MACC Debate (From Random Thread)

[pchan]

1) to reconcile the apparent phenomenon of having the effects of changes in skill/macc/INT be halved below the overall 50% "effective" magic accuracy (or magic hit rate, EMAC for now, apologies if you don't like my acronym) "threshold" and unaffected above 50% EMAC.

2) to explain the effect of HQ elemental staves crossing the 50% threshold. So this model has a basis, although I am not proving the model is true but forwarding an explanation for the data.

That's what I didn't understand with the previous "50% threshold". HQ staves would give either 15% or 30%, while a substancial number of my tests provide intermediate values.

This threshold is meant to be independent of whatever spell you cast. I did not make any statement about every spell starting from a "base" EMAC, which seems to be a point of contention here.

The way I chose to explain this apparent phenomenon is to devise a hypothetical formula based on a difference of "magic accuracy" (MACC) and "magic evasion" (MEVA), both of which are meant to be catch-alls for positive and negative contributions to EMAC without worrying about specifics. 50% EMAC could then be a critical point where MACC - MEVA = 0. Where MACC - MEVA is negative, any changes in MACC (or MEVA) are cut in half.

The model is not really dependent on the spell. Like we've all said (except some people), some spells are more or less accurate. This is obvious for paralyzes since just having paralyzeII in you spell list implies a +2 mac at least.

Moreover, I continued to assume an additive model for EMAC, as opposed to a multiplicative one, where any distinct factor that affects EMAC has its own term and all the terms are added (or subtracted) from each other to obtain the final result. The additive model assumes that each factor is "independent" from one another and has no effect on the other factors. Of course, the additive model can break down pending demonstrable interaction among some of these factors. (Yet based on the data, the long-held idea of an interaction between staff effect and skill is not very credible.)

The additivity of the model is easily checked by reaching the target's magic eva with very different combinations of skill/magic accuracy.

Examples of distinct factors, whether contrived or not, include level adjustment (correction), elemental resistance, resist traits, INT, etc

I strongly believe that level correction and elemental resistance or spell resistance (like resist gravity/death or barfire, barpetrify) translate into + magic eva to a specific spell/element. Note that in the melee hit rate formula, the -2% hitrate per level can be equivalently seen as +4 evasion per level difference.

Also, I cite the melee analogue that is oriented about 75% hit rate, not as proof of the additive EMAC model but as an example where this threshold idea seems to be accepted.

I doubt that the melee model is totally accurate anyway, as there is no agi check. The check could be hidden in the "eva" number however.

2. Is a model for "effective" magic accuracy practical?

If the model(ette) is true, perhaps you can estimate an "effective" MEVA value for some mob of interest. This can be useful for hypothetical garden-variety testing of new equipment or abilities (uh, good luck finding more than a handful of people to do that with satisfactory sample sizes... credit to pchan) with unquantified EMAC (or MEVA) bonuses. If what we think is 391 total MACC (here, MACC is the catch-all) really corresponds to 50% EMAC for a specific mob that is somewhat convenient for testing (like a Qiqirn ranger), then we can avoid crossing this threshold and confounding our results.

In the spreadsheet, there are plenty of different tests with 1k sample size. Each gives you a landing rate and an error. When you solve the equation backwards you find an interval for magic evasion. For example : line 73 has 281 skill, 22 mac, dint=+10. According to the model this gives 313 magic acc. We observe 70.9% landing rate with 2.66% error. This means that the actual landing rate is between 68.24% and 72.56% (with 95% probability, whatever that means). The model then suggests a magic eva between -18.24 and -22.56 so
it gives a range for meva in [290.44,294.76]. You can repeat this for all the lines and you find various intervals, all of them contain 291 and 292.

4. Use regression

This is regarding the generated graph. Of course, confidence intervals are based on the data

I already adressed that, the error I give is 3% which is the maximum you can get regardless of the data.

But, you don't need to generate confidence intervals for each point.

You do. Every test has it own error and is used to determined an approxiamtion of meva.

Just use linear regression based on the assumption of piece-wise linearity where appropriate. Using this reasonable assumption will make the interval estimates of EMAC, given input level, more efficient than interval estimates that ignore the assumption. Then you can also check for violations of regression assumptions and lack of fit to the data.

I don't think that using linear regression in the right idea, as the main part of a test would be to prove the non linearity, as it is the main difference when compared to melee hit rate. Proving that the hit rate is piecewise linear by already assuming it will behave differently depending on a certain value of mac (that you have to find somehow ..) is not exactly proving piece wise linearity.

5. Not controlling for Qiqirn level

Technically speaking, this is not very important assuming that the samples are balanced with respect to level differences (and assuming no other potentially confounding factors) because we are concerned with the change in EMAC with skill level. If that seems unpersuasive, think of randomization in clinical trials.

The qiqrn levels are controlled, there are only two types of them. Assuming that the levels are are balanced for a give INT would just add unecessary noise. The fact that all the qiqrn that had 65 INT gave the same XP is enough evidence to me. The same was done on my sleep tests on imps.


To summurize what is left to do (to me)

- quantify level difference

How much magic eva per level. This could be tricky because changing levels will/can also change the INT of the mob.

- quantify elemental resist.

I have some idea for a test. It would be based on getting nuked by friar's lanterns. We need a formula like 5 fire resist = 1 meva or something like this.
- more investigation about int

My test seem to show that stacking int does not add much acc after a point. I've been working on that.

EDIT: Based on what Thorny says, I don't know now. As Robonosto mentioned, the difference should be small given that Pchan nuked relatively equal numbers of 76 vs 77 mobs, but would be nice to have a 'perfectly executed' data set of 40,000.

That's completely irrelevant TBH, as the 3 lines with 184 XP have less than 300 sample size. If anything it gives an idea about level correction.

[Argettio]

I don't have much to add, this topic has moved along a lot since I went to bed.

The data set that pchan has now posted is being checked by people smarter than me, and it seems to be holding water.

Level correction : If it does anything , it's modifying the magic evasion, not the effect on INT on resists.

Correct me if I am wrong
*Resists happen when the target of the spell has 'too much' magic evasion.
*Increasing INT increases your magic evasion (and therefore your resist rate)

So if your magic evasion is 'corrected' then any effect from your INT will also be corrected.

The best analogy I can think of is DEX on melee accuracy. You have to sum all you skill, acc and dex gear to get your total accuracy and then that number is corrected.

But I think I will sit back and see where this goes, if nothing else pchan has provided some useful data that warrants attention (to which he deserves some thanks).

Unlike some people I don't hate on pchan, I am not a fan of his posting style and have disagreed with his view point at times, but I am not here to troll. I just wanted to voice my reasons for not accepting this model (yet) and continue the discussion.

[Argettio]

How much magic eva per level. This could be tricky because changing levels will/can also change the INT of the mob.

Sorry for double post.

If you can establish the INT value of both the high and low level mob you can use burn to adjust the INT of the high level mob to match that of the lower level mob.

* 1 - 39 INT = 1hp/tic (-5 stat)
* 40 - 69 INT = 2hp/tic (-7 stat)
* 70 - 99 INT = 3hp/tic (-9 stat)
* 100 - 149 INT = 4hp/tic (-11 stat)
* 150 - ??? INT = 5hp/tic (-13 stat)

You could reduce the mobs INT by 9-13 as a lvl 75. That should allow enough control.

EG your low lvl mob has 65INT and your highier level mob has 67INT. You can burn the lower lvl one with < 99INT and the higher one with 100-149INT.

They will both then have 56INT.

Not perfect, but a start

[pchan]

I don't have much to add, this topic has moved along a lot since I went to bed.

The data set that pchan has now posted is being checked by people smarter than me, and it seems to be holding water.



Correct me if I am wrong
*Resists happen when the target of the spell has 'too much' magic evasion.
*Increasing INT increases your magic evasion (and therefore your resist rate)

So if your magic evasion is 'corrected' then any effect from your INT will also be corrected.

Yes and no. Yes if lowering the magic evasion makes you cross the red lines "5%" "50%" and "95%".

[CDF]

All right, so I checked some of the "Stone" sheet today just to check a few things.

1. Appropriateness of the additive EMAC model below 50%

I used rows 3-17 to check for a linear relationship between the computed MACC score (comprised of delta(INT), elemental magic skill, magic accuracy, and staff bonus) and EMAC below 50% and got the following results.

                            Data Set                      WORK.QIQIRN
                            Distribution                     Binomial
                            Link Function                    Identity
                            Response Variable (Events)        success
                            Response Variable (Trials)          total

                             Number of Observations Read          28
                             Number of Observations Used          15
                             Number of Events                   4579
                             Number of Trials                  12227

                             Criteria For Assessing Goodness Of Fit

                  Criterion                 DF           Value        Value/DF

                  Deviance                  13         11.9312          0.9178
                  Pearson Chi-Square        13         11.9303          0.9177

                                 Analysis Of Parameter Estimates

                                    Standard     Wald 95% Confidence       Chi-
    Parameter     DF    Estimate       Error           Limits            Square    Pr > ChiSq

    Intercept      1     -0.9692      0.1050     -1.1750     -0.7633      85.12        <.0001
    macc_score     1      0.0051      0.0004      0.0043      0.0058     162.95        <.0001

For this data, delta(INT) was less than or equal to 10. The estimate for the MACC score is consistent with the idea that any bonuses (or penalties) are cut in half below 50% EMAC.

The linearity assumption appears valid (as we would hope). There could be problems if level adjustment came into play.

2. Appropriateness of the additive model, broken out by specific factors

Since we are interested in model checking, we can break down the MACC score by its constituent parts and see how EMAC varies with changes in each.

                           Data Set                      WORK.QIQIRN2
                           Distribution                      Binomial
                           Link Function                     Identity
                           Response Variable (Events)         success
                           Response Variable (Trials)           total

                             Number of Observations Read          15
                             Number of Observations Used          15
                             Number of Events                   4579
                             Number of Trials                  12227

                             Criteria For Assessing Goodness Of Fit

                  Criterion                 DF           Value        Value/DF

                  Deviance                  11         11.0852          1.0077
                  Pearson Chi-Square        11         11.0933          1.0085

                                 Analysis Of Parameter Estimates

                                      Standard     Wald 95% Confidence       Chi-
  Parameter         DF    Estimate       Error           Limits            Square    Pr > ChiSq

  Intercept          1     -0.0879      0.0869     -0.2581      0.0823       1.02        0.3115
  macc               1      0.0049      0.0005      0.0039      0.0058     108.91        <.0001
  int               1      0.0057      0.0013      0.0031      0.0083      18.05        <.0001
  staff        1     1      0.1600      0.0164      0.1278      0.1922      94.70        <.0001

The parameter estimates are consistent with previous work based on EMAC being below 50%. The HQ staff bonus is around +15% EMAC and the magic accuracy effect is about +0.05% per 1-point increase.

Recall that lodeguy did not check for the effect of INT below 50% EMAC. Here, the effect of INT appears to be halved as well.

3. Estimating observed magic evasion

You could either obtain individual MEVA estimates for each value of MACC score or base MEVA estimates on the assumption of a linear relationship between MACC score and MEVA (based on the regression model and the estimates of slope and intercept).

In either case, this involves using the estimate of the true EMAC value for a given MACC score. (These estimates will be different depending on which approach you take.)

I noticed that the interval estimates for MEVA (pchan's approach) are 93.57% confidence intervals (z-score of 1.85 used). Why did you use that? At any rate, the 95% confidence intervals are slightly wider.

The other approach uses the information from all other samples (summarized in the slope and intercept estimates) to provide narrower interval estimates, so they are generally more efficient than estimates treated in "isolation." Here are the results using this approach:

Score     E^[Y|X]     E^[MEVA|X]    CI lower     CI upper
---------------------------------------------------------
240       0.24777     290.446       286.308      294.584
246       0.27819     290.362       287.054      293.67
251       0.30354     290.292       287.622      292.962
256       0.3289      290.22        288.09       292.35
261       0.35425     290.15        288.376      291.924
265       0.37453     290.094       288.39       291.798
266       0.3796      290.08        288.358      291.802
268       0.38974     290.052       288.254      291.85
270       0.39988     290.024       288.102      291.946
271       0.40495     290.01        288.01       292.01
274       0.42017     289.966       287.684      292.248
276       0.43031     289.938       287.436      292.44
277       0.43538     289.924       287.306      292.542

This is what I meant when I said "use regression."

4. Appropriateness of the additive EMAC model above 50%, broken out by specific factors

I am focusing on rows 75-86 here. This data is more problematic because delta(INT) exceeds 10 for some samples. But, if I recall correctly, INT remains the same regardless of Qiqirn level. Also, there appear to be two different levels of qiqirn involved. In particular, rows 75-78 are based exclusively on a single qiqirn that gave 230 EXP. I assume the other qiqirn were of a different level.

We think that the effect of INT is halved for delta(INT) >= 10, so let's see if that's the case here by excluding the two samples with INT = 65.

                             Number of Observations Read          10
                             Number of Observations Used          10
                             Number of Events                   6571
                             Number of Trials                   9790

                             Criteria For Assessing Goodness Of Fit

                  Criterion                 DF           Value        Value/DF

                  Deviance                   5          7.5070          1.5014
                  Pearson Chi-Square         5          7.4390          1.4878

                                 Analysis Of Parameter Estimates

                                      Standard     Wald 95% Confidence       Chi-
  Parameter         DF    Estimate       Error           Limits            Square    Pr > ChiSq

  Intercept          1      0.2456      0.0683      0.1118      0.3793      12.94        0.0003
  macc               1      0.0112      0.0007      0.0098      0.0127     231.28        <.0001
  int                1      0.0034      0.0007      0.0021      0.0048      24.76        <.0001
  staff        1     1      0.2972      0.0217      0.2546      0.3398     186.92        <.0001
  level        1     1      0.0063      0.0168     -0.0266      0.0392       0.14        0.7067

There are some borderline cases here from the standpoint of statistical significance, such as macc and INT. Checking the residuals, the Alkalurops data (row 84) seem "extreme." Interestingly, the level adjustment effect is not close to significance (are rows 76-78 for a different level qiqirn than the other rows?).

Of course, the identification of delta(INT) = 10 as a critical point was done informally. Let's try applying the model for samples where delta(INT) <= 10. (I removed the level effect from the model.)

                             Number of Observations Read           5
                             Number of Observations Used           5
                             Number of Events                   2725
                             Number of Trials                   4424

                             Criteria For Assessing Goodness Of Fit

                  Criterion                 DF           Value        Value/DF

                  Deviance                   1          0.4115          0.4115
                  Pearson Chi-Square         1          0.4118          0.4118

                                 Analysis Of Parameter Estimates

                                      Standard     Wald 95% Confidence       Chi-
  Parameter         DF    Estimate       Error           Limits            Square    Pr > ChiSq

  Intercept          1     -0.4354      0.1674     -0.7635     -0.1074       6.77        0.0093
  macc               1      0.0088      0.0019      0.0051      0.0125      21.97        <.0001
  int                1      0.0127      0.0022      0.0085      0.0170      34.09        <.0001
  staff        1     1      0.2833      0.0190      0.2460      0.3205     222.08        <.0001

INT is indeed more efficient below delta(INT) 10 than above.

5. Appropriateness of the additive EMAC model above 50% based on MACC score

Using the MACC scores, there is a statistically significant lack of fit to all twelve samples (p-value .0026... compared to the saturated model) even accounting for level difference (p-value .0701).

So, it may be worth checking model fit to the data where delta(INT) <= 10 and where delta(INT) >= 10. Start with the first case where delta(INT) <= 10:

                             Number of Observations Read           5
                             Number of Observations Used           5
                             Number of Events                   2725
                             Number of Trials                   4424

                             Criteria For Assessing Goodness Of Fit

                  Criterion                 DF           Value        Value/DF

                  Deviance                   2          4.5021          2.2510
                  Pearson Chi-Square         2          4.5427          2.2713

                                Analysis Of Parameter Estimates

                                      Standard     Wald 95% Confidence       Chi-
 Parameter          DF    Estimate       Error           Limits            Square    Pr > ChiSq

 Intercept           1     -2.5469      0.1985     -2.9360     -2.1578     164.60        <.0001
 macc_score          1      0.0104      0.0006      0.0092      0.0117     268.48        <.0001
 level         1     1      0.0136      0.0162     -0.0181      0.0454       0.71        0.3995

And for the second case with delta(INT) >= 10:

                             Number of Observations Read          10
                             Number of Observations Used          10
                             Number of Events                   6571
                             Number of Trials                   9790

                             Criteria For Assessing Goodness Of Fit

                  Criterion                 DF           Value        Value/DF

                  Deviance                   7         20.7036          2.9577
                  Pearson Chi-Square         7         20.4193          2.9170

                                      Standard     Wald 95% Confidence       Chi-
 Parameter          DF    Estimate       Error           Limits            Square    Pr > ChiSq

 Intercept           1     -2.7372      0.2020     -3.1331     -2.3413     183.64        <.0001
 macc_score          1      0.0110      0.0006      0.0098      0.0123     294.82        <.0001
 level         1     1      0.0335      0.0148      0.0044      0.0626       5.10        0.0240

While superficially the MACC score is not really different in either case, there is a lack of fit to the data given delta(INT) >= 10. Checking the residuals, row 84 (Alkalurops) stands out like a sore thumb (std. residual above 3). Removing the Alkalurops data gives the following result:

                             Number of Observations Read           9
                             Number of Observations Used           9
                             Number of Events                   5760
                             Number of Trials                   8785

                             Criteria For Assessing Goodness Of Fit

                  Criterion                 DF           Value        Value/DF

                  Deviance                   6          8.7710          1.4618
                  Pearson Chi-Square         6          8.7827          1.4638

                                Analysis Of Parameter Estimates

                                      Standard     Wald 95% Confidence       Chi-
 Parameter          DF    Estimate       Error           Limits            Square    Pr > ChiSq

 Intercept           1     -2.6206      0.2091     -3.0303     -2.2109     157.14        <.0001
 macc_score          1      0.0106      0.0007      0.0093      0.0119     254.43        <.0001
 level         1     1      0.0349      0.0149      0.0056      0.0642       5.46        0.0194

6. Conclusion

The so-called additive EMAC model looks pretty good below 50% EMAC. It seems ok above 50% EMAC with some caveats about level correction, INT, and Alkalurops. At any rate, the data support this sharp dichotomy oriented about 50% EMAC.

Here, I didn't provide an explanation for why the Alkalurops data seem unusual relative to the rest of the data (above 50% EMAC). I mean, the MEVA estimates seem consistent (around 290). I would have to look at it later.

If you think I am looking at the data incorrectly, please provide an explanation. I already checked to see if I made any input errors.

[Yugl]

What's the best piece for a resistance set:

INT+5 or INT+2 Magic ACC+2?

[Ultimagi]

What's the best piece for a resistance set:

INT+5 or INT+2 Magic ACC+2?

Magic Accuracy +2 will provide more accuracy against weaker enemies.
Intelligence +3 will provide more accuracy against harder enemies, which matters more.

[pchan]

Still trying to decrypt what you said because I basically stopped statistic at the central limit theorem.

I noticed that the interval estimates for MEVA (pchan's approach) are 93.57% confidence intervals (z-score of 1.85 used). Why did you use that? At any rate, the 95% confidence intervals are slightly wider.

Yeah I don't know the value by heart, the correct value would be 1.96 so 6% wider, so a maximum error of +/- 3.1%.

Recall that lodeguy did not check for the effect of INT below 50% EMAC. Here, the effect of INT appears to be halved as well.

I wasn't aware of that but the fact that INT behave the same wether you are under 50%, above 50% or you cross the line is strong evidence that INT translates into magic accuracy and not hitrate.

here are some borderline cases here from the standpoint of statistical significance, such as macc and INT. Checking the residuals, the Alkalurops data (row 84) seem "extreme." Interestingly, the level adjustment effect is not close to significance (are rows 76-78 for a different level qiqirn than the other rows?).

This row 84 gave me trouble also. It works as if the last 10 INT add 1% hit rate while they are supposed to give 0.5% since you go from dint=10 to dint=20. At one point I thought that adding magic accuracy to your gear would increase the amount of INT that can give 1% hitrate (or 1 mac). In the end I just assumed the error was significant. So while you measure 80.7%, since you get 2.3% error, you could very well get 78.4% landing rate which is aprrox 10% more hit rate that row 81, which has -10 mac.


You also made some comments about qiqrns being of different levels.
I'm 100% positive that all the qiqrns, except row 18-20, had all the same INT. The first lines were done long ago, so I'm just assuming that I forgot to check the INT for those 3 lines, but still kept the results. When I didn't fill the "exp. " column It just means that I knew I had the correct level.

[CDF]

Well, I don't know if anyone was interested in the enfeebling stuff, but with sleep resists following a trinomial distribution, the observed samples are consistent with the probability distribution p, p(1-p), and (1-p)^2 for no, half, and complete resists. With this in mind, it seems probable that stuff like Bind, Gravity, and Paralyze may have three levels of resist, too, just easily obscured by the fact that their durations are themselves random.

Sure, maybe all this stuff on enfeebling is "common knowledge," but it's pretty nice to have some supporting data sets.

[Callisto]

Regarding Paralyze, I still have yet to have time to go back out and get a reasonable sample size since I lost my initial batch of testing on Steelshells, but I found it to have a very consistent duration when I was actually timing it. I want to say it was 2:00 or 2:30 using Para II Lv. 3...I could see the 3 tier thing working there as well, though Bind is really funky.

I'm sure I'm not the only one who has seen Bind randomly eat 1k+ damage without breaking on an EM-T mob like Groundskeepers, yet if I do so much as 1 damage on something like Despot/Ulli it always breaks immediately...assuming part of that is a level correction, but I definitely thing there's more to that spells in terms of 'full landing' it.

[pchan]

Na, I used both max enfeebling and max mnd gear for paralyze and cast on a tough qiqrn, and the duration was random. Now it's possible that you can get a half resist which duration is random .. gl testing that.

[Uzor]

Could be like bind and be the potency or something. Iunno, never thought about Paralyze getting partial resists, but its possible I guess.

As was mentioned earlier in the thread, Bind does seem to have partial resists. On normal mobs Bind can sometime withstand 1000 damage before it breaks, and other times a Bio breaks it. On Suzaku my bind usually didn't last very long, except for once when it lasted through a Tier-4 spell. Bind might be random as well though, and this was way back when RDM was my only 75 job.

[CDF]

When I speak of Bind being partially resisted, I meant its duration possibly being cut in half like how magic damage is cut in 1/2, 1/4, or 1/8 when you get a resist. Of course, it can be outright resisted, too. This is based on my current perception that the variability of duration is not that large, but can appear that way when resists come into play.

Anecdotally speaking, it is not terribly uncommon for Bind to wear off in under 10 seconds on Despot when I am soloing it (BLM). The (un)reliability of Bind here can make the difference between a sub-15 minute attempt and one that stretches out to say 30 minutes.

This would also be analogous to Aspir, where it's pretty clear that resists manifest as poor results on, say, Kaiser Behemoth (Apollyon), like 16 MP drained, even though the amount of MP drained seems to be random (uniformly distributed perhaps).

The range of MP drained without a resist certainly is not between 16 and 111 for KB though.

As for the amount of damage to break a Bind, I would "wildly" speculate that there is a possibly non-fixed probability (perhaps depending on level difference) of breaking the bind upon inflicting some direct damage. It would be interesting if skill or INT actually affected this, too.

[pchan]

I went in halvung to test what fire resist do exactly. I only have some weak sample size but still, it's pretty clear that 1 fire resist equals to 1 magic evasion. Basically, I would get nuked by friar's lanterns, while having the same int as theirs.

getting nuke 50 times by burn/firagaIII/flare/fire4 with zero fire resist : not a single resist.

getting nuked 50 times with 91 fire resist (barfire) : not a single resist either

101 fire resist : ~92% landing rate

This means that 95% happens at around + 100 fire resist.
137 fire resist : 63% landing rate for the bomb
168 fire resist : 39%


Interestingly, going from 101 to 168 brought the landing rate from 95% to 39%. In theory, you would neeed 67 magic evasion for that, and going from 101 to 168 is also 67 ..lol. So I'm convinced (even though sample size is low) that 1 fire resist equals to +1 magic evasion. With this in mind you would need an additionnal 68 fire resist to cap : 236 fire resist and you resist all (95%) of those level 76 bombs spells.

[Ultimagi]

I'm sure I'm not the only one who has seen Bind randomly eat 1k+ damage without breaking on an EM-T mob like Groundskeepers, yet if I do so much as 1 damage on something like Despot/Ulli it always breaks immediately...assuming part of that is a level correction, but I definitely thing there's more to that spells in terms of 'full landing' it.

The spells you're using to do that "1" damage, probably Dia or Bio, always break Bind.

[Yugl]

The spells you're using to do that "1" damage, probably Dia or Bio, always break Bind.

I remember seeing someone say this in an old Standard thread, and the poster removed their statement after they were PMed a pic of someone casting Bio without breaking bind.

Edit: Found the thread

http://www.bluegartr.com/forum/76258...ml#post2798279

[Ultimagi]

My Bind doesn't break to Blizzard III very often, but always breaks to Bio II, but whatever.