.3 pp	-0.4 to +2.9

Bossk + Aggression47.0%

#	Card	Win rate change	95% interval
1	Enraged	+2.3 pp	+0.4 to +4.1
2	Inferno Squad — We Can Grieve Later	+2.2 pp	+0.4 to +3.9
3	Qimir — Everyone Has a Weakness	+1.8 pp	-0.0 to +3.6
4	Filthy Dianoga	+1.7 pp	+0.0 to +3.4
5	Adamant Ewoks	+1.6 pp	-0.2 to +3.3
6	Arena Nexu — Starved For Prey	+1.6 pp	-0.2 to +3.4
7	Mining Guild Trespasser	+1.4 pp	-0.3 to +3.1
8	Tusken Raider	+1.4 pp	-0.4 to +3.2
9	Cid Scaleback — Can't Be Trusted	+1.3 pp	-0.5 to +3.0
10	Occupation Officer	+1.2 pp	-0.6 to +3.0

Wicket + Command45.7%

#	Card	Win rate change	95% interval
1	Emerie Karr — For Your Own Good	+3.0 pp	+1.3 to +4.8
2	Enraged	+2.7 pp	+0.9 to +4.5
3	Leia Organa — These Are My Friends	+2.4 pp	+0.7 to +4.1
4	Arena Reek — Fixture of Ceremony	+2.2 pp	+0.5 to +3.9
5	Adamant Ewoks	+2.2 pp	+0.5 to +3.9
6	Captain Tarpals — Grand Army Captain	+1.9 pp	+0.1 to +3.7
7	Verdant Fortress	+1.8 pp	+0.1 to +3.6
8	Teebo — Striped Hunter	+1.7 pp	-0.0 to +3.4
9	Champions of Endor	+1.7 pp	-0.1 to +3.5
10	Child of Dathomir	+1.5 pp	-0.3 to +3.2

Jar Jar Binks + Cunning44.5%

#	Card	Win rate change	95% interval
1	Boss Nass — Otoh Gunga Boss	+4.1 pp	+2.4 to +5.8
2	Sol — Compassionate Guardian	+3.3 pp	+1.6 to +5.1
3	Yoda — Trickster In Exile	+3.2 pp	+1.5 to +4.9
4	Numa — Still Fighting	+3.0 pp	+1.4 to +4.7
5	Grand Army Marine	+2.5 pp	+0.9 to +4.2
6	Yord Fandar — Devoutly Disciplined	+2.5 pp	+0.8 to +4.2
7	Phee Genoa — Liberator of Ancient Wonders	+2.2 pp	+0.6 to +3.8
8	Territorial Mudhorn	+2.1 pp	+0.4 to +3.8
9	Howl	+1.9 pp	+0.3 to +3.6
10	Captain Tarpals — Grand Army Captain	+1.7 pp	+0.0 to +3.4

Highlights

• Emerie Karr makes the top ten for eight of the fifteen archetypes, and Arena Reek and Inferno Squad for six each.
• Mae adds nine points to Doctor Hemlock + Cunning, the biggest effect of any card measured, and six to Bossk + Vigilance.
• Bestial Bond adds five points to Tarfful + Vigilance.

Takeaways

• If you open Tarfful, he's a very strong choice, and any of his three base colors can work. Vigilance is the strongest.
• Without Tarfful, Doctor Hemlock on a Command base is the next best thing, and the rare Darth Vader and Darth Sidious archetypes are worth building when your pool has them.
• Chewbacca is a good fallback if you don't get one of the better leaders. With good Cunning pulls, he is a solid choice.
• Avoid Tarkin, Maz, Omega, Cham, Maul, and Osha if you have anything better. They are just not very good in this limited meta.

Unfortunately, due to model training issues, this experiment wasn't ready to publish until after Prerelease events had mostly completed.

Questions about our findings? Email admin@swoolab.com.

Try sealed against the model