PV Revision Public Review - Feedback Wanted

[Descronan]

I'm not sure if discussion on the PV calculation is closed or not, but I was doing some odd calulations trying to come up with a relative armor value for each increase in TMM.

First off, I assumed medium range with 4 skill as our average To Hit number. This gave me the percentage chance of getting hit based on 2d6 for each increase in TMM.

T#   % Hit
6   72.22
7   58.33
8   41.66
9   27.77
10   16.66
11   8.33
12   2.77

Then I assumed an average of 5 rounds or attacks per game and used a 0 TMM as the base. That gave me a number of extra hits per game.

Diff %
0
13.89
30.56
44.45
55.56
63.89
69.45

The difference in percentage gave me a relative number of hits per game.

Hits in 5 Rounds
0
0.7
1.5
2.2
2.8
3.2
3.5

Then I assumed an average of 3 damage per unit on the battlefield so that gave me a relative value of armor.

AVG Eqiv Armor
0
2.1
4.5
6.6
8.4
9.6
10.5

So if armor is worth 2 PV each that gives us a TMM PV Cost table that looks like:

TMM/PV
0/0
1/4.2
2/9
3/13.2
4/16.8
5/19.2
6/21

So anyway. That's my thought process. IMO you could simplify the calculation to +4 PV per point of TMM. Movement rate might need a separate mod.

[Xotl]

Getting ready to leave the country for the summer, so there's been little in the way of replies for me, but no, the discussion is absolutely not closed: I'll post when it is.

Still actively soliciting feedback.  Hoping to find some time to close this out next week.

[Xotl]

EDIT: removed, thinking it over some more.

[Xotl]

Any chance I could get ARM implemented on your sheet, Joel?  Should be fairly simple, as it's just "ignore the first chance for a critical".

[Joel47]

Yeah, that looks pretty easy. I'll do that over lunch.

[Xotl]

Here's the most recent sheet.  It removes the Glass Jaw mod altogether: I'm no longer convinced it's anything more than additional math in light of the fact that we'll be rebalancing artillery.

https://www.dropbox.com/s/px3fu6669e40jzu/Alpha%20Strike%20PV%20-%20Output.xlsx?dl=0

I'm at the point where I'd largely call this done.  There are some suboptimal units still, but the goal was never to remove that entirely (and there were suboptimal units before; I suspect this will just shift a few of those around), but rather to ensure that the most broken units are fixed (RIP 23-point Gunsmiths; use 'em while you can, Feddies).

I'd very much welcome more testing with this, using either spreadsheet or your own personal approach.  At this point I'm looking for broken matchups, where a unit gets consistently slaughtered, rather than ones that seem off by a few points.  If you find those, I'd welcome your analysis of what's going wrong, and how to fix it.  If something comes up, I'll still look at it.

[sadlerbw]

RIP 23-point Gunsmiths; use 'em while you can, Feddies

Oh I will....I will.

Thanks for all your effort on this Xotl. It is appreciated!

[Joel47]

Version 1.3
Feature: Added handling for ARM sa
Bug fix: CombatUnit.__init__ now *copies* lists instead of using a handle to the list instance. Combat damage (e.g., ARM being used and deleted) does not affect the original list.  :-[
https://github.com/Joel47/AlphaStrikeDamageSimulator

[Xotl]

Cool, thanks!

Could you post a complete list of specials that work the program as of now?

[Joel47]

RFA, SHLD, AMS, RAMS, ARM, CR, CASE, CASEII, BHJ2, BHJ3, RHS, STL, ENE

Used for other specials: LRM, SRM, IF

[ianpelgrim]

I don't know how to run your testprogram however i think that the superheavy mechs are too expensive point wise.
if you compare the Kodiak II standard to the ARS-V1A "Hera" Ares both 69 points. although the ARES has more armor it has less damage and a 2 lower TMM due too being large. you could argue that the Ares could stand still to reduce difference to 1 lower TMM. but that will keep the fight at long range where the damage difference is a lot bigger.
Can anyone run this fight and see if the superheavy almost never wins it?

[Xotl]

I updated the instructions on page 14 on how to use the python program.  If you find that one too intimidating, however, there's still this sheet, which is excellent for pairing off units and much quicker and easier to use (just copy and paste units from the Output spreadsheet I uploaded, set the range and skill levels if you want, and press Go):

https://drive.google.com/file/d/0BxH6vkK1TLalRTNxcVZ0WWJGaGM

Hera:
4"   Armour 15, Structure 10     7/7/3   OV 2, LG

Kodiak II:
8"   Armour 10, Structure 5     9/9/6   OV 2

Kodiak can decide the range.  At long the Kodiak needs 7s (about three hits every five rounds), while the Ares needs 9s (about two hits every seven turns).

Mathematically, at long range we're looking at about 3.5 damage a round for the Kodiak, and 0.8331 damage a round from the Hera.  At that rate, the Kodiak starts inflicting criticals on the Hera on round 5, while by the same time the Hera has done only 3-6 damage.  Generally, the Hera will lose at long range.  Of course, you're not always going to have the luxury of dueling across the battlefield at long range, and at medium it works out differently.

Still, I think you're right that there's a disparity.  The problem is that, while the system does take LG's to-hit penalty into account, it only does so to the point where the unit's defense mod drops down to 0.  In other words, if the unit's defense modifier hits 0, the system stops.  However, at the tabletop the penalty from LG/SLG/VLG is still applied to a unit even if it has 0 TMM, so that effectively it has -1 defense, and that's definitely worse than 0.  PV need to take this into account somehow.

Cheers for pointing this out.  Thanks to you, a correction will make it in.

[ianpelgrim]

Great finally might try to put an ares on the table O0.
Thank you i'll see if i can think up anything else

[Xotl]

Okay, that was a fairly easy fix.  There's only 10 units in the game right now that would have a negative defense mod, but for those that have it, they'll get a 10% discount on their armour and structure (Defense Factor of 0.9).  So the Hera would drop to 65 points.

[Joel47]

Version 1.4
Improvement: Support for LG, VLG, and SLG when calculating effective movement mod
New feature: Added readme.txt with list of supported specials (manually updated) and link to forum thread with Xotl's instructions.

https://github.com/Joel47/AlphaStrikeDamageSimulator

[Xotl]

Thanks Joel.

I've also updated the Output sheet to take into account the LG fix.  You can get it at any of the links I've posted earlier, including the first post of this thread.

[wantec]

Sorry for jumping in at the end here, but I saw the Osteon Prime and Stormcrow Z, both 76 points in the revised scale (if I'm reading the spreadsheet right). Pretty similar stats, except for speed, armor, & structure. In 1000 sims it seems to be averaging 80-100 more wins for the Osteon. Is that within acceptable for now? Just asking since I haven't followed it all that closely.


EDIT: Oh wait, I found one that might be broken simply b/c of the unit, the Charger CGR-1X1, 58 points. The best it does against the other 58-point units is against the HTM-30Z Hatamoto-Godai, where it gets in the mid to low 300 wins out of 1000. Since this is another fast unit, 16", like the Stormcrow, should there maybe be an upper limit to the impact speed can have?

[Joel47]

If you're using my Python script, what range method are you using? I ask, because a 1-v-1 fight (what my script simulates) is different from what we expect to see in game; different range algorithms will give different results, as each makes different assumptions in an attempt to be more "realistic" than a straight up 1-v-1. My script should not be used for more than rough balancing (at least until I get the budget for a dev team and a proper AI algorithm  O0 ).

Edit: Looking at the units in question, I don't see any unsupported specials that would affect the duel and aren't present on both units. That's why I suspect it's just the slight bias towards size present in my script due to its inability to model tactics.

[wantec]

I used the 1v1 sim xls file. I just took the file as-is and plugged in some different units out of curosity.

[Xotl]

The Charger has a lot of non-direct-firepower things gobbling up its PV.  It's not designed for a straight-up firefight, but is going to provide a lot of alternate battlefield utility.  It's targeting artillery, it's co-ordinating lances and adding to Initiative, it's blanketing the field with super ECM, it's detecting even the most hidden of units.  All this has combat value, which is why it costs PV, but the Charger should most definitely lose in a one-on-one fight to a machine designed only to directly kill things.

I'll look at the Osteon vs. Stormcrow, although a 10% variance doesn't sound too bad.

[Xotl]

Also, I've just updated the spreadsheet again.  This has the aero PV correction I made in the PV aero thread; aero players should be pleased to see some nice PV drops for their units.  I also discovered while implementing this that the spreadsheet did not have a value for submersibles, which meant that they weren't being charged for armour at all.  As such, anyone fielding Neptunes and the like is gong to notice some pretty sharp point increases. :)

[Joel47]

Using range_determination = fast_unit_causes_slow_approach:

Osteon Prime: 520
Ryoken (Stormcrow) Z: 415
Ties: 65
Average battle length: 9.0

Using range_determination = fast_unit_minimizes_damage:

Osteon Prime: 534
Ryoken (Stormcrow) Z: 462
Ties: 4
Average battle length: 26.1

Using range_determination = random:

Osteon Prime: 659
Ryoken (Stormcrow) Z: 191
Ties: 150
Average battle length: 4.1

Conclusions:
1. Random range is a terrible range model for comparing units with wildly different movement profiles. (But it's there because it's a good simulation of a crowded, low-terrain battlefield.)
2. fast_unit_minimizes_damage means that in a real world game the fast unit's player had better be careful walking to his car post-game now that he's wasted everyone's time for eight hours.
3. I'm still seeing people complain that tough units cost too much. I think this is an argument against that.

Edit:
4. I bet that win rate swings the other way (slightly) if I ever get around to modeling n-vs-n combat, where groups of units face off. The big units' potential to overkill, leaving damage unused, might be better modeled then.

[Joel47]

As such, anyone fielding Neptunes and the like is gong to notice some pretty sharp point increases. :)

Did anyone else, upon reading that, feel a sinking feeling?  :D

[NeonKnight]

2. fast_unit_minimizes_damage means that in a real world game the fast unit's player had better be careful walking to his car post-game now that he's wasted everyone's time for eight hours.

I'm not a Maths guys, and I'm simply following the thread to be in the knows, but that made me chuckle!

[GoldBishop]

Considering how the Osteon Prime has more Armor than the Ryoken Z has Armor and Structure combined, I'm not surprised at it's win total.

What is disappointing is the cost of that Ryoken Z: its PV jumped a heck of a lot due to that supercharger (speed jump from 12" to 15").

However... I don't think comparing these two is fair, considering both units would actually utilize (differently) the reach of their NOVA equipment.  Probably shouldn't consider this case too closely.

I have a sinking feeling, and its not related to the submarine joke...

[Xotl]

Always good to double check programs by running math ourselves (also make sure you downloaded the latest version of the sheet at the link I posted last, as it fixes some bugs).

In this case, the Ryoken can dictate the range easily.  I would assume it would prefer medium (feel free to run the other ranges to double-check).  That means it needs 7s to hit, while the Osteon needs 9s.

The Ryoken needs 4 hits to win.  Its second and third hits will cause criticals.
The Osteon needs 2 hits to wins.  Due to the amount of damage it inflicts, it will never cause a critical.

The Ryoken needs ~6.8 turns to win, rounding up to 7.  This ignores the two critical chances it gets, which can only swing things further its way, but aren't guaranteed.
The Osteon needs ~7.2 turns to win, rounding up to 8.

The Ryoken generally will win, but it will be very close; given enough games, the Osteon will readily pull out wins of its own.  The revised PVs look very good to me, which make me happy.

I'm also noting that the Osteon has higher IATM ratings than it has standard damage ratings.  I'm surprised that that is possible.

[GoldBishop]

75 PV is a ridiculous amount to pay for a 55-ton Mech.  This cost makes the Ryoken Z "untouchable" considering how many assault mechs - and even a few Superheavies - far cheaper to field.

I expect this level of pricing on the Osteon Prime - all that armor, structure, and crit resistance makes it's PV worth the longevity on the battlefield.

Its not a fair fight in my eyes.  These are specialist Mechs that should use their Nova CEWS to greater effect in their own network (sync'd per the C3/NOVA rules).  The range for them will vary based on ally positions, not their own.

[nckestrel]

The Osteon Prime in the spreadsheet doesn't match the MUL, probably the muL was corrected after the spreadsheet was pulled.

[Scotty]

75 PV is a ridiculous amount to pay for a 55-ton Mech.  This cost makes the Ryoken Z "untouchable" considering how many assault mechs - and even a few Superheavies - far cheaper to field.

When did a 'Mech's size start determining how much a design is worth outside of its PV? ???

[Xotl]

The Osteon Prime in the spreadsheet doesn't match the MUL, probably the muL was corrected after the spreadsheet was pulled.

Thanks.  I've updated the Osteon's stats on the Output sheet, both pre- and post-revision (and also updated the submersible units to accurate pre-revision stats; I've updated the MUL with all these changes, too).