Why our Coguire player sometimes buys nothing

Richard here. I work on co-gas. An agent helped me trace the current planner, authoritative mechanics, and completed game evidence for this post, and I reviewed it before posting.

This covers Coguire 0.1.9, the four-player standard variant, and the co-gas marginal-value family. Coguire is an Acquire-style game on a 12-by-9 tile board. Players found seven hotel chains, grow or merge them, buy shares, and finish with the highest net worth: cash plus stock valued at the final chain prices.

The bot is phase-based. It handles tile placement, chain founding, tied-survivor choice, stock disposal during a merger, and the buy phase separately. The public view includes every player's cash and holdings, current chain sizes and prices, our six-tile hand, and the connected board.

The buy decision is the easiest part to summarize:

enumerate every affordable mix of zero to three shares
for each chain, compute the majority/minority payout before and after
add price growth supported by tiles in our own hand
subtract cash cost and the strongest rival's payout gain
buy the highest positive-value mix; buying nothing is always allowed

That last option is important. An older version picked a favorite chain once and kept buying the maximum count. In one completed game it reached a public 13-to-1 lead in Imperial, spent its last $1,800 on three more shares that changed no bonus position, then made nine empty buys with zero cash. The current scorer values only the marginal effect of each proposed share. A share that changes majority or minority standing can be valuable; the eleventh redundant share may not be.

Tile placement uses the same relative-value idea. The bot first classifies the full orthogonally connected component. A tile may stand alone, found a chain, grow one chain, or merge several. It refuses a dead merger between safe chains and avoids trying to found an eighth active chain. Growth is scored as our stock appreciation minus 0.75 times the best rival's appreciation, plus the change in shareholder-bonus edge. A merger also includes bonuses from defunct chains and appreciation in the survivor.

Three rule details shaped the implementation:

• Chain prices use exact size brackets. Sizes 6–10 share one base price, followed by 11–20, 21–30, 31–40, and 41-plus brackets. The chain's cheap, medium, or expensive tier adds $0, $100, or $200.
• Majority and minority bonuses are 10 and 5 times the share price. Ties split the applicable pool and round each payment up to the next $100, so one share can create a discontinuous payout jump.
• In merger disposal, trade means survivor shares received. Each costs two defunct shares. Sending the number of surrendered shares instead makes sell + 2*trade exceed the holding and loses the whole decision.

The active V12 version was tested in four completed seat rotations. It averaged $36,075 final worth, versus $20,575 for its predecessor and $30,900 for the other owned control, while finishing 3.5% below the strongest public control. All 153 actions were accepted with no fallback. That evidence supported replacing the older overbuying version.

The checked-in source also keeps a $400 cash reserve, enough for one maximum-price opening share after founding a chain. In eight later hosted games that reserve raised average worth to $39,175 versus V12's $35,700, but tied V12 four games each and slightly worsened average placement from 2.25 to 2.375. The founder-specific release never activated, so V12 remained the established hosted version. We also tried forcing that founder share; it lost V12 five games to three and averaged $5,150 less.

The main limit is future information. We can price growth from our visible hand, but not from opponents' hands or the unseen tile bag. Our certified model exactly reproduced 2,349 buy transitions; it does not yet certify placement, merger, disposal, or final liquidation. The next useful improvement should start with a completed counterexample in one of those phases, then add the missing payoff term without disturbing the buy accounting.

How do you value one share that creates a majority tie versus keeping the cash liquid?

When is growing a chain bad because it rewards a rival more than you?

What board feature best predicts whether a merger will happen soon?

How much cash do you reserve when your own hand contains no clear growth tile?