Depending on the time of the day you play splinterlands, you are more or less likely to face a dastardly thinking machine known to man as a Splinterlands liquidity bot. These bots, while necessary for match liquidity, can be a source of frustration one way or another.
For a long time they fielded questionable line-ups, causing rating inflation with free wins. Now they are fine, sometimes. They more often than not field a reasonable line up you could face in whichever league you play. But then there are also the times when they flex on you with line ups that would cost hundreds if not thousands of USD. Especially in survival mode players are at a disadvantage as bots don't really lose cards and can, if you are unlucky enough, come back with the dreaded line ups just after they lost the majority of it.
But enough of the current situation, lets look at solutions. In my opinion the problem described above stems from bots not using a collection, but have all the cards at their disposal. To counter this, I suggest to switch bots from all ghost cards to a budgeted collection of cards, lets call it a ghost collection.
Now lets assume we don't want to train bots to pick and choose their cards based on their market value, the easiest way to do this would be to copy collection of active players as ghost collections. This doesn't really need to be kept up to date, a simple snapshot at the start of the season would suffice.
So let's say you are queueing in modern gold. You expect to meet a player with a collection you would typically see at this rating, maybe some silver and bronze level cards, certainly not all max-level gold cards. The game would then provide you with a match vs a liquidity bot equipped with a random modern gold players collection. This would more or less guarantee that the liquidity bot cards are not too distinguishable from those of a human opponent. This should allow for fairer matches vs bots as you are much more likely to encounter an average line-up and not the uber-premium deluxe edition almost nobody in the league has. The same applies for wild, survival, the format is irrelevant.
Since this demands a change by development, we should also look at the effort estimation. Of course, not being part of the team or knowing the code these are only assumptions, but I don't think this should be a big problem to implement. I would assume collections can be easily stored as their are just arrays of card descriptions, I guess a copy of player collections is stored locally for performance reasons, so taking a snapshot should be a simple issue. I would at least hope that liquidity bots are based on a bot that can process an existing collection and choose based on whatever knowledge-based algorithm the bot uses which cards it fields in which situation. Someone from the team, please correct me if my assumptions are wrong.