Manual Draft or Auto-Deal in CONQUEST: What to Claim

Use a manual draft if you want a say in your opening, and claim a complete region first. In 2,400 auto-deal starts on each of three maps, almost nobody began with a whole region. Drafting four Hard bots gave each player 0.72 whole regions on World, 1.01 on Europe and 1.37 on USA.

The claim rule that worked is simple: finish the region with the most bonus per territory that no rival has touched. It beat the bot's own drafting by up to 15 points on USA and 9 on Europe. On the World map nothing we tried helped.

CONQUEST is a RISK-style game and is not affiliated with Hasbro, the maker of RISK. The numbers below describe CONQUEST's rules and its bots, not the board game and not human opponents.

Auto-deal never hands out a region

Auto-deal shuffles the territory list and deals it round-robin, then scatters your leftover starting armies at random across what you got. Nothing in that process looks at regions, and a region bonus only pays when you hold every territory in it.

We counted what each player owned before the first attack: 600 four-player games per row, Hard bots, using the real engine from the game. The table shows the average per player.

Map	Setup	Whole regions	Biggest connected block	Players with a whole region
World (42 territories)	Auto-deal	0.00	3.6	0%
World	Draft	0.72	7.2	71%
Europe (27)	Auto-deal	0.01	3.1	1%
Europe	Draft	1.01	5.5	100%
USA (39)	Auto-deal	0.01	3.9	1%
USA	Draft	1.37	7.7	100%

The second number matters as much as the first. A dealt hand is a scatter of islands: the biggest connected group averaged 3.1 to 3.9 territories. A drafted hand averaged 5.5 to 7.7, which means your armies can actually move to where the fighting is. The Fortify move only works through territories you own in a connected line, so connection is what makes a reinforcement reach the border.

The World map guide covers which continents are worth taking once the fighting starts. A draft decides how much of that you already hold on turn 1.

The board game drafts too, one army at a time

Classic RISK has the same idea. Hasbro's own game guide (the RISK with Two-Headed Dragon edition, Classic RISK setup) says that after a die roll, the highest roller places one Infantry on any territory, and players then take turns placing one Infantry on any unoccupied territory until all 42 are claimed. After that, each player adds one extra Infantry at a time to territories they already hold. Whoever placed the first troop takes the first turn.

CONQUEST keeps the shape and changes two details. The starting armies match the guide for 3 to 5 players (35, 30 and 25). A claim costs one army, as it does in the board game, but once the board is full you place the rest up to three at a time, not one. Seat 1 picks first and also moves first, as in the board game.

A published research result points the same direction. A University of Alberta team built a drafting bot for the RISK clone Lux Delux and reported that territory drafting is important to overall success. That paper is about their own clone and its bots, so treat it as support for the question and not as a ranking of picks.

Claim the cleanest region, not the biggest

We tested four ways to claim territories, with one player seated among three Hard bots. Only the claim phase changed. The hero's armies after the board filled, and every turn after the draft, were played by the Hard bot brain. Each cell is 600 games with Wild Dice, and a fair share is 25%.

• default: the Hard bot's own claim logic, which weighs how many of a region it already owns.
• random: any free territory.
• cleanRegion: pick inside the region with the highest bonus per territory among regions no rival has touched, and keep going until it is whole.
• blockRival: pick inside the region where a rival already holds the largest share, weighted by the region's bonus, to deny it.
• hubs: pick the free territory with the most neighbors, ignoring regions.

Map	Seat	default	random	cleanRegion	blockRival	hubs
World	first	13%	18%	15%	13%	9%
World	last	21%	20%	15%	12%	9%
Europe	first	20%	20%	29%	22%	15%
Europe	last	32%	28%	41%	28%	20%
USA	first	18%	19%	33%	21%	13%
USA	last	38%	28%	44%	23%	17%

One cell of 600 games carries roughly 3.5 points of error either way. The gaps that matter are the big ones: cleanRegion over the default is +9 on Europe in both seats and +15 on USA in the first seat. On the USA last seat it is +6, which is real but smaller.

Two other results are as clear. Denying rivals loses ground: blockRival fell from 38% to 23% in the USA last seat. And chasing hubs is the worst plan on every row, down to 9% on World. Territory count and neighbor count do not pay a bonus. A finished region does.

Why World is different

On World the same rule did nothing, and it even lost 6 points in the last seat. We did not isolate the cause. What the rule does there is visible in the region table: the best bonus per territory is Europe at 5 for 7 territories (0.71), so the rule heads for a 7-territory region with 4 border territories. On Europe and USA the best ratio is 3 for 4 territories (0.75), with regions that are cheaper to finish.

Map	Best bonus per territory	Region size
World	Europe, 5 bonus (0.71)	7 territories
Europe	Scandinavia and Western Europe, 3 bonus (0.75)	4 territories each
USA	Great Lakes, Northeast and Southeast, 3 bonus (0.75)	4 territories each

A sensible reading: on a map with 4-territory regions, fill one early and you start ahead of the Hard bot's own drafting. On World, the small regions (South America and Oceania, 4 territories each, bonus 2) pay less, and the bot's own logic was already as good as anything we tried.

Seat 1 is the weak seat in a four-player draft

With auto-deal and four equal Hard bots, seat 1 won 24.8% to 27.0% of games on these three maps, close to a fair 25%. With a draft it won 11.3% to 19.0%.

Map	Setup	Seat 1	Seat 2	Seat 3	Seat 4
World	Auto-deal	27.0%	25.6%	20.9%	26.6%
World	Draft	11.3%	37.6%	30.5%	20.5%
Europe	Auto-deal	26.1%	21.0%	25.4%	27.5%
Europe	Draft	16.7%	23.0%	28.0%	32.4%
USA	Auto-deal	24.8%	24.5%	26.2%	24.5%
USA	Draft	19.0%	16.6%	31.9%	32.5%

That is 1,000 games per row. It runs against the two-player result in our first-player advantage post, where seat 1 won 92.4% of drafted two-player games. We did not isolate why the sign flips with four players. One mechanical fact in the code: the draft rotates through the seats, so the last seat places its final armies after seeing everyone else's, and seat 1 then moves first with its position already visible. Take that as an observation of the code, not a proven cause.

Bots are not people, so treat this as a warning and not a law: do not assume that picking first is an edge in a four-player draft. Seats follow lobby order, as the first-player post explains, so the host is seat 1.

Practical details of a draft

• It takes more turns than you think. On the World map each player takes 12 to 28 draft turns: you claim your territories one at a time, then place the remaining starting armies three per turn. At four players that is 17 to 18 turns.
• Every pick is on the turn timer. The lobby timer is 30, 60, 90 or 120 seconds. At the 30 second setting, 17 draft turns is at most 8.5 minutes of your own thinking. If the timer runs out, the server makes the pick for you using a Medium-level bot, and a second miss hands your seat to a bot until you return.
• In Capitals mode your first claim is your capital. The engine marks your first drafted territory as your capital, so your first pick is also the one you must protect. How Capitals mode works explains the win condition.
• The second phase matters. After the board fills, three armies per turn is a slow way to build a wall. Our test left that to the Hard bot, which weighs threat from neighbors. A sensible placement is on the borders of your region, since an army in the middle cannot attack anyone.

Everything above used Hard bots. If you play against other levels, the bot difficulty post shows how each one plays, and Medium, Hard and God bots claim with the same logic in the game code, and Easy bots claim at random. You can switch the setup to Manual draft in the quick play setup or the lobby, and the how to play page lists the other options.

How we got the numbers

The script is scripts/research/conquest-draft.mjs in the blog repository. It imports the real engine (shared/engine.js) and the real bot brains (shared/bots.js) and plays bot-only games. The first table is 600 games per row, the claim-strategy table is 600 games per cell, and the seat table is 1,000 games per row. The bots use random choices, so a rerun moves each figure by a few points. The heroes in the strategy test are bots with one rule swapped, so the results describe the engine's rules, not human play.