Speed adjustment considerations

[low_grade]

I understand how probability works. A sample of 88 locos from the same running start will see perhaps +/-1% variance in rail profit in 90% of the trials, so yes, there's a 10% chance of an outlier, though slim chance that it would be by more than a few percent. So yes, I'm keeping this in mind, and at this point attempting to provide some good data to consider. When the variance in results is more than 5%, you can be sure there is a difference in performance between locos. Many of the locos on this list can move up or down a few spots on this list.

Mostly what I'm seeing here is that on this kind of map, if available, with the 1.06 stats, the Stirling and Eight Wheeler and Crampton dominate and averything else is mediocre. Even the Connie, though it'll be a bit better than the Crampton at some things from 1865-1870 while you wait for the Stirling. But looking at some of the modded engines, particularly the American lineup becomes much more interesting (instead of American>Connie>Eight Wheeler with a sprinkling of Dukes.)

I've now done an install with Lirio's pack and tested them. While I appreciate the boost to the Adler and Norris, the Standard 4-4-0 is just too good, and then again his American 4-4-0. Some of his other mods are interesting... Ten Wheeler more of a long distance freight hauler? Camelback is bloody quick now and will most certainly compete with the 2-D-2 for that niche. Nerfing the Eight-Wheeler like he did was certainly in order. I may keep some of his modded engines for use...

I'm thinking I may use in my custom install the following lineup with NA/World/Europ availability all covered hopefully:
1835 Lirio's Adler
1837 Lirio's Norris
1840 Firefly
1942 Hercules 0-6-0 (my Baldwin for the US to be a hauler)
1945 Baldwin 0-6-0
1952 Crampton
1955 DX Goods
1955 my American
1965 my Connie
1970 Stirling 4-4-2
1975 Lirio's Ten Wheeler
1880 Lirio's Duke
1890 Lirio's Eight Wheeler
1892 S3 4-4-0
1896 Lirio's Camelback

I think all these will see good situational applications, though I still think there are gaps. 10 years without some sort of improvent in some area is too long (1855-1865 and 1880-1890, and also sort of the 1840's-1855.)


Updated table here:

[RulerofRails]

I can't figure out why you are having a problem with the maintenance cost prediction. I assume you are using the latest sheet.

The formula used in that sheet is:

Maintenance multiplier to find cost at a specific engine age:
(0.00175*(A^2)+0.0884A+1.0959) * Maintenance cost (straight from the lco file, but in thousands).

Not a pretty formula, but that's what LibreOffice could do with the "Add Trend Lines" tool that gave respectable accuracy. My maths weren't up to the task of finding it manually.

The Eight Wheeler has a base cost of 6k. When the equation is solved for A=4 (Displayed engine age is 4) we get 1.4775 * base cost. That works out as 8.865.

When making the sheet I put in rounding functions so it will read 9k for a Eight Wheeler with Displayed engine age of 4. I mainly tested single locos. From what you recorded it seems like the game may not round off engine maintenance in the ledger (what you really pay) even though this is definitely done in the individual loco revenue/running costs display, which is interesting.

PS. Why do you have two columns (LTD and a YTD) for Mail? This was a one year test right?

[low_grade]

Yeah, I use 0.0017533218*(4^2 years)+0.0883636686*(4 years)+1.0958709528*(listed maintenance cost)=7.0, but in the game the Eight Wheeler cost $8.8k on average. Ah, I think I see the problem, lol, I removed one too many sets of parentheses! Yeah that formula exactly describes my data now, whew!

But fuel is still consistently higher than predicted. No surprises here, but perhaps Lirio's Adler and Norris could be toned down a bit... And as suspected, the original Adler is no better than a Planet... Here's the little first round completed in 1943 same map, Southeast USA.

This is still with the old data. And the extra Mail column was for calculations, I forgot to hide it.

Okay, on to the 20th century for some more testing!

[RulerofRails]

I think I mentioned the load miles hauled isn't accurate. The detail behind that is that I ran a test with one loco doing exactly the same thing (didn't go over end of year so separate from average speed inconsistency), running 10 or so repetitions. The Load miles hauled displayed different each time. Some of them were ballpark in a similar neighborhood, but there was also those that were further out too. That was early on in my testing "binge". Didn't save the specifics unfortunately.

TBH, unless we use wait-to-fill to ensure full car loads we can't get a super accurate fuel cost even when mileage is known. For example if a train on a longer run has more cars than a train on a shorter run (not implausible) that will skew the result lower if we selected the average of the two for consist weight.

I did quite a lot of test runs in controlled environments to double-check the fuel cost formula. So can't imagine I used it all this time with a major mistake like the wrong constant. If you care to define the constant better go for it. As I see it, because mileage and accurate consist weight are estimates of some sort for a typical game it will never be air tight as a prediction.

[low_grade]

Could it just be that fuel costs are higher once we get some grades mixed in?

And a few preliminary results from the period C test. Mallard haters can shut it! Haven't quite finished with the earliest available from 1900-1949, and have some from Lirio's pack to test yet, but I've always heard a lot of hate on the Mallard. Yeah, it's steam, it also averaged 25mph, fastest steam, and most revenue. GG1 is king, followed by the E18, Zephyr, E428, and F3 (then Mallard.) Details to follow.

[RulerofRails]

Grades have nothing to do with fuel cost (at least according to RT3's simplified mechanics) in the game. The formula is hinted by the description when you place the cursor over "Fuel" in the INCOME STATEMENT in the ledger:

Fuel costs for your trains. These vary based on your locomotives (some are more fuel efficient), and on how far those locomotives travel and how much weight they pull.

Yeah those are the usual suspects for that era. The Mallard is good on flat ground. Probably needs a decent cargo density and good express volume to really shine. GG1 is good too on flat ground. Both these locos are seriously nerfed (relatively low pulling power) by the default 1950 weight jump. E18 is a super loco, cheap and good performance (obvious from the stats). If it's available, I don't use any other electric (I've rarely bought the E428). Yep, I've used Zephyrs quite a lot too (look so unrealistic hauling freight ). In C-era are better deal than F3, especially on the flat.

Did you record how much it cost to electrify your network so we can get a costing on extra track maintenance cost as well?

[Gumboots]

Could it just be that fuel costs are higher once we get some grades mixed in?

No. It's only distance x weight. Oil and water are even simpler. They're only distance (weight and grades don't matter).

The only one that varies with grade is sand consumption: it's zero on a 0% grade, and roughly the same as water consumption on a 3-4% grade. It's the same up or down the grade, and regardless of weight.

We tested all this stuff thoroughly. The game's mechanics are far more simplified than we originally expected. What they did was trim everything down to the minimum that they could get away with while still maintaining the illusion of a train game.

As RoR says, the inconsistencies you're seeing would be the result of random partial car loads, along with random variations in consist makeup and length. The fuel consumption formula is very simple.

[low_grade]

$27M to electrify, and track maintenance is $5.6M vs $4.2M for non-electric. Say if you go electric you'll have at least 10 years to pay it off, running 88 locos as I am that's a premium of $47k per electric loco (not penalizing the opportunity cost as I should, admittedly, as if I look to get 20% return, that $27M is a rather high interest loan...) If I count that against the electric locos then the top performer is the Zephyr, followed by the GG1, F3, E18, Mallard, E428, Class A1, N&W Class J, Orca, Lirio's Class A4, and the Class V2, all above $350 profit per train average in this test, Zephyr made $442 per loco.

Dang Lirio messed with a lot of locos... and not always with any dramatic effect...

Chugging away at this, break for work...

Arg, just say I pay the interest, that's $5.4M with a 20% opportunity cost, $2.7M with a 10%, plus $1.4M, then I could be looking at a $77k premium, pushing the GG1 down below the F3 and the E18 below Lirio's Class A4 and the E428 below the Class V2.

[Gumboots]

TBH the default Mallard stats are over the top. Maximum service speed for an A4 was usually in the 100-110 range. To make sense with a 126 mph top speed the game stats would have to use a very low (possibly negative) value of free weight, so that service speeds on the flat with a long consist were reduced appropriately.

Dang Lirio messed with a lot of locos... and not always with any dramatic effect...

She was trying to balance the entire roster, but didn't have the spreadsheets that we have now so was largely flying blind. The repetitive testing necessary back then eventually wore her down, and I think a few locos sort of missed the boat. Also, I remember her saying that she only ever played steam, which I thought was very sensible, so it would follow that her testing of the diesels and electrics might not have been so thorough.

[low_grade]

So far I've only noticed changes with the steam, but it is interesting testing her creations, some pretty cool ones, as I've stated that I'll be using some when I get back to actual playing lol. But it is also a bit tedious going back and forth to see if any changes have been made. Plus the fact that my regular install has I forget how many personal modifications (some of which I've scrapped in favor of Lirio's.) So I have 4 installs of RT3, my custom, 1.06, 1.05, and Lirios that I'm going back and forth between to look for differences. Well not 1.05 yet, though I'll probably run those locos, too, before I get back to Coal Miner's Daughter and get it set up properly for a comparo. Any better suggestions for an intensely graded map to test on?

Edit: just did what I wanted to do before I had to leave for work, account for the cost of the loco. It's an investment. You want a 20% ROI. Actually made it so you could set the desired ROI on the Electrification and the price of the loco, so 15% late game? 30% early game? But at 20% ROI, some big upsets! The GG1 gets moved way down, as does the Mallard. A before Zephyr $402k tops, E-7 and FP7 a bit behind, F3 ($348K), E18, Class A1, GP7, Orca, 2-6-4T Suburban(surprise?!?) , Class V2 ($320k), Mallard, GG1, N&W Class J, G5s 4-6-0, Lirio's Class A4 2-6-2, Lirio's Class P8, my Atlantic 4-4-2, Class 5P5F Black 5, and then finally the E428 at $291k factoring 20% ROI on the investments you've made. The median is $269 profit, the low is $175, Lirio's Challenger. Zephyr is even more ahead on cost, and should be nerfed.

[RulerofRails]

TBH, in real life we generally have a mountain range or two with rail lines through the passes and the vast majority of trackage is on the flats. There are plenty of maps with a significant range in them (your Chile map is a good example, also some maps of western US or Canada, even Pacific Northwest, and the Alaskan maps). Others are more bumpy than mountainous which is more of a bother than anything. Even on maps like Coal Miner's Daughter, a great portion of mileage would be done on low grades (<3%). except that we need to make deliveries up to the mines otherwise if car count is restricted on the uphill, there will be on net substantially more cargo (Lumber, Pulpwood, Coal, etc.) coming down. Mountain haulers only have a niche for the most part.

Possibilities where there may be a fair bit of traffic on grades (was awhile since I played them, so memory isn't exactly clear on whether they could be better for your purpose than Coal Miner's Daughter):
Scrimshaw
Twenty Years in Tibet
Durango & Silverton (small map, not in the archive, found here)

---

No surprise here with the Suburban Tank engine. It needs adjustment.

Care to elaborate on how you are you are calculating engine ROI. Does this change the Eight Wheeler coming out on top in your last test or is it just for electrification?

I could write paragraphs about electrification (ETA: I just did ) . Most of the scenarios we think of "electric" being decent have an event that reduces electric track cost (for example I believe all the choices in the default and Coast to Coast scenarios which give a choice for "Electric" reduce electric track cost). This closes the gap in short term ROI.

In scenarios without a reduction in electric track cost, I think cargo density needs to be quite high for electric to beat diesel (prob. still wont beat the Zephyr) in ROI terms for the short term. Cargo density of the high value cargoes can be influenced by industry building. I like the strategy of waiting for Depression/Recession and electrifying then. But it sort of needs to tie in with a decent replacement age for your existing engines.

However, maps with a haulage goal require hauling lots of cargo regardless of revenue. This can increase usage especially on a main-line. Also, when revenue is lower, running cost becomes more important. So, it can be smart to electrify earlier in this case. I think African Dream fits this category. Still not sure if it's smart to build electric from the beginning.

I did some plays, for example with Steam vs. Diesel vs. Electric recently. TBH, Electric has better economics than Steam, if played with a heavier industry mix (if you do an all industry start and remain heavy on industry, well Electrics from the start may even be a great idea) as I tend to do with Steam than with Diesel, Electric from early in the game can give decent results for sure. Although if cities are further than average apart that can change. So many possibilities.

[low_grade]

Scrimshaw, that was the one I was trying to recall. And testing in the grades is to find which locos move up the list because they don't slow down to a crawl going uphill, and are also pretty decent in the flats. Just curious to see.

And I know it's silly to put a number on it, but I'm asking for 20% ROI or more from pretty much anything, so I'm subtracting 20% of the purchase price of the loco from its profit figure. And this does shuffle the list around a bit! We now have:

Lirio's Eight Wheeler $302k - up 1
Eight Wheeler $296k - down 1
Lirio's Camelback - up 1
S3- down 1
Lirio's American
My Ten Wheeler $268 - up 1
Lirio's Duke - up 1
My Camelback - up 1
Stirling $262 - down 3
Lirio's Ten Wheeler
my Mogul
Camelback - up 1
Duke - up 1
Lirio's Connie II - down 2
Lirio's Standard
Mogul
Crampton - up 1
my American - up 2
my Connie $237
regular Connie $235 - down 3
Lirio's Connie $232
Vittorio
DX Goods - $221
etc down to still $104 for the Adler

Edit: just for fun and since yeah the profits from network expansion in the early phase I suspect can go north of 50% ROI so you want a lot of cheap locos to take advantage of ready cargo, then upgrade to more efficient locos as the game matures, I tried looking at a 60% return, and as you'd expect yes at this point the Eight Wheeler and Stirling move way down. Lirio's cheap locos continue to move up (her Eight Wheeler tops at $262k,) the Ten Wheeler is better than the Eight Wheeler by a wee bit, the S3 holds its spot at 4th with $239k, the Camelback moves up to within no statistical difference from the Eight Wheeler, the American moves way up with $200k net, the Vittorio moves way down to below the Firefly lol with $174k, and my Consolidation becomes clearly the best Connie (I priced it at $100k rather than $120.) This is a list of best high ROI options at a given time in the B period:

1942 Hercules 0-6-0
1945 Baldwin 0-6-0
Crampton 4-2-0
1955 my American or Lirio's Standard 4-4-0 or a regular American or DX Goods
1970 Stirling by a little if you're not using my or Lirio's locos also 1970 Lirio's American but it's just too good really
1974 Duke class is a slight improvement over the American and is a toss-up with the Stirling, if you're not using my or Lirio's locos. else 1974 my Duke or 1980 Lirio's Duke
1886 Mogul is also in the mix, only slightly better than the Stirling but no match for my or Lirio's Dukes
1890 Lirio's Eight Wheeler, or if not using it...
1892 S3 4-4-0 or if not available the Ten Wheeler or better my Ten Wheeler

so if using 1.05 locos, Baldwin, Crampton, American, Stirling, Duke and S3 are the only locos to go with when shooting for 60% ROI figures. So 15 years of Americans, and really the Stirling only netted 3% more, so kind of more like 37 years of no real improvement until you get the S3, if you get it. And 1.06 only adds possibly a few Moguls and the option of the Ten Wheeler, but still basically 37 years of Americans. And I feel like in my playing mindset, while I make business decisions based on 20%, by the time I have industry set up and start with rail, and assuming healthy express traffic, I want to blow up to 20+ cities within a few years to build on that express network and take advantage of new price gradients, and in this period I probably am thinking more like 60% ROI. Rather than 6 Stirlings at $140k a pop, I can connect to another city and run 8 Americans (or still 6 of mine at $55k a pop.) After I've expanded my replacement strategy follows the 20% ROI list, since at this point in a mature game that's more realistic. So the Connie becomes better than the regular American, I think as we all practice in our games in this period, if the Crampton's not available. Then for me the Duke and Stirling, then the Ten Wheeler, but I usually hold off a year and wait for the Eight Wheeler. Well, that and I'm using my mods and some of Lirio's so there's more consistent progress in performance so I have different options every 6-10 years.

[Gumboots]

Any better suggestions for an intensely graded map to test on?

The PopTop California map has plenty of grades and some decent length runs. I'm familiar with that one because I made a variant of it. IIRC NW Pacific is another good one for grades, and the one that goes up the west coast from San Francisco too (can't remember what it's called).

...2-6-4T Suburban(surprise?!?)

Not really a shock. It has reasonable top speed, grunt, purchase price and maintenance costs combined with very low fuel costs. Frankly I regard it as a bit of a super engine. It's not one that leaps out at people at first glance, but its stats make it one of the best all rounders.

With Lirio's stats, she probably didn't test fuel consumption. I know she was aware of maintenance costs, but back then we didn't know how to calculate fuel bills, and testing oodles of minor variations there would drive anyone nuts.

[low_grade]

Thanks, Skrimshaw is perfect with its easy to lay track on big swooping grades.

Right now I'm doing some error correction. Looked at each Reliability and took the average and standard dev of the breakdown chance for each category, then re-ran the outliers a bit. Only bring this up because it appears that we've been deceived yet again:

Reliability/breakdown chance per 1000 miles
10 - 6%
9 - 12%
8 - 18%
7 - 30%
6 - 42%
5 - 54%

this is much higher than the displayed probability... and I'm yes using the formula for miles per year=2.35*12*mph, and yes calculating fuel costs they're somewhat off, but usually not too much and if by much then lower than predicted, and I'm assuming 6.2 loads + caboose, so I'm pretty sure when fuel costs are lower it's because I'm actually averaging 4 or 5 loads, something like that. So I'm pretty sure my mileage estimate is close enough that these breakdown figures are significant.

Also, it appears that breakdowns are not following the rules of probability that I know. Or I don't understand when it's tested. Is it every week? I don't think so... Breakdowns per year are highly variable, sure after error correction the standard dev is just 20% of the average, so not a huge range of variability, but initially the standard dev was more than half the average, in other words, I was quite likely to see for example Good reliability locos running 500-600 miles per year seeing anywhere from 3 to 15 breakdowns in a year with 88 of them going, or a range of breakdown chance from 8% to 30%+. Seeing this is what made me decide to do some re-runs... Although it turns out the result of taking a freakishly bad or good year for breakdowns and replacing it with an average year is a difference of at most 3% in profits...

Oh, and lol it's probably just the name Suburban, it doesn't really inspire awe...

[Gumboots]

We have no way of knowing when breakdowns are tested unless we can decode the relevant parts of the .exe. AFAIK nobody has even found them, let alone figured out how they work.

Ok, these breakdown figures: are you taking oil level into account? That has a big effect on breakdown chance, and will be constantly changing.

[RulerofRails]

Low_grade, did you enable price islands in your 1.06 install?

Price islands are the lock on price (for a short period of time) established between two stations thanks to a recent delivery. For some reason Milo took them out of 1.06. Perhaps he thought it was possible to exploit these with ship at a loss. There is a chance for that, but I never saw one appear in my games so far. If you put a lot of effort in maybe it's possible, but this is a con that in my opinion is outweighed by the positive: they reduce re-hauling in the typical Auto Consist game. Now I play all 1.06 games with them enabled. IMO, re-hauling is a weakness of RT3, so less of that without serious side effects is a win for me.

I mention this only because without price islands (default 1.06) engines with higher performance have more to haul (thanks to increased chance/frequency of re-hauls). I don't think it makes a huge difference, but I would expect to see slightly more profit in default 1.06 than 1.05 in an apples-for-apples revenue comparison. This artificial "supply" of cargo means that there is no absolute "capacity" as in 1.05. Again, it's probably only a small thing, but I thought about it so figured I would mention it.

---

Yeah, reliability is an enigma. TBH, I was surprised that we found some sense behind the Breakdown Chance reading. I definitely see randomness when playing. Just cause one engine has a better rating may not mean that you actually get less breakdowns.

Can you post the data from your B era test in a sheet? I want to do some calculations to see a comparison with my attempt at "typical" mileage prediction.

[Gumboots]

Yeah, reliability is an enigma. TBH, I was surprised that we found some sense behind the Breakdown Chance reading. I definitely see randomness when playing.

Yes, but that is to be expected. It's how probability works.

[Gumboots]

Oh here's another thing: we already know the game's pop-up for speed vs consist is wrong. We also know there is deprecated code all over the place throughout the files and (presumably) the .exe.

So, it's perfectly possible that the breakdown chance has nothing to do with 1000 miles. I do expect it would be related to some fixed distance, but at the moment we don't actually know that that distance is. It may be in game miles or in track pieces, or any other measure. The fact that the chance is quoted as being "percent" would imply that originally the distance was 100 miles. The current supposed 1000 miles is obviously not a percentage as such.

It may simply be that the counter is reset every time the locomotive leaves a station, not by any fixed distance, so in effect it'd be breakdown chance per trip. Although I have no idea how it would then determine when and where to break down on that trip.

Short version: we can only assume it is logically related, somehow, to what we see in the gui. But we don't really know how it's related. So, all we can do is use the given breakdown chance as a relative guide.

[low_grade]

Okay, got through Lirio's edits, a few interesting ones in the 1900-1949 period.

Here's how the best choices break down

S3, Eight Wheeler and Ten Wheeler are still good bets to begin with, Lirio's Duke and Camelback (especially) also shine.
Well, but you have the P8, and Lirio's Eight Wheeler is just as good... but no need to replace wholesale in 1900 all your S3's or Ten Wheelers.

Fs. GR 2-6-2 and my Baldwin are a bit better, and the Class 500 in 1905. And my Atlantic and Lirio's P8 a bit after. Class A1 or Suburban a while later (need a boost to something in 1915 or so...) and then the Zephyr all the way (need a boost to something in the late 1920s os so, and then 1942ish, else nerf the best so others have a chance.)

Yeah, too much data for a screenshot, here's RoR's spreadsheet that I've been building on, my data is on the last page

And, probability works like this, think of coin tosses. Flip once, you get heads or tails, while the average is 0.5 in the long run, the first result is 0 or 1, very far off from 0.5. The more you flip, the closer you will get to 0.5. So if breakdown chance is checked often, there shouldn't be the kind of variability I'm seeing. Your theory of one check per run, though it doesn't make sense to do it that way, actually makes perfect sense interpreting the results. Breakdown chance cannot be getting checked often. Then another random to determine where in the run it breaks down would explain everything.

[Gumboots]

What's the Fs GR 2-6-2? Never heard of it.