Final Fantasy Tactics: Level Up/Down

[Talaysen]

Not a stat topic, but still good info that I don't think should be wasted.

From mc

This contains a summary of the growth formulae, and how they work in level-up-down.  Following which it contains the percentage increases of level up/down of various classes at a few different levels.  Finally there is an Appendix for the duelling metagame with Maxed Stats FFT characters (since, really, what else are you going to do with all those stats?)  Not sure if this topic belongs in the unranked games (FFT is ranked...but level-up-down FFT is not...barring really, really weird interpretations I guess...).

[size=18]The Stats[/size]

The idea of Level Up Down is to level up in a class with good stats, and then level down in a class with bad stats, and repeat this process a few times to max out your stats.  There are five stats that change when you gain or lose levels in FFT: HP, MP, Speed, PA, MA.  HP and MP have an in-battle cap of 999.  PA and MA have an in-battle cap of 99.  Speed has an in-battle cap of 50.  In addition, there is an internally stored number which will not go higher than 16777215 (which translates into 1024 of a stat given a 100 multiplier).  This means that classes with lower multipliers on HP/MP will not be able to cap them (for instance, Ninja tops out at 716 base HP, 511 base MP).  All other stats can easily pass their in-battle caps of 50-99, given enough patience.

[size=18]Up-Down Formulae[/size]

From the BMG, the bonus you get from levelling up is...

bonus = [current_RX / (C + Lv)]

(Where Lv is the level you are levelling from).

And the amount you lose from levelling down is...

decrement = [current_RX / (Lv + C)]

(Where Lv is the level you are levelling down to).


It's actually easier to think of these in multiplicative terms

new_RX = current_RX * (C+Lv+1)/(C+Lv)

Over, say, three levels this will become...

(C+Lv+3)/(C+Lv+2) * (C+Lv+2)/(C+Lv+1) * (C+Lv+1)/(C+Lv)
= (C+Lv+3)/(C+Lv)

So in general if you are levelling from Lv_low to Lv_high it's...

new_RX = current_RX * (C+Lv_high)/(C+Lv_low)


The idea of level-up-down is to gain stats by levelling up in a class with low C-value, and levelling down in a class with high C-value.  However note that if the C-values are equal, you actually lose stats (at a very slow rate):

current_RX * (C+Lv+1)/(C+Lv) * (C+Lv-1)/(C+Lv)
= current_RX * (1 - 1/(C+Lv)^2)

This is why you use different C values.

(C_up+Lv+1)/(C_up+Lv) * (C_down+Lv-1)/(C_down+Lv)
=1 + (C_down - C_up - 1)/((C_Up+Lv)*(C_down+Lv))

In short, the percentage you gain is (roughly) equal to the difference between the C-values, divided by the level squared.  In other words, it is much, much more effective to level-up-down at lower levels.  This quadratic decay has nothing to do with FFT growth being quadratic--FFT growth is linear.  No, it has to do with you're getting a linear bonus on the way up, AND a linear bonus on the way down, which multiply for a quadratic bonus.  (It is very important to think about the gain from level up down in terms of the percentage or ratio at which you're increasing a stat, because the overall effect is exponential).

When gaining N levels we get....

(C_up+Lv+N)/(C_up+Lv) * (C_down+Lv-1)/(C_down+Lv+N-1)
= N(C_down+Lv-1)/((C_up+Lv)*(C_down+Lv+N-1)) - N(C_up+Lv)/((C_up+Lv)*(C_down+Lv+N-1)) + 1
=1 + N*(C_down-C_up-1)/((C_up+Lv)*(C_down+Lv+N-1))

Very similar to the previous equation, just we multiply by the number of levels we go up...but also divide by a somewhat larger number.  In particular, as N goes to infinity, this becomes....

=1 + (C_down-C_up-1)/(C_up+Lv)

Which means if we levelled up to infinity, and then levelled back down to zero, our stats would not be maxed out (even though they would be maxed at infinity).  Basically, above a certain level, the fact that it's (C+Lv) will cause the Lv to drown out the C, meaning there's no real difference between one growth and another, so going up in one growth and down in another will have no particularly noticeable effect.

[size=18]Up-Down Charts[/size]

HP: up Mime (6) down: Bard/Dancer (20)
1-11: 61.90% bonus (100% efficient compared to low levels)
11-21: 19.12% bonus (36% efficient compared to low levels)
21-31: 9.63% bonus (19% efficient compared to low levels)
31-41: 5.86% bonus (12% efficient compared to low levels)
41-51: 3.95% bonus (8% efficient compared to low levels)
51-61: 2.85% bonus (6% efficient compared to low levels)
61-71: 2.16% bonus (4% efficient compared to low levels)
71-81: 1.69% bonus (3% efficient compared to low levels)
81-91: 1.36% bonus (3% efficient compared to low levels)
91-99: 0.91% bonus (2% efficient compared to low levels)

HP: up Ninja (12) down: Bard/Dancer (20)
1-11: 17.95% bonus (100% efficient compared to low levels)
11-21: 7.61% bonus (44% efficient compared to low levels)
21-31: 4.24% bonus (25% efficient compared to low levels)
31-41: 2.71% bonus (16% efficient compared to low levels)
41-51: 1.89% bonus (11% efficient compared to low levels)
51-61: 1.39% bonus (8% efficient compared to low levels)
61-71: 1.07% bonus (6% efficient compared to low levels)
71-81: 0.84% bonus (5% efficient compared to low levels)
81-91: 0.68% bonus (4% efficient compared to low levels)
91-99: 0.46% bonus (3% efficient compared to low levels)


MP loss: up Mime (30) down: Bard/Dancer (20)
1-11: -11.83% bonus (100% efficient compared to low levels)
11-21: -6.71% bonus (55% efficient compared to low levels)
21-31: -4.31% bonus (35% efficient compared to low levels)
31-41: -3.01% bonus (24% efficient compared to low levels)
41-51: -2.21% bonus (18% efficient compared to low levels)
51-61: -1.70% bonus (14% efficient compared to low levels)
61-71: -1.34% bonus (11% efficient compared to low levels)
71-81: -1.09% bonus (9% efficient compared to low levels)
81-91: -0.90% bonus (7% efficient compared to low levels)
91-99: -0.62% bonus (6% efficient compared to low levels)

MP: up Ninja (13) down Bard/Dancer (20)
1-11: 14.29% bonus (100% efficient compared to low levels)
11-21: 6.25% bonus (45% efficient compared to low levels)
21-31: 3.53% bonus (26% efficient compared to low levels)
31-41: 2.27% bonus (17% efficient compared to low levels)
41-51: 1.59% bonus (12% efficient compared to low levels)
51-61: 1.17% bonus (9% efficient compared to low levels)
61-71: 0.90% bonus (7% efficient compared to low levels)
71-81: 0.71% bonus (5% efficient compared to low levels)
81-91: 0.58% bonus (4% efficient compared to low levels)
91-99: 0.39% bonus (4% efficient compared to low levels)

MP: up Summoner (8) down Bard/Dancer (20)
1-11: 40.74% bonus (100% efficient compared to low levels)
11-21: 14.47% bonus (40% efficient compared to low levels)
21-31: 7.59% bonus (21% efficient compared to low levels)
31-41: 4.70% bonus (13% efficient compared to low levels)
41-51: 3.21% bonus (9% efficient compared to low levels)
51-61: 2.33% bonus (7% efficient compared to low levels)
61-71: 1.77% bonus (5% efficient compared to low levels)
71-81: 1.39% bonus (4% efficient compared to low levels)
81-91: 1.12% bonus (3% efficient compared to low levels)
91-99: 0.75% bonus (3% efficient compared to low levels)


Speed: up Ninja (80) down: anything else (100)
1-11: 2.13% bonus (100% efficient compared to low levels)
11-21: 1.74% bonus (82% efficient compared to low levels)
21-31: 1.45% bonus (68% efficient compared to low levels)
31-41: 1.22% bonus (58% efficient compared to low levels)
41-51: 1.05% bonus (49% efficient compared to low levels)
51-61: 0.91% bonus (43% efficient compared to low levels)
61-71: 0.79% bonus (37% efficient compared to low levels)
71-81: 0.70% bonus (33% efficient compared to low levels)
81-91: 0.62% bonus (29% efficient compared to low levels)
91-99: 0.45% bonus (27% efficient compared to low levels)

Speed Loss: up not Ninja (100) down not Ninja(100)
1-11: -0.09% bonus (100% efficient compared to low levels)
11-21: -0.08% bonus (83% efficient compared to low levels)
21-31: -0.06% bonus (71% efficient compared to low levels)
31-41: -0.05% bonus (61% efficient compared to low levels)
41-51: -0.05% bonus (53% efficient compared to low levels)
51-61: -0.04% bonus (46% efficient compared to low levels)
61-71: -0.04% bonus (41% efficient compared to low levels)
71-81: -0.03% bonus (36% efficient compared to low levels)
81-91: -0.03% bonus (32% efficient compared to low levels)
91-99: -0.02% bonus (29% efficient compared to low levels)


PA: up Mime (35) down: Bard (80)
1-11: 13.58% bonus (100% efficient compared to low levels)
11-21: 9.57% bonus (72% efficient compared to low levels)
21-31: 7.14% bonus (54% efficient compared to low levels)
31-41: 5.56% bonus (42% efficient compared to low levels)
41-51: 4.45% bonus (34% efficient compared to low levels)
51-61: 3.65% bonus (28% efficient compared to low levels)
61-71: 3.06% bonus (24% efficient compared to low levels)
71-81: 2.59% bonus (20% efficient compared to low levels)
81-91: 2.23% bonus (17% efficient compared to low levels)
91-99: 1.57% bonus (15% efficient compared to low levels)

PA: up Ninja (43) down: Bard (80)
1-11: 9.09% bonus (100% efficient compared to low levels)
11-21: 6.67% bonus (74% efficient compared to low levels)
21-31: 5.11% bonus (57% efficient compared to low levels)
31-41: 4.05% bonus (46% efficient compared to low levels)
41-51: 3.30% bonus (37% efficient compared to low levels)
51-61: 2.74% bonus (31% efficient compared to low levels)
61-71: 2.31% bonus (26% efficient compared to low levels)
71-81: 1.97% bonus (22% efficient compared to low levels)
81-91: 1.71% bonus (19% efficient compared to low levels)
91-99: 1.21% bonus (17% efficient compared to low levels)

PA (poor man's style) up Knight (40), down Chemist (75)
1-11: 9.76% bonus (100% efficient compared to low levels)
11-21: 7.02% bonus (73% efficient compared to low levels)
21-31: 5.31% bonus (56% efficient compared to low levels)
31-41: 4.16% bonus (44% efficient compared to low levels)
41-51: 3.36% bonus (35% efficient compared to low levels)
51-61: 2.77% bonus (29% efficient compared to low levels)
61-71: 2.32% bonus (25% efficient compared to low levels)
71-81: 1.98% bonus (21% efficient compared to low levels)
81-91: 1.70% bonus (18% efficient compared to low levels)
91-99: 1.20% bonus (16% efficient compared to low levels)

PA up Ninja (43), down Dancer (50)
1-11: 2.27% bonus (100% efficient compared to low levels)
11-21: 1.59% bonus (70% efficient compared to low levels)
21-31: 1.17% bonus (52% efficient compared to low levels)
31-41: 0.90% bonus (40% efficient compared to low levels)
41-51: 0.71% bonus (32% efficient compared to low levels)
51-61: 0.58% bonus (26% efficient compared to low levels)
61-71: 0.48% bonus (21% efficient compared to low levels)
71-81: 0.40% bonus (18% efficient compared to low levels)
81-91: 0.35% bonus (15% efficient compared to low levels)
91-99: 0.24% bonus (13% efficient compared to low levels)


MA: up Mime (40) down: anything else (50)
1-11: 3.66% bonus (100% efficient compared to low levels)
11-21: 2.52% bonus (69% efficient compared to low levels)
21-31: 1.84% bonus (51% efficient compared to low levels)
31-41: 1.41% bonus (39% efficient compared to low levels)
41-51: 1.11% bonus (31% efficient compared to low levels)
51-61: 0.90% bonus (25% efficient compared to low levels)
61-71: 0.74% bonus (21% efficient compared to low levels)
71-81: 0.62% bonus (17% efficient compared to low levels)
81-91: 0.53% bonus (15% efficient compared to low levels)
91-99: 0.37% bonus (13% efficient compared to low levels)

MA loss: up anything but Mime (50) down: anything but Mime (50)
1-11: -0.33% bonus (100% efficient compared to low levels)
11-21: -0.23% bonus (72% efficient compared to low levels)
21-31: -0.18% bonus (54% efficient compared to low levels)
31-41: -0.14% bonus (42% efficient compared to low levels)
41-51: -0.11% bonus (34% efficient compared to low levels)
51-61: -0.09% bonus (28% efficient compared to low levels)
61-71: -0.08% bonus (23% efficient compared to low levels)
71-81: -0.06% bonus (19% efficient compared to low levels)
81-91: -0.05% bonus (17% efficient compared to low levels)
91-99: -0.04% bonus (15% efficient compared to low levels)

[size=18]In Practice[/size]

If you want an uber character quickly, get yourself some PA.

If you're looking to max everything, there are basically three problem stats: Speed, MA, and MP.  Speed can only be raised by Ninja, MA can only be raised by Mime, and Mime kills MP.

If you're looking for a maxed stats female Ninja... (see appendix for why you might want this class in particular)

Level 1 maxed stats requires:
57 cycles of 1-20 Ninja, 20-1 Dancer
63 cycles of 1-20 Mime, 20-1 Dancer
14 cycles of 1-20 Summoner, 20-1 Dancer

Level 99 maxed stats requires:
34 cycles of 1-20 Ninja, 20-1 Dancer
42 cycles of 1-20 Mime, 20-1 Dancer
10 cycles of 1-20 Summoner, 20-1 Dancer

While MA does grow faster than Speed with 1-20 cycles, it also has a longer way to increase.  (MA needs to double 3-5 times; Speed needs to double 1.5-3 times).  MP needs to double 4 times/7 times depending on level 1/level 99, but it's much faster to double (nearly every Summoner cycle).

Now, realistically you'll probably do shorter cycles with Summoners (1-10, say) and longer cycles with Ninjas (1-30), as MP growth is very front loaded, and Speed growth is not very front loaded, but this gives a rough idea.

[Talaysen]

From mc

[size=18]Appendix: Max Stats Duelling Metagame[/size]

Concentrate Ninja

Ninja
Concentrate
Teleport
Talk Skill
MP Switch
(Setiemson, Ribbon, Earth Clothes)

The idea is to have Concentrate and Two Swords, which essentially cuts through any defences (Hammedo and Golem would be the exceptions, and they have issues).  4 move = kills units 5 panels away.  Add Talk Skill for Persuade (range 3 you die now) = kills units at 7 panels away.  Add Teleport = kills units 12 panels away 50% of the time (and the other 50% the opponent still needs to bridge 12 panels).

Meliadoul/Orlandu

Holyswordsman/Divine Knight
Talk Skill
Teleport
MP Switch
Maintenance
(Germinas Boots, Excalibur, Escutcheon II, Ribbon/Grand Helmet, Robe of Lords/Chameleon Robe)

(Note: there's also a Two Swords variant which...loses to other Meliadoul/Orlandus, but has a couple of advantaged matchups).

The idea is to have move 5 and Teleport, which in combination with Persuade makes for a 13 panel kill 50% of the time.  Basically sit around Persuade/waiting until you have 140 CT, Persuade-Wait (60CT vs 0CT), Break their Haste Item-Wait (55 CT vs 75 CT), and doubleturn (130 CT vs 125 CT).  Alternatively, if they cast Haste on themselves instead of using Auto-Haste, Persuade 30 times until it wears off.  Obviously this plan doesn't work if they have Maintenance or two Auto-Haste items.  If you're wondering why this trick doesn't work even without killing their haste,  it's because you can have at most 60 carryover CT from your previous turn.  From the BMG:

If unit moved and acted -->  new_CT = min{60, CT - 100}
If unit moved but did not act --> new_CT = min{60, CT - 80}
If unit acted but did not move --> new_CT = min{60, CT - 80}
If unit neither moved nor acted --> new_CT = min{60, CT - 60}

On Orlandu's equipment, Grand Helmet stops Sleep (critical against all the Talk Skill) and Chameleon Robe stops Death (against Talk Skill you do need to be worried about Death Sentence).  He's still vulnerable to Berserk.

Anti-Meliadoul

Maintenance +  Hammedo = WIN

More realistically (since Hammedo leaves you wide open to Holy Explosion/Concentrated Throw Chaos Blade, neither of which are survivable without some trick on top of Robe of Lords...such as Defence Up or bad compatibility)....

Maintenance + MP Switch + Evasion

But...wait, both Meliadoul and Orlandu have this already!  Well the trick is to use it with a skillset that doesn't get shut down by Maintenance or Evasion (like, say, Elemental).  But wait, Orlandu has a skillset like that already!  Well the trick is to use this setup on someone status-immune.  But wait, he can null Berserk if he sacrifices a bit of movement, and heal MP!  Okay, so this setup has to be prepared for a potentially longer haul (and it's not beating Concentrate Ninja anyway) so...

Geomancer
Maintenance
MP Switch
Move MP Up
Talk Skill
(Escutcheon II, Setiemson, Ribbon, Robe of Lords)

Note that Talk Skill isn't really needed here (other than to Berserk stray Orlandus...which are probably handleable anyway) so theoretically you could do something like Summon Magic for Golem to give an outside chance against Concentrate Ninja.  (Though, actually thinking on it it's probably a very bad matchup even if Golem resolves successfully--basically relying on Teleport failing since Persuade stops a lot of CT manipulation tricks that might have worked, and otherwise they both kill in two hits).

Anti Orlandu

Mostly you take advantage of hitting him with status.

Mustadio
Maintenance
Draw Out
Finger Guard
Teleport
(Mythril Gun, Feather Mantle)

First of all, Haste self with Draw Out.  The idea here is to kill Orlandu with Arm Aim (since he WILL be using Grand Helmet--Mimic Daravon so badly rapes him otherwise).  Move 3 + 8 range is enough to pass Move 5 + 5 range (Holy Explosion), so this will be advantaged (though obviously you can raise the percentage to a slaughter with MP Switch and Germinas Boots if you wanted).  The reason for Finger Guard, Feather Mantle is to have a decent chance against Meliadoul (Maintenance stops Mighty Sword, Finger Guard stops Berserk with Mythril Gun, then Kikuichimoji and hope she either fails her Teleport or misses through Feather Mantle).

The other Anti-Orlandu move is Steal Heart, which tends to be a win and has the same range as Persuade but better vertical tolerance.

Two Swordsing Meliadoul/Orlandu

Same as above, just replace Excalibur+Escutcheon with...

Chaos Blade+Excalibur

The reason for Excalibur in the second hand is that it doesn't instant-lose to Persuade Lock+Break Autohaste (because it takes two breaks to kill haste).  The damage difference isn't worth worrying about, though as it really comes down to nobody can set both Maintenance and Defence Up, so unless we bring compatibilty into this, Excalibur's dealing 999 so worrying about Chaos Blade's damage is silly.

At any rate, this setup mostly exists to do better against the anti-Meliadoul/Orlandu setups (and does worse against the Maintenance Meliadoul/Orlandu setups).

Other

Dance/Math Skill are good at preying on people who didn't max out MP.  Math Skill also has some use for preying on people who are not using Talk Skill (eventually Pray Faith will hit)--particularly people who are using Golem.

Quick the idea would be to get two turns in one clocktick thereby doubleturning your opponent.  I just don't know if it works, though (slow actions would need to resolve after you've just gotten a turn--so I guess slip behind your opponent in CT so that he acts first on a clocktick, move, and then get an additional turn during the following slow action phase?  Your opponent can also see this coming and manipulate their CT.  Potentially insane, just confusing so I haven't figured out if it works, plus a little slow).

Jump You can use it to be invincible for a turn...but your opponent will nearly always be able to double-wait and get the next turn when you'll be on the ground.  Otherwise...range 8 and 999 is...mediocre, but then so is Elemental which was included in a setup (four move class and can status Orlandu would be the clear advantages there; the subtle advantages are not being shut downable by CT manipulation at times, and gets stopped by Blade Grasp which...nobody ever uses, but still).