Resdesigning RCI

[ptarth]

• I had missed your RCI.
• It looks nice.
• My wishlist of changes
• plot RCI by year (for 100 years, updated per month
• Add buildings with, start date, end date, onset, offset, recurring fields
• Add events with, start date, end date, onset, offset, recurring fields
• In retrospect specific types of events aren't necessary, just have a generic one and we can label it whatever.
• Add formula variants that take a given RCI and apply equation X (i.e., rescaling)
• Add formula variants that manipulate RCI gains (i.e., recalculating)
• I should also probably talk about my rough current RCI changes.
• Balance RCI on min/max of -100/+100, 0 neutral
• Plan balancing around Technologically advanced races will have an RCI of 70 in 100 years.
• 100 years seems to be the game length that Arcen was aiming for as being the maximum length of 90% of all games
• 70 is high enough to "feel" large, but low enough that extra growth can still be handled.
• Make all RCIs have trends all the time. Increase trend size to +/-0.25 per month, and have a 2% chance of switching per month.
• This is give some variance in RCI which give trends a purpose.
• Models current systems that do have yearly cycles.
• Trend size will be discussed later, roughly 25% of a building
• % chance of switching should be set to model shifts based on X year cycles. This was a ballpark figure.
• Reduce maximum impact of Events to 20 for small events, 30 for medium events, 50 for major events
• Current events drop RCI by hundreds & thousands, ~-1 per month for 100 months is common
• Knocking it down to 33%, 50%, and 75% of normal max RCI  seems like a reasonable starting place.
• RCI buildings have a maximum count of 1 (or divide all following values by the max number)
• RCI buildings currently vary between having 1 to 5 maximum counts.
• There doesn't seem to be a reason.
• In my mind, this represents the proportion of the planet that has ready access to this type of building.
• As such 1/max = % of planet covered. Sticking with a constant max = 1 or 5 seems reasonable and straightforward.
• Increase building effects to +10 one time increase and +0.1 per month
• There are K buildings (and technologiest) that effect RCI Z. However, they are accessible at different years.
• Do a summed weighted effect by building up to year 100 for fixed effect L and recurring effect M. Set it equal to 70.
• Solve for M, L. Find out K.
• I ballparked +10 and +.1. It isn't right and it needs to vary per RCI.
• Give Hydraal player the ability to reduce RCI by 1 of a race. Do not give them a reputation or relationship penalty for it.
• Allows the player to manipulate RCI is both directions without resorting to Racial Specials.
• The impact is now reasonable compared to events, but it is more cost effective for the player to build them buildings.
• Space Batman doesn't advertise his sabotage.
• If Space Batman advertised, soon Space Batman would be shot on sight and not allowed to spend 10 months handing out "medicine".
• Scaling
• I played around with NichG's simulator and with my own code and found that scaling was unsatisfying.
• If I did something that caused an impact of 10, I wanted to see the RCI change by 10. Not by some weighted function of 10.
• I also didn't want to see decay happening, showing that I just wasted all my effort on working on a dispatch mission.
• Giving buildings minor recurring boosts and a larger one time boost produced the effect of slowly accelerating RCI growth.
• I think that if Racial/Planetary AI is adding allowing the Races options to sacrifice RCI for other things, RCI will end up even more natural and reasonable looking.

[NichG]

What might be useful would be if you could make a 'faked' (e.g. hand-drawn) graph of what you see as a satisfying RCI pattern over the course of a 100 year game. That'd give me some idea of how to induce that kind of structure mathematically.

Question: for the unsatisfying feeling of the 'weighted effect', would it resolve the problem if the game told you what the change would be after the weighting? E.g. 'this action will adjust the RCI by +9.3' rather than 'this action will adjust the RCI by +10', even if that means that actions will not always have constant impact?

Other random comment: If decay is particularly unsatisfying to observe, that suggests to me that the best way for the player to interact with RCI would be to limit their modifications to permanent adjustments of the baseline (e.g. buildings and technology), rather than dispatches (also in light of the comments in the other thread about removing RCI dispatches, I think its a good idea to include that in the analysis)

Update on the state of the code: I've nearly got things set up so you should be able to choose between a bunch of different models for the different cases. I've had to put in some models for things like the budget and the feedback between RCI and population growth/death, and there are some pretty arbitrary choices I've made there. One thing that I've included because I think its worth considering is to generally treat the effect of RCI as exponential on various processes. For example, the budget model I'm using says that each 1 billion population generates resources equal to exp(C*Economy/100). That means that positive RCI and negative RCI proportionately have the same effect - e.g. if C=2 then +100 multiplies economic production by 7 and -100 divides it by 7. I've added a switch to toggle between that and the linear model. The GUI part of this is going to be a little bit of a slog, to make it fairly responsive to trying out different combinations and to plot things reasonably.

The result is going to be a bit big for a JSFiddle I'm afraid. Would a GitHub repository be something you could work with if I were to put up the code/interactive page on one?

[ptarth]

• re:example - I have some code generated RCI examples. I'll upload some pictures when my computer is done with its current job.
• re:weighted effect - I tried that on my coded version and found that I got annoyed that the effects of things changed. One month I was +2 RCI the next it was +1.4, but nothing changed.
• re:decay - Fixed one-time changes are equivalent to permanent baseline adjustments. That's how I've been modelling it, and it feels right.
• re:code - I've done most of my work in R on my local machine. I like the idea of nonlinear effects of RCI, just not nonlinear changes to RCI.
• re:gui - for my own work, I've never bothered with a gui and just hard-coded everything.
• GitHub - Whichever.

[ptarth]

Example RCI Medical Simulation Event Log

#Event History
#Year 0 - Starting point 0, trend - 1/
#Year 10 - Building +10 +2.5/
#Year 20 - Building -5 -2.5/trend changes to +2.5/
#Year 30 - Disaster -20 - 5/
#Year 35 - Year 10 Building lost
#Year 75 - Hydraal assists +10/
#Year 100 - Disaster ends
#Year 110 - Building +10 +2.5/
#Year 120 - Technology +10 +2.5/
#Year 130 - Building +10 +2.5/
#Year 140 - Building +10 +2.5/

The black line is the linear model. The green is the sigmoid. Currently I think that the linear model is fine, as long as we keep the RCI manipulations reasonable.

[NichG]

I want to be careful about the GUI in particular because its clear that presentation has a large effect on how a given model 'feels' to play with. E.g. you seem to be bothered when things are presented in a way that is inconsistent with the effect you expect - either the effect you expect due to being directly informed (RCI will go up by 2, but goes up by 1.5) or effects that you expected due to past exposure to those actions (this action gained +2 RCI last month, but this month it just gains +1). So picking out the right way to inform the player about consequences is, I think, going to be key to making something that both feels right to play and at the same time has good stability/plasticity/etc properties.

It may well be that it makes sense to e.g. use nonlinear RCI updates for 'hidden' degrees of freedom in the simulation, but use linear ones for whenever the player pokes things directly because of this presentation issue.

One thing I want to play with in particular is how it impacts the feel if the player has access to projections. In other words, imagine you have a button that just says 'develop Economic technology' and when you click on it, it gives you a graph with two lines - one being the Economic RCI trend line extended 20 years into the future as things stand, and the other being the Economic RCI trend line that you would obtain given the player action (assuming no random events/influences between now and then). It then gives you a button that says 'commit' and a button that says 'cancel'. So that way, the player can be fully aware of the consequences of particular actions within the context of an underlying model that has hidden degrees of freedom that can make things murkier (like e.g. what you were finding with the actual RCI change values in game compared to what you expected).

Edit:

For your example graph, is that the kind of curve you want, or just what the model produced? I'm saying, if you give me a shape I can probably come up with the math that generates that sort of shape on average, so don't worry about being too constrained by the specific models.

[ptarth]

• The graph I provided uses the events I listed to produce the curves.
• I believe that sort of plot is most useful for getting a feel for what changes in RCI mechanics do.
• It has nothing to do with the actual mechanics, just with the presentation of the data.
• Being model agnostic it also lets us compare models of RCI in identical circumstances to see which seems more right.
• I wrote a longer post using when I created the model, but tossed it because it wasn't useful. Thus I might not realize I didn't explain anything.
• The nonlinear model is Reported.RCS = (200/(1+exp(-1*Actual.RCS/100))-100)
• The linear model is purely additive and unbounded.
• I attempted to add your model, but I wasn't sure what unit you wanted to integrate over, so I gave up in the end.

[NichG]

Hm, I think we're miscommunicating here.

What I'm saying is, putting aside a specific mathematical model of RCI, what should an RCI trace from a 'good' game look like? I.e. if we ignore how it was generated and just think of what kinds of results would be satisfactory.

For example, I think I'd generally want something like the image attached as my ideal RCI behavior.

- It does something without the player's direct intervention and occasionally goes on excursions but doesn't get lost in the boonies.
- When the player intervenes, they can change the resting point through extended effort, but they're not chasing the details around.
- Disasters are fundamentally short-term severe periods, with effects that are eventually healed back to baseline automatically at long times.
- Over the course of the game, RCI slowly tends towards more extreme results (extreme negative would involve e.g. a planet choosing to sacrifice Environment for Economy on the long term).

[ptarth]

• I agree. That's what in my event log and the plot.
• The scale though is dominated by the comparison of the linear and nonlinear models.
• I don't think disasters should "heal" automatically. "healing" should be a product of effort (by Racial AI or player).

[NichG]

Okay, so I think in terms of design at least, the major point of actual contention is disasters. I'd prefer them to heal because that means that the structure of RCI in the solar system is dominated more by intentional things, and can't just be permanently wiped out in one blow by the whim of the random number generator. I'd rather disasters be something to deal with, rather than something that says to the player 'well I should just not bother with RCI because I can't do anything to protect it'.

Also, since random events tend to be more negative than positive, this means that the eventual fate of a planet's RCI in any sufficiently long game is basically to go deeply negative.

I'd suggest that to resolve this, we look more carefully at what the design goal of disasters is/should be, and what is actually achieved by the various ways they could be implemented.

1. Is the point of disasters to give the player a short-term strategic curve ball that they have to react to in order to regain equilibrium? E.g. 'I want the Acutians to invade - crap, their economy just got smashed, I have to repair it or wait for it to recover before I can move forward with my plans, or find someone else for the time being'
2. Is the point of disasters to give the player something they have to work to fix? E.g. 'Okay, the Peltians just got the super-bug. If I don't get on this, it will spread and decimate the entire solar system. This is my priority for the next 10 years.'
3. Is the point of disasters to bias planets in a random but persistent way, setting the scene for the player's strategy? E.g. 'Okay, the Andorian planet has been turned into a wasteland, so from now on I can't use them in any strategies involving military force'

I've been using case #1 as my model. Maybe case #2 is more what you're thinking? Hopefully not case #3, since I think that's pretty bad design.

For case #2, I would suggest that if disasters can permanently move the baseline (e.g. they don't eventually heal) then all disasters should message the player with advanced warning before they reach that point and have certain triggers that the player can use to try to interrupt the permanent damage stage. For example:

"Andorians experience global warming from Economic surge - For the next 5 years their Temporary Economy will increase by +1 per month and their Temporary Environment will decrease by -1 per month. If at this point their total Economy RCI is positive, they will experience a permanent -20 shift to their Baseline Environment."

So if the player acts in time and responds correctly, the changes are temporary; otherwise, they're permanent (modulo future events that change the baseline). This prevents disasters from just being an arbitrary strategy-destroying element, while still requiring effort from the player (or racial AI) to prevent permanent damage.

-------------------
The rest of the disagreement appears to be miscommunication.

For your plot, I would say that the linear example is unstable, and passing it through a sigmoid just hides that instability. As the game progresses in your example, the ability of the RCI result to be moved systematically decreases (because the linear thing driving the nonlinear function is off in +600 land, so you'd have to shed 500 points of RCI before you saw it move noticeably and there'd be no real effect from shedding those 500 points). Thats why I've been arguing that the system should use nonlinear changes rather than just a nonlinear filter.

There are other ways to achieve that which would be more development work but which might introduce less confusion (for example, quests unlocking as you get more extreme that produce more extreme changes, so if you're at -500 there's a quest that'd give +300, but if you're at -100 then that quest is not available). Basically though, I really think that this 'running away to +/- infinity' phenomenon is the core of what needs to be removed from RCI, because that's what renders the system untenable later in the game.

[ptarth]

• I agree that we seem to be coming from the same places, but aren't able to communicate well for some reason.
• By reducing the impact of events to the 20-50 range, runaway events are stopped.
• By giving AIs the ability to directly change RCI like the Hydraal Dispatch missions, we can moderate the effects further.
• Putting a cap on the maximum gain provided by dispatch missions may also be reasonable. But I'd want to see it work out first.
• With events being both positive and negative, some of the drift is reduced.
• The recurring benefits from buildings being higher means there is pressure forcing RCI up.
• Isn't the difference between 1,2, & 3 simply the relative size of the effect of the event?
• Baseline in the sense that you are using it, only is meaningful if there is pressure to return to the baseline. Otherwise it is just a point on the scale. I don't think that a fixed baseline is necessary at this juncture. I believe the majority of the work in improving the mechanic can be focused on relative gain balancing and one-time and recurring RCI changes.
• I'd need to see your definition of unstable.
• I do believe the system should push towards positive infinity, being either capped by limiting time, the scale of the additions, or via sigmoidal or log scaling.
• I point to the GNP link I posted earlier as validation, compared to the past, humans production levels continue to rise, perhaps to infinity.
• My example behaves as it did because I shoved in massive parameters to show the difference between the two models. I believe that more reasonable parameters, based on the previously posted logic, should present a more reasonable system.

[NichG]

• Putting a cap on the maximum gain provided by dispatch missions may also be reasonable. But I'd want to see it work out first.

I sort of like a version of this as a way to both have nonlinear large-scale changes but also make every change locally linear. For example, lets say you break down the RCI into a bunch of sub-items. You have 5 negative sub-items that can each be 'worth' up to -20 and 5 positive subitems that can each be 'worth' up to +20.

Any given action of the player or planetary AI would influence these particular sub-items. When you max out a sub-item, the option to push it further goes away. So you know that +1 RCI is always +1 RCI, but you can only push it by 20 points before you have to use a different method. The different sub-items might grow or shrink at different rates when you push them, so that implements the effective nonlinearity.

For example, lets say we have some Environment factors like:

- Recent Nuclear fallout (0 to -20): Repairing occurs at +0.5/month, bombing inflicts a -10 in this category, no worsening action available.
- Global Warming (0 to -20): Repairing occurs at +0.2/month, worsening actions occur at +0.5/month
- Recycling Programs (0 to +20): Improving occurs at +1/month, worsening actions occur at -1/month.
- Terraforming Initiative (-20 to +20): Improving occurs at +0.1/month, worsening actions occur at -0.1/month.

That gives you predictable response, 'you get what it tells you', 'the same action always changes things by the same amount', and also captures the kind of nonlinearity that I've been aiming at.
You can quickly push the RCI around, but when you want to go to the extreme ends everything slows down because there are fewer actions left that can improve/worsen the situation further.

• Isn't the difference between 1,2, & 3 simply the relative size of the effect of the event?

Not at all, it has to do with the timescale that it takes to overcome the event. For example, lets say a disaster strikes in each of the three models that decreases the Economy RCI by 90 points.

Case #1: The effect lasts for, say, a year and then goes away, leaving untouched whatever slow permanent improvements you've been making to the Economy: the disaster is a temporary debuff. Assuming no player involvement, then if you look at the RCI before the disaster and the RCI ten years later, they'll be roughly the same (assuming no tech development, other events, buildings, etc).

Case #2: The effect lasts for a year and then either goes away if you've responded to it and dealt with the problems, or becomes permanent/semi-permanent if you have not. If you deal with it, then two years after the disaster the planet's Economy will be back to where it was before the disaster. If you fail to deal with it, then ten years after the disaster the planet's Economy will still be below what it was before the disaster (assuming no tech development, other events, buildings, etc)

Case #3: The effect lasts for a year, after which it has made a significant change to the planet's Economic RCI, such that more or less no matter what the player does, if you look at it 10 years later then the Economic RCI will be lower than it would have been if the disaster had not occurred.

• Baseline in the sense that you are using it, only is meaningful if there is pressure to return to the baseline. Otherwise it is just a point on the scale. I don't think that a fixed baseline is necessary at this juncture. I believe the majority of the work in improving the mechanic can be focused on relative gain balancing and one-time and recurring RCI changes.

I disagree with this.

• I'd need to see your definition of unstable.
• I do believe the system should push towards positive infinity, being either capped by limiting time, the scale of the additions, or via sigmoidal or log scaling.
• I point to the GNP link I posted earlier as validation, compared to the past, humans production levels continue to rise, perhaps to infinity.

This is in fact my definition of 'unstable' - if you go to infinite time and the system's variables go to infinity. As far as realism, that not really relevant if it leads to bad gameplay. I'd be willing to debate the realism, but I feel like it'd be a monumental and pointless distraction and this thread is already quite scrambled.

Things that grow exponentially have a number of bad characteristics when it comes to delivering a consistent gameplay experience. Exponentials are controlled so strongly by differences in growth rate combined with differences in starting point (when there's non-linear coupling) that often you lose the game before you realize you're playing. Also, once things get underway, the conclusion becomes obvious long before you're actually done playing the game, so you're forced to slog through a 'conquer the rest of the map' stage where you can't lose, but you haven't performed the keystrokes necessary to confirm the win yet. Just look at all the work Civilization has to do to prevent the 'smallpox' strategy from dominating the game (the strategy of settling new cities as quickly as you can produce settlers during the early game). They use all sorts of nonlinear effects like unrest scaling with city count, culture costs increasing with city count, etc in order to suppress that exponential curve.

[ptarth]

• We seem to be going in circles. Let's try something more concrete.
• Attached are two plots.
• They are 1000 simulations of RCI.
• The black is the linear model. The green is the sigmoidal rescaling of the linear model.
• In them, I plot current RCI only being affected by a trend per month for 100 years (10 months per year).
• The trend is a monthly change of +/-0.25.
• Every month there is an independent 2% chance of switching to the opposite trend.
• Mean months between changes is 50 months, from a sample of 100,000 months
• Median: 35, Min: 1, Max 482, SD 50

There are 4 obvious parameters.
1. Trend effect size- Larger trend effect sizes would increase the range of variation.
2. Chance of switching - Larger chances would decrease the range of variation.
3. Scaling - Showing linear and sigmoidal.
4. RCI-dependent delta RCI Scaling - Per nichG's approach.

How do you feel about this version of RCI trend?

[NichG]

Both of these are 'unstable', but the sigmoidal one hides it using a mathematical sleight of hand.

The easiest way to see this would be if you were to plot, as a function of time, the amount of player effort that would be needed to return the RCI value to zero, assuming the player is using a fixed-rate dispatch which doesn't also increase in ability as the RCI grows (e.g. the player can apply +1/month or -1/month for N months). Remember to count only total months spent, not total change in RCI (otherwise the average will be zero because plusses and minuses will cancel).

In both the sigmoidal and linear cases, this quantity will on average grow like sqrt(t). That's the mathematics of a 1D unbounded random walk, which is what you've got here. When t->infinity, that means an infinite amount of effort is required to return the RCI value to zero. That's what I mean by 'unstable'.

Now, you can do a version of this where you have buildings that add/subtract a certain amount per month. It turns out that this is even more unstable, because now the effort required to return through zero scales like t instead of sqrt(t). Even worse is if buildings are being constantly built to go a particular direction, because then the average effort required to return through zero scales like t^2.

My problem with all of these is that they diverge when t->infinity. That means that the game becomes increasingly hard to influence as time goes on. This will remain true regardless of whether you put a sigmoidal filter on it, because the number the player is influencing is the (hidden) linear value, not the sigmoidally filtered value.

[ptarth]

I don't understand why you have a problem with an unstable system. Isn't divergence at infinity a good thing? Otherwise, the player can simply do nothing and the system will proceed to self-correct back to baseline (or at least stay within a standard limit). That is to say if a player ignores a mechanic for 1 month or 100 years, they would have to spend the same effort to adjust the system to put it wherever they want it.

For gameplay purposes, I see divergence as good, the problem being that you don't want it to get too far out of a reasonable zone in 75% (some arbitrary number) of cases. However, if you ignore the collapse of the Peltian ecosystem for 50 years, it shouldn't just return to life with a tiny bit of effort.

Currently, we have a system that gets too far out of range too quickly (-2000 +), but that's just because the events are so large (and negative) and positive effects so weak.

Let's look at your proposed cases

Not at all, it has to do with the timescale that it takes to overcome the event. For example, lets say a disaster strikes in each of the three models that decreases the Economy RCI by 90 points.

Case #1: The effect lasts for, say, a year and then goes away, leaving untouched whatever slow permanent improvements you've been making to the Economy: the disaster is a temporary debuff. Assuming no player involvement, then if you look at the RCI before the disaster and the RCI ten years later, they'll be roughly the same (assuming no tech development, other events, buildings, etc).

Case #2: The effect lasts for a year and then either goes away if you've responded to it and dealt with the problems, or becomes permanent/semi-permanent if you have not. If you deal with it, then two years after the disaster the planet's Economy will be back to where it was before the disaster. If you fail to deal with it, then ten years after the disaster the planet's Economy will still be below what it was before the disaster (assuming no tech development, other events, buildings, etc)

Case #3: The effect lasts for a year, after which it has made a significant change to the planet's Economic RCI, such that more or less no matter what the player does, if you look at it 10 years later then the Economic RCI will be lower than it would have been if the disaster had not occurred.

I made one change which was to make Case 3 start at -150 rather than 90 due to the overlap. The attached plot shows your 3(+1) cases.

• Case 1 - Returns to baseline without player intervention
• Case 2A - Returns to baseline with player intervention
• Case 2B - Would return to baseline, but player ignores it
• Case 3 - Player can't affect it

The only difference between Case 3 & 2 is that in Case 3 the amount of change necessary to adjust the line is too large for the player to do anything about. The solution is to increase the relative effect the player has, which makes Case 3, Case 2. The problem I have with Case 1 is that they player can ignore events completely and they will just go away. An event has no long lasting effect, so why should the player care about it?

[NichG]

I don't understand why you have a problem with an unstable system. Isn't divergence at infinity a good thing? Otherwise, the player can simply do nothing and the system will proceed to self-correct back to baseline (or at least stay within a standard limit). That is to say if a player ignores a mechanic for 1 month or 100 years, they would have to spend the same effort to adjust the system to put it wherever they want it.

Yes, this is basically what I want. The idea is that the player's ability to adjust the scenario should be roughly the same whether its game year 1 or game year 100. Otherwise what happens is that all the important game decisions are frozen in early on, and the late game is just playing out the 'destiny' that you set up during the short time-frame that the scenario was actually mutable, which is generally dull gameplay (lots of time but few actual decision branches, and if you screwed up early on you can in principle leave the game in an unwinnable state without being aware of it for a long time).

I'm comfortable with some degree of freezing-in like that, e.g. due to advances in a race's technology or the presence of additional buildings making it harder to shift things by a bit, but divergence towards infinity means that the game becomes arbitrarily unplayable the longer you wait. For example, if the first 20 years has RCI values that start around 0 so are nearly trivial to shift, but then the last 80 years has RCI values that hover around +50 or -50 and so require a (non-diverging) fixed amount of effort, then that's okay.

This divergence thing has shown up in the complaints other people have about RCI too. Its not just me being mathematically fixated. I've seen at least three posters on these various threads comment that in the late game the RCI values become really big or really negative and you basically can't shift them without sitting and doing dispatches for a year. This is actually something people are encountering and being turned off the game by.

For gameplay purposes, I see divergence as good, the problem being that you don't want it to get too far out of a reasonable zone in 75% (some arbitrary number) of cases. However, if you ignore the collapse of the Peltian ecosystem for 50 years, it shouldn't just return to life with a tiny bit of effort.

If you want to be able to have very long games (which seems to have been a goal in the recent patches) then divergence is bad not just because it renders late-game actions meaningless compared to early-game actions, but also because its really hard to control when the length of the game has a large variance. Essentially divergence amplifies all the little errors, sources of noise, etc that you make in the game design the longer the game takes.

Take an exponential growth curve. If I want populations to be about 10 billion people by 100 years, and the game starts with populations of 500 million, then what growth rate should I choose on average? Well, I need the population to be multiplied by ~exp(3) over 100 years, so the time constant of growth should be 0.03. But now lets say the game goes 10% longer than I expected (110 years) - now I'm at 14 billion people, or 40% higher than my target. If game goes 30% longer than expected (130 years) I'm at 25 billion people, or 150% higher than my target. So thats pretty error-amplifying, and the same is true if I screw up the growth constant by similar percentages.

So its extremely hard to give a consistent gameplay experience because the numbers you try to balance around are a moving target.

The only difference between Case 3 & 2 is that in Case 3 the amount of change necessary to adjust the line is too large for the player to do anything about. The solution is to increase the relative effect the player has, which makes Case 3, Case 2. The problem I have with Case 1 is that they player can ignore events completely and they will just go away. An event has no long lasting effect, so why should the player care about it?

In Case #1, the player cares because the event influences their immediate strategy. If I get the Acutians to go to war with the Thoraxians and I'm in the middle of a 5 year ground invasion, then suddenly the Acutians get an event which tanks their economy and makes the pressure let up, then I have to do something about that event even if it isn't going to affect the price of tea in Acutia 10 years down the road. The effect it has on me is that my immediate plans must be revisited and adapted to a change in the current situation. If I don't adapt, the Thoraxians survive the invasion or even invade Acutia. So the permanent consequences have to do with specifically how the immediate and large change in a race's parameters influence my current plans. That is what defines Case #1 - the game design focus for disasters is on the player's here-and-now decision making.

The difference between Case #2 and Case #3 is the effort required versus timing of that effort. The intent of Case #2 is that if you act immediately, the effort required to fix the situation is much less than if you act in a delayed fashion. So in Case #2, a disaster further makes the player drop their current plans and take care of the disaster, or face the long-term consequences. Separating temporary perturbations to RCI from long-term ones allows us to separately balance the mechanical 'effect' of not interfering and the 'effort' required to fix it in short and long terms. In Case #2, if the baseline drops by -10, that is literally impossible to fix (because there's only a finite set of ways to raise the baseline) - but the temporary effect can be -90 without committing us to making the long term cost of non-interference -90.

So, concretely. If you act within a year, the effort needed to 'fix' the disaster in Case #2 is '10 units'. If you fail to act within a year, its '1000 units'. In Case #3, the cost is the same whether you act now or later.