Perfomance files - how do they work?

Been playing around with a personal 2010 mod, and trying to get the performance file right, but I don't know what all the numbers do.

[Team #00]
Name=Mclaren,Mercedes
Performance=781,786,3519
First Driver=1,Jenson Button,14237,445,14380,549
Second Driver=2,Lewis Hamilton,14139,745,14289,854

For the team, i'm guessing the first 2 numbers are race & quali perf, but what's the next number - reliability?
Again, for the drivers I see them as two pairs, race & quali (which way round i'm not sure), but I have no idea what the second number in each 'pair' is. Consistency or something?

Can anyone shed any light on this, or point me in the direction of a useful help page. I've tried the FAQ and didn't find anything.

The second and fourth number set for the driver is variance, or chance. Reliability is also a way to describe it. The higher the number, = more chance of a breakdown, and vice-versa ok. The first number set is qualy, and the third is race. For the team number set, the first is qualy, the second is race, and the third is variance.
For the driver number, the higher the number, the better the performance. With your example, Jensen Button is clearly the superior driver.

Your file....copied here:
Name=Mclaren,Mercedes
Performance=781,786,3519
shows one number too many for the team variance. It shows a massively high chance of failure. I would think that 319 or 519 would be better.

---

[www.mediafire.com] Some say you should click it, you know you want to. [www.gp4central.com] <----GP4 Central

As Mortal said...


[Team #00]
Name=Mclaren,Mercedes
Performance=781,786,3519
First Driver=1,Jenson Button,14237,445,14380,549
Second Driver=2,Lewis Hamilton,14139,745,14289,854

Here's each item explained:

[Team #00]
Name=Mclaren,Mercedes
Performance=781 [QUAL POWER],786 [RACE POWER],3519 [RELIABILITY: 3519 isn't actually very high, by default on GP4 the most reliable team is about 1500 I think and the most UNreliable team is about 8000. Less than 1000 will make cars very rarely fail. But the last several years of course reliability has greatly improved in F1 so maybe all values less than 1000-1500 would be reasonable. The best way to test is to set them to a value, e.g. 2000, then run a simulated race and accelerate time, and see how many retirements you end up with, then work from there depending on how many retirements you want to get - although the number's just a chance don't forget - it's not a guarantee)]
First Driver=1,Jenson Button,14237 [QUAL PERFORMANCE],445 [QUAL PERFORMANCE RANGE: Lower = very consistently at this performance level, Higher = Can be much better, and much worse],14380 [RACE PERFORMANCE] ,549 [RACE PERFORMANCE RANGE]
Second Driver=2,Lewis Hamilton,14139,745,14289,854

Ok?

That's a great help.
Thanks

TeamBPR Wrote:
-------------------------------------------------------
> As Mortal said...
>
>
>
> Name=Mclaren,Mercedes
> Performance=781,786,3519
> First Driver=1,Jenson Button,14237,445,14380,549
> Second Driver=2,Lewis Hamilton,14139,745,14289,854
>
>
> Here's each item explained:
>
>
> Name=Mclaren,Mercedes
> Performance=781 ,786 ,3519
> First Driver=1,Jenson Button,14237 ,445 ,14380
> ,549
> Second Driver=2,Lewis
> Hamilton,14139,745,14289,854
>
> Ok?

wrong

driver's first perf value is race,the second is race's range,the third value is qualify's perf value and so on.
luckily i still remember nacciari and walter_rohl's tutorial

---

My GPGSL Career



Edited 1 time(s). Last edit at 06/05/2010 12:05PM by senna9194.

I'm sure it's Qual then Race - it's always been that way round, maybe it's not like that in GpxPatch perf file - I've always used it as that.

Regardless, even if I got the Qual/Race the wrong way round, the rest is still correct - Ginnerchris so give it a go and experiment with the figures mate

There's no need to try and be clever by just saying "wrong", is there? I never understand why people just turn advice topics into a competition to see who can be the smartest and to try and make other people look dumb. That's not why I use a Forum. I was just trying to help, I won't bother next time.

TeamBPR Wrote:
-------------------------------------------------------
> I'm sure it's Qual then Race - it's always been
> that way round, maybe it's not like that in
> GpxPatch perf file - I've always used it as that.
>
> Regardless, even if I got the Qual/Race the wrong
> way round, the rest is still correct - Ginnerchris
> so give it a go and experiment with the figures
> mate
>
> There's no need to try and be clever by just
> saying "wrong", is there? I never understand why
> people just turn advice topics into a competition
> to see who can be the smartest and to try and make
> other people look dumb. That's not why I use a
> Forum. I was just trying to help, I won't bother
> next time.

i've not tried to show that I'm the smartest,and never tried to make other people look dumb.I've just wanted to correct you.that error is almost common.I had the same problem when i had to set a custom grid manually,because shaddix's tool had problems with new java version.That's because I always thought that race values came after qualify's one.Then i checked nacciari and walter_rohl's tutorial and confirmed my error.that's all.i wrote "wrong" because it's the first word that i thought

---

My GPGSL Career

How do you accelerate the race. I have heard this a few times but have never knew how to do it.

---

Retro Liveries on the SMD-ZG02![www.grandprixgames.org]
Retro Renders 2.0 on the SMD-ZG02LN![www.grandprixgames.org]

Hit escape, and choose accellerated time (or something similar) from the menu

I have always "assumed" simple maths can be employed to control likely maximum and minimum grid/race finishing positions. By adding the drivers quali perfomance and his quali perfomance range this will give his best result and subtracting his quali performance range from his quali performance will give his lowest result. The result being relative to the other driver's similar max/min calculations. Of course the team power needs to be considered too. But I am not certain about this. There's bound to someone around who can verify it.

There is a nifty performance calculator included in Lo2k's Team Editor. It's under the "Drivers" tab.

Edit: I note that Team Editor calls the driver's figures "Ability" and "Variation" so there's a clue there!





Edited 1 time(s). Last edit at 06/06/2010 01:00PM by Mr Pibb.

Mr Pibb Wrote:
-------------------------------------------------------
> I have always "assumed" simple maths can be
> employed to control likely maximum and minimum
> grid/race finishing positions.

From limited testing in the past, I don't think this is the case. The best example is probably one of the minnow teams that still manages not to qualify on the last row. I haven't tested with equal team BHP - so maybe this has something to do with it.

---

X (@ed24f1)

How do I get the to be green striped within the game, what number is what?

At the end of a performance file you have (or could have) a section like this:

[Tyres]
Team #00=2
Team #01=1
Team #02=0
Team #03=0
Team #04=1
Team #05=0
Team #06=0
Team #07=1
Team #08=0
Team #09=1
Team #10=0

where 0 stands for Bridgestone (=> wheel_bridgestone_0.tex), 1 stands for Michelin (=> wheel_michelin_0.tex) and 2 for McLaren (=> wheel_mclaren_0.tex) type of tyres.
So, asuming that you have the green striped tyres named wheel_bridgestone_0.tex (_1, _2 for lower lods), then with the above example you will have Teams #02, #03, #05, #06, #08 and #10 having the soft compound (green stripes) in game.
Please note that the team numbering starts with a 00 for the first team!!

---

My personal GP4 files at : www.4shared.com ______ My Blog at : Gitanes GP4 work

* Performance Files * LoadingScreens GoogleMaps * GP4 Tracks and Cars Databases

By the way, regarding the order of all the values, in a performance file:
first comes race, then comes quali!

So it is:
[Team #00]
Name=Mclaren,Mercedes
Performance=781 [=race power], 786 [=quali power], 3519 [=variance]
First Driver=1,Jenson Button,14237 [=race performance], 445 [=race variance], 14380 [=quali performance], 549 [=quali variance]
Second Driver=2,Lewis Hamilton,14139 [=race performance], 745 [=race variance], 14289 [=quali performance], 854 [=quali variance]

The variance defines the consistency of a driver during the race or the propabilty to have a constant performance.
A large variance number for the team performance means a high propability for one or both of their cars to suffer a mechanical failure during the race!
A large variance number for the driver performance means a high propability for that driver to suffer an accident during the session!

---

My personal GP4 files at : www.4shared.com ______ My Blog at : Gitanes GP4 work

* Performance Files * LoadingScreens GoogleMaps * GP4 Tracks and Cars Databases

After more than 15 years without GP4 I installed it and downloaded 1998 mod. The mod itself was good, but how performance files are organised, is questionable. The I look for more performance files and all of them lacked realism. There are two major problems with user created performance files:
1. They are based on the results and not on the real pace.
Like, if a rider has finished low or failed his qualifying attempt, he will have low performance, it's especially visible in rainy conditions. Like Frentzen on Williams had low performance in Austria 1998, lower than Shinji Nakano on Minardi in the same race. This is not real.
2. Most of performance files have the same engine power. I understand where it comes from: if you only need to adjust driver performance to come to the same gaps, it's easier than to adjust two parameters. Problem: cars underperform on fast circuits like Hockenheim or Spa and overperform on slow like Hungaroring. You have a crazy grip which cars obviously didn't have in 1998. And you cannot really use your engine to overtake as they have the same power.

This is why I decided to correlate performance files with times. But before I present here the result, it's important to resume the environment.
Base values used are: 800bhp, 15418 grip, 0 variance.
1. Correlation is performed in qualifying conditions, 0% rain, 60min session, 10 cars at the same time on track. This permits every AI driver to complete 4 flying laps before the end of the session with the least risk of traffic. So you are sure that 2-3 flying laps are traffic free.
2. 1998 circuits are used for correlation (in fact I used 1997 trackpack w/o Jerez, but the rest is the same as in 1998).
3. For each simulation I gave the same bhp, grip and variance values to all 22 drivers, mean value for each simulation was retained for the correlation.
4. Ace level is used

Bhp is varied from 700 to 800 with step of 25
Grip (or driver performance) is varied from 15018 to 15818 with step of 200 (I know it's a close range, it should be expanded to about 13-16k to draw better conclusions)
Variance is changed from 0 to 1000 (0, 100, 200, 350, 500 and 750)

Just to mention here, there are many misconceptions about the order of parameters:
[Team #00]
Name=Williams,Mecachrome
Performance=765,775,1969
First Driver=1,Jacques Villeneuve,14020,580,15031,296
Second Driver=2,Heinz-Harald Frentzen,14286,580,14785,296

It's:
For teams:
765=race bhp; 775=qual bhp; 1969=failure chance (according to GP3edit 4096 corresponds to 10% failure chance, but to my sense it's higher). Btw, I haven't yet correlated failure chance to be sure.
For drivers:
14020=race performance; 580=race variance; 15031=qual performance; 296=qual variance.

The most important conclusions (gaps in % of 1998 best weekend times, most of them are pole times except A1-Ring and Monza):

1. The most obvious: bhp and grip effects are track specific.

1.1. Bhp has the strongest effect (obviously) in Hockenheim, Monza and Montreal.
Like you have 0.8% bonus for each 25hp, but the dependence is not linear. For some reason, sometimes you have 0.9%, sometimes 0.65%.
Hungaroring is the least impacted by bhp: only 0.37%-0.47% bonus for each 25bhp.
Thoughts? Results can be polluted by internal variance or by mismatching gear ratios (for some specific bhp they can have different effect)

1.2. The same counts for grip.
Hungaroring is the most impacted with 0.57% per 200grip followed by Barcelona, Buenos Aires, Magny-Cours and Silverstone (all between 0.53-0.56%).
The least impacted are Hockenheim (0.29%), Interlagos (0.36%), Monza (0.37%) and A1-Ring (0.39%). Reason? Less corners = less impact.
But again, sometimes you have 0.05% less, sometimes 0.05% more for the same circuit.

2. Variance = 0 does not mean results will be the same.
In fact, there's still a huge spread in lap times (given that I compare 22 drivers which has 4 tentatives every session) even if the variance is 0.
But you can have 0.2-0.9% between the fastest and the slowest. This delta is also track specific. If we talk about standard deviation, it's 0.21% for Interlagos (max) and 0.065% for Imola (min). I don't know why, but it's a recurrent pattern. It's reproduced on various combinations of bhp and grip values. And Interlagos and Buenos Aires had the highest dispersion, Imola and Barcelona the lowest. On every single test.

3. Variance of 100-200 is practically invisible on some circuits.
For example: I modified variance from 0 to 100. On Melbourne I got closer field results for 100 (SD=0.126%) than for 0 (SD=0.164%). Then I increased it to 200 and got SD=0.155%. Only at values exceeding 350 I began to see some kind of effect. This is also track specific. For circuits with basic low variance like Imola you see dispersion immediately.
But even at high variances between 500 and 1000 when the effect of basic variance should be less visible there are situations when you increase the variance and the field becomes closer. It's still unclear how variance works disperses lap times.

4. What is sure that variance improves lap times.
And this is reproduced on all variance values and all circuits.
Again, the most impacted are grip-dependent circuits like: Silverstone, Hungaroring, Barcelona, Buenos-Aires and Magny-Cours.
Like, you adding 200 to driver performance in Hungaroring reduces your lap time by 0.55%. Adding 200 to variance reduces it by 0.22%.
The trick is: it's non-linear and track specific.
Increasing variance from 0 to 200 gives a different effect than from 800 to 1000.
Increasing variance from 0 to 200 gives sometimes 0.4 sometimes 0.7 of pure increasing of driver grip. In farmers words: sometimes is variance of 200 equivalent of 80-140 increase of driver performance.

This latter problem was already discussed on grandprixgames. In fact, in GP4 a random value from 0 to 1 is multiplied by variance and then added to grip (driver performance). This explains why it gives near the half of performance on average (although to my sense, the average bonus is slightly higher than 0.5).
This is different to GP3 variance where a random value from -1 to 1 is multiplied by variance (so the average remains the same).

So these are my findings grosso modo after some correlations. My times are still far from 1998 results, but they are already playable.

Hi Uncle. It might have been better to start a new thread, rather than resurrecting on old one, but regardless, as a fellow alchemist I hope you don't mind if I add a few thoughts....

"1. They are based on the results and not on the real pace."

Generally, for each event you have one recorded real world Q time per car, one set of race results and one fastest race lap time per car in each race (although in the some of the 1970s seasons and earlier, which is where I tend to specialize), you sometimes only have a single FL time for just ONE car in the race! It's not a lot to go on, so how do you propose to measure 'the real pace' exactly? Some mods use just one PF for the whole season, which is probably the easiest solution to that problem, but I can already sense that you don't like that idea any more than I do.

But this is a key question for you really, because you seem to want your PFs to be an objective representation of performance (which is great), but if you're going to base your approach on something like 'the real pace' you would need to be able to identify what this is objectively, wouldn't you?

"2. Most of performance files have the same engine power."

Well perhaps; but mine never do. However, you say "if you only need to adjust driver performance to come to the same gaps, it's easier than to adjust two parameters" and I have to challenge (and reject) both of those ideas. The gaps between the cars can be adjusted most accurately by adjusting one single variable. It's too long-winded to explain here but it's set at 10.3391 in the default game* - and if you adjust it in relation to the length (or more precisely the best Q time) of each track you can make the gap from first car to the last much more accurate - and this allows the car/driver performances to stay rooted in the real world data while avoiding the need to 'adjust driver performance to come to the same gaps'. Subjective opinion has absolutely no place in my PF creation. Everything is based on the objective real word data.

* I first learned about this figure when I downloaded PRBlanco's 2008 PFCalculator many years ago. He never adjusted the value in his model, but he used it in one of his formulas - and while my PFCalculators definitely have their origins in his initial creation, I went on to use the value in conjunction with the Q FL time to create a realistic spread from fastest to slowest car for each track in an objective way without skewing the underlying data and without any subjective input from me. You need to be a fairly advanced Excel user to understand my PFCalcs these days, which is why I've never released any copies of it to anyone, preferring instead to make PF sets for certain people who share things with me, but if you're interested I can give you an outline of how it's done (and maybe offer you a set of the PFs that I would make for use in 1998 for you to compare with yours).

"Problem: cars underperform on fast circuits like Hockenheim or Spa and overperform on slow like Hungaroring"

Well, yes and no. If you want the game FL time to match the RW FL times you can do that to perfection and you want the maximum speed of the fastest car to match the RW speed you can do that too - and you can even do both simultaneously on quite a few tracks, but there comes a point where you have to acknowledge the limitations of the game engine, as the original developers did. This is why you can set the laptime for each track in it's magic data file, but on the very next line you get the opportunity to tweak the clock speed. If GP4 were a full blown industrial simulation you wouldn't need to 'cheat' like this; but it isn't. It's just a game (albeit a rather special one). But point is, it is quite possible to replicate both top speed and laptimes fairly accurately on most circuits (even if you have to play with the clock speed a little to do it), but on some circuits it's absolutely impossible to do both. For example, try achieving anything like that on any of the Brands Hatch circuits (except for Atticus's engineering version) and you'll eventually have to concede that it can't be done without abandoning any ties with RW recorded data completely (and I suspect you'll be no more keen to do that than I was).

I hope this doesn't come across like I'm attacking you by the way. I'm definitely not, but it was after messing with Brands Hatch (and some other problematic tracks) over and over for months that I finally had to accept there's only so much that can be done. However, now that I am a bone fide GP4 PF/MD nerd, I could discuss these things with you all day (because for me, it's incredibly fascinating and I love to hear other people's opinions), but instead I'll just highlight one other little problem that I think you'll have to tackle eventually (and which shaped my thinking in the past).

"I know it's a close range, it should be expanded to about 13-16k to draw better conclusions"

Maybe so, but here's the problem. See how much you enjoy driving a car with a 13000 grip value. The AI doesn't care, but for me (and I suspect for most humans) it's just no fun at all; because it's like you are driving on ice or with four flat tyres. This is actually what started me on my own journey to create the 'perfect PFs', because the authors in 'the early days' who relied heavily on adjusting the grip value to set the spread of the field accurately usually left the backmarker cars in a totally undrivable state and I wasn't prepared to accept that.

So grip has it's limitations (and I would say you're pretty much forced to keep it within a range of about 16500 max to 14750 min or even more), which brings you back to power. Some early authors would set their grip to a common fixed value and instead they adjusted the power to spread the field. This was better, but it highlighted another problem. Power is a VERY dominant value in the game. A car with only slightly more power than another (like, say, +3) will almost always beat a car with the same grip unless something else happens, like an accident etc. So mods which relied solely on power variations didn't really work to a degree that would satisfy you and I either! Some races, yes. Most races maybe. But in every race? Not a chance.

So, what's left? Well, there's variance, but as you have observed already, we have less control over that than we'd like (because GP4 chooses exactly how much variance to add to each car in each Q or R session, depending on the maximum value you set up - and it never SUBTRACTS it, which makes it even less useful than it might have been; which as you note, is one area in which GP3 did better than GP4).

So, just to sum up this very superficial overview so far. Power has to be used very sparingly or it destroys the fun and unpredictability of the game, grip has to be kept within very restrictive tight parameters or it just becomes annoying and variance can't be 'trusted' to do what we want it to.

Conclusion: it's not RFactor, but we are addicted to the GP4 AI, so where do we go from here?

Well, I don't have all the answers. I can only try to raise some of the problems you face (and there are more that I won't cover now, such as what is known as the 'start line acceleration' problem), but I hope that has resonated with some of your observations up to now.

I would suggest however, that at some point you might like to investigate something called the Bellini Patch (which you might have come across in your GP3 days). I made an over-long post about it here on the GP4 forum years ago and at the time I viewed it as the answer to our prayers, but as I was to discover later, that too has it's own problems; especially in multiplayer games. It's still worth a look though, because while it leaves the PFs to control the performance of the CC cars, it allows you to create a different grip value for the human cars which is a BIG step forward in terms of using grip as a major performance determinant (but creating undriveable cars for the human into the bargain) highlighted above.

The fact is, at least as I see it, perfectionists like you and I are forever doomed to arrive at some kind of compromise between our science and our art (but even then you'll find you can't please everyone which is why I only give PFs to my closest colleagues). And with a final sentence like this: "My times are still far from 1998 results, but they are already playable." it looks like you might already be approaching that point.

Good luck on your journey - and if ever your travels do lead conclusively to the Holy Grail, please be sure to let us know OK?

PS: A quick note about failure rates. You were right, but here's the 'official' explanation.

100% failure in a 100% race is 16384, so 10% is 1638 (ie pretty close to your estimate: well spotted!).

However, it's worth noting that 100% failure is never precisely 100% because sometimes a car will break down after it crosses the start/finish line for example - and very occasionally a car with 100% failure rate doesn't fail at all, so there must be some hidden GP4-determined random factor at play there too.

Also interestingly, if not particularly relevant for you, is that the 16384 figure for a 100% failure in 100% race distance races (which SDI says is correct - and he would know), means that 100% failure in a 15% race (which is the race distance I normally use) should therefore be 109215 (ie 16384 * 6.666), but the game only supports a figure of up to 65536 (any higher and the car will never fail at all), so the game can only manage the failure rate in games down to 20% race distance accurately (16384 * 5). As above, the game forces us to compromise the 'purity' of our initial goals at every turn; but I guess that's why we love it!

Hi Noog
(Better call me Dominique or Stan, my name IRL. "Uncle" came from shitstorms on Russian forums where I paternalistically taught Petrov fans: don't mess with Uncle Dominique.)

Firstly thanks for this detailed answer. In fact this is the reason I didn't create a separate topic. There are already many and FAQ redirects here. So when I'm done with all kinds of correlations, I can create an integral PF FAQ with all my gathered data and correlation results.

Last things first:
100% failure in a 100% race is 16384
Yes, I felt it was somewhere here, but from my earlier memories (Alonso title years) I thought you could never reach 100%, max 80%. 2001 values for Ferrari, as I can remember, are around 900 and there were already 3-4 failures. I can also remember I put 16384 to both of fictional cars (for 2005 I created Arrows with Davidson and Montagny). But GP3edit definitely misled me.
Technically I cannot understand how you can reach 100% failure. For me it's like a radioactive decay. You have each lap a certain probability of failure (or rather chance to survive) which cumulates to a given value to the end of the race. A geometric progression with ratio<1. Your survival chances are multipiled with each lap by this ratio r, but it never achieves 0.

Generally, for each event you have one recorded real world Q time per car, one set of race results and one fastest race lap time per car in each race
That's the trick. For qualifying performance I take the best time of the weekend, which can be also FP1-FP3, or Warm-Up. A good bypass if the qualifying was rainy.
Race pace is more complicated. In fact, this was about 75% of my job. As far as I'm interested in 1998, lap times are available even for free. Basically, I take lap times, remove 1st lap, in-out laps and laps that are slower/faster than the previous by 1.5% of the best time of the weekend. Then the time for adjustments:
1. If the strategy is different: 1/2, 2/3 stops. Add average lost time for extra stop and divide by the number of laps.
2. For retired drivers extrapolate to the teammate or someone on the same strategy with comparable pace. This is however open for interpretations.
Like Hakkinen retired being 2nd between Coulthard and M.Schumacher in San Marino. Coulthard was significantly faster than M.Schumacher, only in the end he had problems and slowed down. So if I correlate Hakkinen with Coulthard, he gets the same extra time and will be behind Schumi. If I correlate him to Schumi, he will be the fastest. Different outcomes based on methodology. I decided finally to correlate him to something in between. Or sometimes you have a Trulli-train, sometimes an early retirement. Everything is here subjective, but it's the best way I could handle.

When I used a 1.00 version I had a generic file, was jealous on you all here, but things were much more simple. I got an average gap of each driver, deleted like 20% his worst results, then took an average and a variation, put it into GP3-edit, then attributed half of the gap to the team, half to the driver. Quick fix and still playable. But now I have 1.02 and I want to profit of the opportunity.

but there comes a point where you have to acknowledge the limitations of the game engine
This is true, but I don't want to be a complete perfectionist. For me, having on average 0.00% difference to 1998 pole times is ok. It is achieved at 800bhp and 15411 grip (close to my initial guess of 15418). So my field still underperforms in Monza, but overperforms in Melbourne. But: I have +2.5% in Monza instead of +4.1%. And -1.3% in Melbourne instead of -0.4%. For me a fair exchange. The most important are the gaps between cars and here I find my correlation still incomplete due to a narrow range (15018-15818).

See how much you enjoy driving a car with a 13000 grip value.
I will not. It's not for me, it's for Ricardo Rosset, he was many times outside of 107%. More precisely, 12923 for Rosset vs 15195 for Takagi on the same circuit.
And secondly, to built a regression graph which is precise enough, you need many points and a wide range even a little bit outside of the range you have to use.
My current set of results is insufficient to something more than a linear polynomial. If I try to build a quadratic or cubic, then the regression gives incorrect results: more grip leads to slower cars which is a nonsense. But this is how regression analysis works: your X-point is outside of range of validity = your Y-point can be somewhere you don't expect is.
The only solution is: enlarge the studied range. So in my case: somewhere between 12900 and 15900.

P.S. I thought the max value of grip was also 16384 (a round number in the binary system), but then I accidentally increased grips of some drivers up to 17000 just to look how it works and they were much faster than those who had 16384.

For the rest I will answer late, I'm falling asleep now

Hi Dominique

"I can create an integral PF FAQ with all my gathered data and correlation results"

I'm sure that will be very useful to many here.

"Technically I cannot understand how you can reach 100% failure. For me it's like a radioactive decay. You have each lap a certain probability of failure (or rather chance to survive) which cumulates to a given value to the end of the race. A geometric progression with ratio<1. Your survival chances are multipiled with each lap by this ratio r, but it never achieves 0."

And that is a brilliant description that deserves a place in your PF FAQ. Radioactive decay... Very nice indeed.

"For qualifying performance I take the best time of the weekend, which can be also FP1-FP3, or Warm-Up. A good bypass if the qualifying was rainy."

Ah yes. I'd forgotten that you have a wealth of data available to you in the later years and it seems you're making very good use of it. My interest in RW F1 was laid to rest with Senna really, so I've done almost all of my work between 1972 and 1995, when RW data was usually very limited - so I just wanted to create a model that was consistent and fairly simple, where I could just drop in the RW data I could gather and have something emerge from the other end with little or no input from me. But yes, now that I think about it, had I started out in the later years I too would have based my PFs on all the available data as well to come up with a more accurate 'real pace' value. Again, nice work by you and I'm sure those who prefer the modern F1 era will be looking forward to your PFs already.

(Which is not to criticize anyone else who makes PFs for the later mods by the way, because I've never once loaded a mod later than 2005 so for all I know they may well use all the available RW data too!)

"Race pace... Basically, I take lap times, remove 1st lap, in-out laps and laps that are slower/faster than the previous by 1.5% of the best time of the weekend. Then the time for adjustments:
1. If the strategy is different: 1/2, 2/3 stops. Add average lost time for extra stop and divide by the number of laps.
2. For retired drivers extrapolate to the teammate or someone on the same strategy with comparable pace."

Again, probably a lot more work than I would like to take on (and simply not possible anyway in the earlier years where data is sparse), but who can fault that approach?

"I got an average gap of each driver, deleted like 20% his worst results, then took an average and a variation, put it into GP3-edit, then attributed half of the gap to the team, half to the driver".

Another nice idea. I use a similar technique to make use of the RW championship points table in relation to car power. Crudely, the more RW points a driver scored in the RW season, the more of a power bonus he gets, but within limited parameters, so that the most successful driver's power never rises to more than +7% in relation to the least successful - and I have a mechanism where a small proportion of driver 1's success is allocated to driver 2 as well to make sure that a 'bad' driver in a good team still drives a relatively powerful car (but he gets kept under control by losing a bit of his grip as a result). It's pretty arbitrary, but it works reasonably well to keep things 'looking right' on the track.

Similarly with variance, I calculate it using all the data I have (losing the bottom 20% like you do would probably be more accurate, but I'm lazy), but then I make sure that the maximum amount of variance given to any driver can never be more than 550 and any reduction at that point is reflected proportionately across all the other drivers, because of the unrealistic impact it can have on a race when a driver gets particularly blessed by the game.

(There's one old mod I remember playing with the originally released PFs many years ago and outsider Manfred Winkelhock overtook me on the outside of a corner at an absolutely outrageous pace and that was the day I learned that variance had to be kept under control! IIRC his variance was 1200 or something and he must have bagged all of it in that race).

"I don't want to be a complete perfectionist".

Are you sure about that?

"For me, having on average 0.00% difference to 1998 pole times is ok."

I mean, are you really sure? Go on, admit it man, we're both crazy! We started out with the honourable goal of mastering GP4 and somewhere along the way we became enslaved by it instead. But unlike a real slave we can always press the power button and walk away..... can't we?

"See how much you enjoy driving a car with a 13000 grip value."
"I will not. It's not for me, it's for Ricardo Rosset, he was many times outside of 107%"

And,

"My current set of results is insufficient to something more than a linear polynomial. If I try to build a quadratic or cubic, then the regression gives incorrect results: more grip leads to slower cars which is a nonsense. But this is how regression analysis works: your X-point is outside of range of validity = your Y-point can be somewhere you don't expect is.
The only solution is: enlarge the studied range. So in my case: somewhere between 12900 and 15900."

I understand, but one of my primary objectives was always to make the backmarkers drivable by the human, because it was always my intention to create a true 'career mode' for GP4 spanning a certain number of seasons (which is still something of an unfinished project) and that meant I had to find a way to make even a weak car retain enough of the F1 feel to make it enjoyable to drive. Plus, the person I work with on the mods likes to make every single car that ever took part in a GP (whether it qualified or not) so he also wanted to have fun driving his backmarker creations - and together those drove me down my path. Once again though, it's interesting to see how you and I have attempted to overcome the game limitations in our own ways.

"I thought the max value of grip was also 16384 (a round number in the binary system), but then I accidentally increased grips of some drivers up to 17000 just to look how it works and they were much faster than those who had 16384"

Yes. I'm not sure what the max grip value is, or even if there is one, but as you say, it's certainly not 16384. However, I took my cues from the original game and while I don't have an absolute limit that I impose, it's very rare that a car ends up with more than 16600, because it doesn't feel right in my older mods - but in the post 2001 mods I'd probably allow it to go higher than that.

In fact, I'm still considering the best way to regulate the power/grip relationship in my mods. Originally I tried to make the power realistic for each year, so in the mid-70s for instance, each car would be given an intial seed value of 440; but whereas a car with 800 power feels about right with 16384 grip it feels like way too much in a slower car. However, because of my desire to keep all backmarkers drivable it was impossible to reduce the grip to where it needed to be while maintaining a close relationship to the RW results, so I've currently set my PFCalc model to start with the 2001 peak default power (860) and I reduce it by a certain percentage each year, which I don't recall, but it means that by 1970 (where my PFCalc currently begins) peak power is down to 674. Now this of course is absolute nonsense, because not only is that way too high historically, but it takes no account of rule changes and huge fluctuations in max power over time (turbos, non-turbos etc), but I decided to allow that major compromise because I wanted to feel a linear progression from year to year and I quickly realised that the game worked best the closer you could stay to the original parameters of 860/16384 (this also impacts on the start line problem too, but that's for another time perhaps).

Again, to return to our theme yesterday; we are forced to compromise our ideals time and time again, but from our conversation so far, which I am enjoying very much, I am pleased to congratulate and salute a fellow slave.

Good work!

Hi Noog,

I've done almost all of my work between 1972 and 1995, when RW data was usually very limited
Ironically lap times are available exactly since 1996, at least at Pitwall: [pitwall.app]
My area of interest is more or less when my peak of interest for real F1 was: end of 1997 to 2008, maybe extended to 1994 when I saw my first race, Barcelona 1994.
P.S. What is RW data?

For 1.02 version I have developed the following algorithm to sort out anomalies (valid both for qualifying and race):
1. Go to drivers pace (discussed in the last post)
2. Subtract best of teammates to get gap to the car. Obviously one of the values is 0, but the other varies.
3. Calculate mean and SD (standard deviation) of each driver's gap.
Normally about 99.7% of results lies 3SD to the left or to the right of the mean, but here it's not the goal.
4. Mark all the results that lay 1SD to the left/right of the mean. They are declared anomalies. Most of them "too slow" (like Villeneuve in Monaco), but some "too fast" (Irvine in Monza, many Rosset results). So exclude not per se too slow or too fast times, but unusual gaps to team's best.
5. I declare Mean±1SD as the new normal for the driver at that Grand Prix.
6. Then I enlarge my standard variation value (296 for qual, 580 for race)
6.1. For too slow drivers so that the lower bound of driver's pace corresponds to his real 1998 result (which I earlier declared "abnormal".
6.2. For too fast drivers I move the upper bound similarly so his 1998 result remains achievable.
If of course, 1998 results aren't already within my standard variation.

This way 2-4 results per driver are adapted (mean) and in about 10 cases in total the upper/lower bound of grip moved. I needed to create an Excel macro for this purpose. I can say things could be easier if Magnussen was not replaced by Verstappen. If complexified the programming.

one of my primary objectives was always to make the backmarkers drivable by the human
I thought about this. In fact when I rode for Frentzen in A1-ring, I was frustrated by his low grip (which was not real because he retired early in the race).
But at the second view, I also didn't like the idea of equal bhp. So I decided to take real bhp values from 1998 and compensated them with according grip. In most cases, grip values are between 14000-15200, so it's still drivable.
On the other hand, 1998 was a season when cars were undrivable: narrowed from 2 to 1.8m (less stability against roll), grooved tyres. I'm sure if I model 1997 results, grip will be higher. In 2000 again, only for 1999 I expect more grip problems.

Moreover, if I found some cars "too" undrivable, I would of course also reduce bhp to achieve a certain level of drivability. But for me it's also important to feel the difference in drivability/engine power from one season to another. Also I understand that underperformance on some circuits comes from car setup. My runs are all on the same setups which is non-sense, so you can achieve 1998 times just reducing wings in Monza/Hockenheim.