Username:
B
I
U
S
"
url
img
#
code
sup
sub
font
size
color
smiley
embarassed
thumbsup
happy
Huh?
Angry
Roll Eyes
Undecided
Lips Sealed
Kiss
Cry
Grin
Wink
Tongue
Shocked
Cheesy
Smiley
Sad
page  <- 12345678910111213141516171819 -> <- 1 .. 16 .. 19 ->
--
--
List results:
Search options:
Use \ before commas in usernames
Is the Death Necklace strat intended to replace the gold glitch?  If so, then you'd have to do without the Full Plate until level 9 or 10, which means no Rimuldar grinding until then... you could try grinding there, but it's extremely risky until the critters there are in the safe zone.  My segment had a >50% mortality rate at level 7 there, for instance.  And it seems rather likely that eight levels of less optimal grinding around Kol would lose all the time you gained from skipping the gold glitch.

Regarding the spreadsheet, it's too rudimentary at the moment to be useful for drawing conclusions.  Here are some things that stood out to me on looking it over: whatever the basis for the "Heal time" column, it's the same between levels sixteen and seventeen, even though you only have Healmore for the latter; the "Heal time" column sometimes becomes negative, but is still added to the total time; for Starwyverns and Werewolves, the "HP lost" column drops drastically between levels sixteen and seventeen, even though the average numbers of turns is constant (no strength gain) and neither creature moves into the safe zone at that level; the "Total: fight" calculation has a typical fight (with no ambush) having the monster take one more turn than the player, when it should be the opposite.  The spreadsheet would need much improvement before you could start relying on it.

Anyway, you can rule out Zone 17 pretty quickly.  No monster there is in the safe zone, so you'd be losing ~40 HP per battle, and regaining ~16 between battles.  So you'd only be able to go three or four battles between Healmores, taking about a minute per Healmore (there's no convenient desert zone for healing).  This is neglecting herbs, but they're weak in comparison.  So in roughly ten minutes, you'd have to leave the dungeon to heal.  The trip to Charlock's basement takes about four minutes.  So, being generous, that's three extra trips to Charlock, since after all you'd have to heal one more time before attempting the Dragonlord.  Taking the 140 XP/min from your spreadsheet for granted, that's ~22 minutes of grinding plus ~12 minutes of walking, for ~34 minutes to level, compared to the ~25 minutes that seems typical in the Hauksness desert.  And it would probably take more than three trips, since you probably wouldn't be crazy enough to only heal at ~25 HP.

I do appreciate any attempts to make the speedrun more rigorous, though, so hopefully my objections won't be too discouraging.
Edit history:
Crow!: 2014-06-02 05:18:24 am
Crow!: 2014-06-01 09:04:06 pm
Crow!: 2014-06-01 09:01:46 pm
What's that gemma?
Quote from Lhexa:
Is the Death Necklace strat intended to replace the gold glitch?  If so, then ...
Probably, though not necessarily.  It just seems to me that being at 1200G at 5 minutes is wildly better than being at 30G + 11 EXP, or at 8 minutes, being at 1200G + 7 EXP is better than being at 50G + 25 EXP.  (Edit: was off by 50G).

Maybe it would turn out that re-visiting the Mountain Cave later for the gold glitch would still pay off, or maybe there would be some other sensible way to gain the rest of the needed money (some variant of the repetitive mountain cave dive+grind that Feasel tried a while ago comes to mind).  But I'm reasonably certain that gaining the first series of levels would be insanely fast with that much equipment to work with.

Quote:
Regarding the spreadsheet, it's too rudimentary at the moment to be useful for drawing conclusions.  Here are some things that stood out to me on looking it over: whatever the basis for the "Heal time" column, it's the same between levels sixteen and seventeen, even though you only have Healmore for the latter;
Whoa, oops!  For some reason I thought you learned that at level 16.

If I put the player onto grass tiles inside Hauksness (increasing average time to find a battle by 4 seconds but making the area almost safe), the EXP/min goes down to 116, which is less than the desert, which is at 124 (both of these are after the corrections below).

Quote:
the "Heal time" column sometimes becomes negative, but is still added to the total time;
This is alright so long as the player at some point needs to heal.  Here's an exaggerated example to demonstrate: Suppose there was a zone with this set of enemies:
1x Red Slime, 4x Green Dragon
The healing due to Erdrick's Armor leading up to Red Slime fights would contribute toward the healing you need to do to cover the other fights.

Where this breaks down is if you remain perpetually in a safe zone, such that your HP is max and some portion of Erdrick's Armor's healing is wasted.  I have now therefore added a "safe zone?" entry for each zone, and healing will be ignored if the area is a safe zone, rather than magically saving fractions of seconds from thin air.

Quote:
for Starwyverns and Werewolves, the "HP lost" column drops drastically between levels sixteen and seventeen, even though the average numbers of turns is constant (no strength gain) and neither creature moves into the safe zone at that level;
I'm taking my data from this battle simulator.  http://www.woodus.com/den/games/web/dwsim/battle.htm
I didn't write that simulator myself; if there's a serious bug there, then that would be good to know.  Amusingly, the Starwyvern fights actually take longer at level 17 than at level 16.  In case it matters, I'm assuming the name Z.

One thing I'm not accounting for is the battles that go so wrong the simulator predicts the hero losing.  This is quite rare in regions where you'd actually want to grind, but at present I'm not using the statistics from those fights at all.  It's not immediately obvious to me what to do about them.

At Level 16, Starwyverns are listed as losses in about 5% of fights, whereas 0.5% of level 17 fights are losses.  My currently inaccurate method of handling the lost fights might be responsible for the level 17 fights being listed as longer; the worst fights are automatically sorted out for level 16.  The simulator definitely seems to think that level 16 -> level 17 is a defense break point, though, listing damage taken as going from 11-22 to 0-15.

Quote:
the "Total: fight" calculation has a typical fight (with no ambush) having the monster take one more turn than the player, when it should be the opposite.
Oops! That's an easy mix-up to fix.

Quote:
... The trip to Charlock's basement takes about four minutes ...
That's good to know.  That means that every 9 HEALMOREs costs an extra 4 minutes, so that each HEALMORE costs an extra ... 26.7 seconds.  Ouch.

Adding that in, the best predicted EXP/min I'm getting there is ~113 (down from ~139, which is where it was after correcting the player/enemy turn error).  That this involves adding Axe Knights to the "run away" list makes the problem pretty obvious.  It's notably less than the predicted EXP/min in either Hauksness or the desert.

Quote:
I do appreciate any attempts to make the speedrun more rigorous, though, so hopefully my objections won't be too discouraging.
Not at all! Some people take criticism hard, but I'm not one of them.  I want to be right, not to convince people that I was right.
Attachment:
Intruding N313 and F014
How about combining the death necklace and the gold glitch since you would go to the mountain cave to do the gold glitch?  This means that the gold glitch grinding would be reduced by 1210 with my route.  It would have to be done segmented due to luck, but I think it could speed up a segmented run.  I tested going to Rimuldar at level 1 on emulator using only a few save states(exit Brecconary, enter cave, exit cave, enter Rimuldar), and it is possible though it will require lots of luck, of course.  Here is what I think the starting route would look like: (The amounts in parenthesis is the character's GP after that transaction.)

--Segment 1--
go to Brecconary
sell torch (124 GP)
buy 2 herbs (76 GP)
buy dragon scale (56 GP)
equip dragon scale, but don't sell it yet
mad dash to Rimuldar hopefully using only 1 herb before the passage cave which is required at level 1
buy a key (3 GP)
die (1 GP)
save

--Segment 2--
go to the mountain cave (using the other herb if necessary)
pick up the herb and the DEATH NECKLACE (use that herb if necessary)
die (0 GP)
GOLD GLITCH the remainder of needed gold
save

--Segment 3--
sell death necklace for 1200
sell dragon scale for 10
buy stuff
etc.


If this speeds up being able to buy the half or full plate and/or the large shield along with the broad sword, then the initial leveling would be very fast and awesome.
Edit history:
Lhexa: 2014-06-05 08:18:42 am
After double-checking, Starwyverns and Werewolves do indeed go into the safe zone at level seventeen, so that was my mistake.  It's certainly odd that the average number of turns to kill a Starwyvern would go up, since there's no strength gain, and thus no different chance of the Starwyvern healing.

The Hauksness desert zone was the place where I noticed the negative "heal time", and it was certainly inappropriate there.  In a more general case you would have to make sure that the total "heal time" is positive, even if (as you justify) some of its entries are negative.

You're probably still overestimating the Charlock grind, due to splitting the Charlock trips up into fractional time costs per Healmore.  After all, if you spend 19 Healmores, you still have to take three whole trips.  Anyhow, at this point you could go ahead and measure the various times to frame-level precision, if you want to improve the spreadsheet.

Dunnius: In a segmented run the gold glitch is suboptimal, because you get Erdrick's equipment so early.  In a typical run the gold glitch pays for itself in gained time around level nine, but the segmented run already gets Erdrick's Armor at level eight.  And the easiest improvement to make to the segmented run is to get Erdrick's equipment at level seven instead, making the gold glitch even less effective in comparison.  The segmented run already does get the Death Necklace, though.
Edit history:
Crow!: 2014-06-06 08:06:37 am
Crow!: 2014-06-05 12:52:13 pm
What's that gemma?
I've improved the grinding spreadsheets. I now account for losing fights (they're treated as fights where you lose all HP; whether you get the EXP or not can be set for each zone since some areas would have to run from a bad fight while others could just spend a resource and go onward), and I now have a system for guessing the amount of time players' and enemies' turns will take.  I'm now underestimating EXP/min compared to Lhexa, which makes some sense given his 2 standard deviations policy, but it might also mean I'm overestimating how long certain parts of a fight take.

I studied a variety of grinding strategies, largely toward answering what the follow-up to an early Death Necklace grab would be.  It seems like every sensible zone at level 7 is 18-20 EXP/min, unless you also have both the large shield and the full plate (and of course the Broad Sword), in which case Wolves don't have to have SLEEP cast on them and so Rimuldar is 29 EXP/min.

----

If I'm looking at how to get a level 1 head start, I suppose I might as well consider the possibility of a level 1 Metal Slime as well.  But that doesn't really pan out...

To get to zone 11, you have to traverse ~2.5 battles' worth of Wyvern territory.  Running away is at ~9-35% odds depending on the enemy, and to survive these fights you have to not only run away on the first try, but also not get ambushed (which uses the same calculation).  So that's 5 times you have to succeed with dodge rolls, for a ~1/625 chance to get through to Zone 11.

Then you must find a Metal Slime with 20% odds (if you get a different fight, you get a ~2.25% chance of trying again at another 20%, which is negligible), then you have to win the fight at ~9% odds.

All in all, I'm estimating that, assuming you get through the Skeleton areas (which I'm less certain how to calculate), you get a Metal Slime kill every 40,000 attempts.  That doesn't sound reasonable.

-----

I tested most of a RTA run, with moderate tool assisting.  I outright luck-manipulated the Death Necklace chest, a level 12 Axe Knight fight, and a level 13 Erdrick's Sword dive.  While I didn't try to luck manipulate anything else, it is certainly true that the fact that I knew I had save states to fall back on if things did go wrong meant that I could play more dangerously than I otherwise could.

I did a level 2 Death Necklace as described a couple posts ago, with this follow-up:
- Rush to Garinham, fleeing from enemies.
- Sell Death Necklace, buy 6x Herb, 2x Torch, Hand Axe, Chain Vest.
- Level up to 4 in the Magidrakee hills south of Garinham.
- Sleep in the Inn in Garinham.
- Level up to 6 in the Mountain Cave's first floor, repetitively grabbing the Herb chest blind and fighting all enemies.
- Light a torch, plunder the Torch chest and the two Gold chests.
--- Run from Warlocks, fight everything else.
--- You have 2x torches so that, if your HP value warrants it, you can skip the Torch chest.
- Repeat until almost out of resources.
- Grind in the desert near Kol until level 8 + 1500G is achieved.
--- Pick up the Fairy Flute and Wings.
- Go to Rimuldar, buy Broad Sword when possible.
--- On the way to Rimuldar, avoid Wolves unless at full HP, and if at high HP use HURT vs Metal Scorpions.
- Grind at Rimuldar.
--- Until level 8, Wolves and Scorpions are both Sleep targets.  Stay away from Warlocks until level 8.
--- Once you have 1000 gold, buy Half Plate.
- Keep grinding here until 150 EXP short of level 12.
--- No, don't fight the Wolflords.
--- I'm skipping Full Plate, which saves 2850G but costs 3.78 minutes in grinding by failing to bring Wolflords into a profitable range, according to my spreadsheets anyway.
--- This time cost is 2x faster than if I had spent that much time in a zone with 5x Goldman fights.
- Buy keys, use Wings, and head to Garinham.
- Get the Cursed Belt, Silver Harp, and level 12.
--- Run from Wraith Knights, of course.
--- I skipped Large Shield.  This was a pain later on when Wyverns didn't get the defense break until I obtained the Silver Shield, so I think a real RTA would pick it up rather than risking several Wyvern fights in a row.
- Get Erdrick's Armor.
--- Fight relevant enemies along the way, of course.
- Grind to level 13 + 14800 gold in the northern desert.
- Get the Silver Shield and Erdrick's Token.
- RETURN and get the Staff of Rain.
- Get the Rainbow Drop.
- Head to Charlock and dive for the sword.
--- SLEEP + run is safer than just running vs Stonemen.
- Death-warp out.
--- Once you have the sword, you can actually farm here, just not sustainably or safely.  Go ahead and grab some EXP until your MP runs out.
- Head to the southern desert and grind to level 18 or 19 and go for the win.

In the end, I made it to the final grind about 20 minutes ahead of Feasel's PB.  It's hard to say how much of this was the route and how much of it was having nothing go wrong at the hard parts thanks to the power of save states.

Essentially, I had the problem that my game state looked like a sped-up version of the segmented run, rather than a sped-up version of the single segment run.  Working without Full Plate slowed things down (there's no reason to leave Rimuldar's moat until level 12), and setting up the gold glitch was too slow to justify using it to steer myself back into that direction.  My guess is there's probably 10 minutes to be saved by some variant of the death necklace strat, and it's not clear to me whether making it so that getting an attempt off the ground takes 5.5 extra hours of grinding for a level 2 death necklace would be worth it to a single segment runner.

-----

So, I changed gears and investigated what this would mean to a segmented run.

I did a level 1 Death Necklace, followed by the same strategies as in the spoiler above, up until a level 7 Broad Sword purchase.  I then used the wings to get back to Cantlin (EDIT: uh, Tantegal...), grabbed the Silver Harp and the Large Shield at Garinham, meanwhile finishing off level 8 via grinding along the way (primarily floor 1B, but enemies on 2B are fought while diving).  I arrived at the level 8 grinding point at 72 minutes, matching Lhexa's state at 77 minutes.  I visited the King to save after the Death Necklace, while Garinham and the Axe Knight are their own segments, for 1 fewer segment than Lhexa's run.

Also, since I'm virtually out of cash on hand when I exit Rimuldar, I could do the Axe Knight at level 7 without worrying about losing money to death, speeding up the Garinham dive / grind, and cutting another segment.  Given the strategies used, the primary difficulty for killing the Axe Knight would be the worse fleeing chances (and therefore worse ambush chances) to get a turn in the first place; the fight itself wouldn't really be any different.

So, I went ahead and did that, too.  This time I arrived at 65 minutes.  That's 12 minutes ahead of the current SDA run, in 2 fewer segments.  So that's cool.  I have attached to this post the movie file (playable in FCEUX) where I achieved this.  I could make a Twitch recording if people are interested.

-----

@ Dunnius: segmenting the run means doing the Axe Knight at level 8 (or 7, apparently), which happens around the 60 minute mark, at which point your armor gets obsoleted.  The gold glitch, once set up, makes 190 G/min, which means spending ~17.5 minutes opening chests to get the Full Plate + Broad Sword, on top of the ~13 minutes to dive to Rimuldar twice and the Mountain Cave once.  So, compared to doing what I outlined above, you have ~30 minutes to grind with Broad Sword + Full Plate that has to compete with ~54 minutes of grinding with Hand Axe + Chain Vest + Small Shield.  Accomplishing that is not completely out of the question, but it seems unlikely to me.
Intruding N313 and F014
Yeah, it is too bad that the gold glitch is slow.  I kind of figured that the full plate could be a bit much since it is expensive (the magic armor's healing would be nice, but the cost is out of the question).  What about broad sword + chain mail + large shield?  That would cut around 10 minutes of the gold glitch compared to the full plate, but I'm not sure if it could save more than 14 minutes of enemy grinding if it requires weaker enemies than the full plate.

I'm still thinking about this route for a single segment, which would be a pain no matter what due to the death necklace.  I suppose some initial leveling would have to be done so that the trip to Rimuldar would be more survivable to reduce the number of resets, but I don't know how much would be needed and whether it would be worth it.
Edit history:
Lhexa: 2014-06-08 06:44:23 pm
I'd be interested in a Twitch recording, Crow!  I'm honestly glad that you've found such a nice improvement to my route.  I never tested Mountain Cave grinding, since it didn't occur to me to manipulate its herb chest.  Also, the next improvements to make would be to get both Erdrick's Armor and Sword at level seven, so you wouldn't need to ensure enough gold for a Broad Sword at level seven.

A level one metal slime kill wouldn't be a mere toss of the dice, but a rather complicated RNG manipulation.  The RNG always seeds on the same value (for a given save file), and is called twice per step, so there are in effect two "lists" of encounters that you might go through.  You would chart out as much of these lists as you could, then stop walking briefly before known encounters -- the slowest possible pause is fairly easy to do on a controller and has a specific number of RNG calls.  The problem, of course, is that the lists become harder to chart, and it becomes harder to identify your spot on the list, the further along you are on it.  In my segments I could usually anticipate how the first minute or two of encounters would play out.

It would still be an incredibly difficult undertaking, mind you...
What's that gemma?
As requested, I've made a twitch recording of the proof-of-concept TAS.  Audio commentary is provided.
http://www.twitch.tv/iicrowii/c/4431543
I've written a bot in Lua that beats Dragon Warrior automatically.  This was discussed half-jokingly in feasel's stream chat a while ago.

Here's a video of the bot beating the game in 6:25:22 http://www.twitch.tv/bml2422/c/5184175
Here's the source code: https://github.com/dwbot/dwbot

Joking aside, this might be a cool way to quickly test out new route ideas.

Suggestions, bug reports and pull requests welcomed!
HELLO!
Heh a few of us were just discussing that bot with Error.  We were really questioning the code until Error noticed your per-monster scripts only apply to when the bot is grinding.  Having it only run from Starwyverns after running out of MP for Stopspell seemed weird until we noticed that. Smiley
It's pretty amazing that a bot could be coded to speedrun an RPG, even one as simple as DQ1. Really nifty.
Neat!  I'm reminded of the bot from Angband, which is a similarly bare-bones game.
Edit history:
bml: 2014-10-08 03:39:11 am
bml: 2014-10-07 09:50:14 pm
bml: 2014-10-07 12:05:01 am
I've been running some numbers with DWBot, and I'd like to share.

First is the success rate of getting Erdrick's armor at level 12 & level 13.  The script started in Haukness with full health and 6 herbs, and counted a success if it left Haukness with Erdrick's armor.  100 attempts were made at each level:
Code:
                         lv 12    lv 13
erdrick's armor         28/100   51/100


Then there's the question of how much time getting Erdrick's armor at level 12 saves.  The times in the following table measure how long it takes to grind from level 12 to level 13, plus the time it takes to grab the armor:
Code:
num trials=5                         time        
strategy                      min     avg     max
1. grind to 13 then armor   29:27   31:07   32:11
2. armor then grind to 13   23:39   25:03   26:30


Next is the success rate of getting through Charlock and beating the Dragonlord at level 18 & level 19.  We start at Charlock's entrance with full health, MP and herbs:
Code:
                         lv 18    lv 19
charlock + dragonlord   13/100   75/100


One thing that all the speedruns did, that the bot didn't yet do, is fight a few battles inside Haukness then leave during the grind to level 18.  I tried it out with the bot and here's what I found:
Code:
num trials=5                                 time        
strategy                              min     avg     max
1. stay out of haukness           1:38:46 1:39:49 1:40:37
2. dip into haukness at 17        1:40:35 1:42:51 1:45:22
3. dip into haukness at 15        1:42:35 1:44:15 1:46:02

In other words, it seems to be faster to grind to level 17 outside Haukness and never set foot inside until you're level 18.  My interpretation of this is that while the monsters inside Haukness yield more XP, they take longer to kill and that lowers your overall efficiency.

Finally, I ran a bunch of tests on the best way to fight Starwyverns.  This is relevant because they are the 40% encounter inside Haukness, which is where the grind to level 19 takes place.  We start in Haukness at level 18 with full HP and MP and grind until level 19.  Only 5 runs were made apiece:
Code:
num trials=5                         time                  num
strategy                      min     avg     max  starwyverns
1. melee to death           20:38   21:32   22:15          189
2. stopspell after heal     20:37   21:31   21:50          209
3. run after heal           20:42   22:13   23:54          211
4. always stopspell         20:45   22:56   23:58          159
5. always run               24:25   26:15   27:03          288


Caveats for this data: the bot does not play perfectly, and I didn't do nearly enough trials to get statistically significant numbers.  Nonetheless I think the data is interesting.
Ok so I know this is not the best place to ask this but I am on my first playthrough of this game. I am enjoying it but I cannot find erdrick's sword. I read a few spoilers from here and from a guide before I started but I dont know where the sword is exactly. I think I read it was around Cantalin but I'm not sure. So I was wondering if someone could give me a small hint of someone to talk to or an area to search. I am level 19 now which I know from this thread that many people take on the final boss at this level so I would like to find the sword soon. Thanks for any help
Learning to Stream
Quote from psirockin123:
Ok so I know this is not the best place to ask this but I am on my first playthrough of this game. I am enjoying it but I cannot find erdrick's sword. I read a few spoilers from here and from a guide before I started but I dont know where the sword is exactly. I think I read it was around Cantalin but I'm not sure. So I was wondering if someone could give me a small hint of someone to talk to or an area to search. I am level 19 now which I know from this thread that many people take on the final boss at this level so I would like to find the sword soon. Thanks for any help



It's in Charlock in yon depths of doom and despair. Seek ye yon pathway through the Dragonlord's madness and you too can arrive at the chest that shall garner thee fortune of fortunes.
Thanks. I was wrong then. I was looking there but didn't find anything. Nice how you made the text fit too.
Edit history:
bml: 2014-10-10 03:00:46 am
Because of the way the RNG works in this game, I was able to make the bot perform exactly the same run on both PRG 0 and PRG 1 versions of the game.  Here's the result:
Code:
PRG 0      5:45:51
PRG 1      5:48:02

This also shows that the two versions of the game are functionally identical, but I'm guessing that was already known.
Dapper as fuck.
Isn't the difference something text related?  Amazing to see you save just over 2 minutes based on that.
HELLO!
THY HITS -> THY HIT POINTS or something like that.
Dapper as fuck.
Ah ok.  Shame my memory sucks, because I'm pretty sure I had the PRG 0 one as a kid.  I seem to recall I only had it because of my NP subscription.
Sadly, the various speedrunners seem to only ever find the "thy hit points" version.  There was some debate earlier in this thread of whether the other version was ever widely distributed.
HELLO!
My understanding is that the THY HIT POINTS version was the one that went out for free with Nintendo Power, so it's just overwhelmed the other version.

At least it's easy to adjust for the difference
Dapper as fuck.
Oh, am I backwards?  I thought the NP one was the one that was sought after.  hahah
HELLO!
Quote from philosoraptor42:
Oh, am I backwards?  I thought the NP one was the one that was sought after.  hahah

Yeah it's the other way around.
Edit history:
Asvarduil: 2014-11-16 03:18:43 pm
I've done a Dragon Warrior I speedrun on the Game Boy Color version (the Dragon Warrior I & II compilation, released in the U.S. in 2001.)  I used the English translation.  I did an Any% Glitchless route, which means not taking advantage of the Chest Glitch for the Choker, and not taking advantage of the fact that the Dragon+ respawns if you die while holding Lady Lora (unless you die naturally, in which case it becomes a mitigation strat.)

You can read my route notes here
You can watch my current PB here

Most of the run actually isn't too bad, but it does get boring during the requisite grinding parts (Lv.9-14, Lv.16-19.)  Right now, my work involves figuring out ways to grind more efficiently, such that less time is spent grinding.  Unfortunately, you can't get rid of said grinding at all; at Lv.18 you have a 33% chance of winning the Dragonlord II fight; at Lv.19 that jumps to 95%.  Subsequent levelups only add decimal points to your chances of winning.

As far as endgame grinding is concerned, I actually like the Blue Dragons found on B3 and B4 of Charlock Castle.  They grant ~170XP each, and from Lv.16 to Lv.19, it takes 3000XP to gain another level.  Thus, every 20th Blue Dragon grants a level up.  What's more, they're rather common.  My first two runs, I've tried to grind Red Dragons on B5-B7 of Charlock, but despite their good XP yield (350XP each, best random encounter in the game), they're not quite as common as Blue Dragons.  Starwyverns just suck.