A study on blacksmithy (How To Train Your Blacksmith)

The researched blacksmithy training recommendation is:

1. 0..32~35: you'll want to train up the skill from NPCs as high as possible. Blacksmith guildmasters are the best at that, and they have quite a lot of variance between one another, so you might want to try several smith shops and pick the one that can train you the highest.
2. up to 49.6: daggers
   
3. up to 94.0: short spear
4. up to 100.0: plate gorget

These skill levels are displayed skill, which is the sum of your knowledge on the subject (real skill) combined with your physique, and represents your ability in crafting. In principle, the most economic means to train up smithing would involve locking your STR as low as possible, but for now it's unknown how big the cost savings would be practically.

The training cost for this path is (at least for low STR):

• NPC to 50.0: ~2.7k ingots
• 50 to 100.0: ~34k ingots

Update: This assumes that created items are smelted back into ingots. Smelting is actually a mining ability as CaptainMorgain noted, so mining skill will also affect the total ingots consumed.

What is this based on?

Before working on blacksmithy, I researched recommendations to mastering the skill. After all, it's a mighty expensive skill to train so it's worth putting some effort into the means. The large variance between recommendations got me wondering, how does one really measure the skill gain vs. used resources per item to come up with a recommended training program?

Quite a few hours of smithing (and endless ingots) later I had my hands on a chart that provides evidence on which items to make and at which skill level. And here it is!



...err, okay, that still needs some work. If you've ever felt like skill gain is erratic, that's because it is! At first you spend endless time, effort and resources in vain only to find that the next gain will be accompanied by several in short sequence even at higher skill levels. But I had the data, and after applying a low pass filter to look at trends within the high variety input, I this graph:




About interpreting this graph:

• I included 2 runs of the short spear and plate gorget for reference on the variance you might expect within a single item run
  
• the low pass filter does reduce the original erratic input to general trends, but the implementation details might leave mathematicians wanting to look into it a bit more; feel free to ask for details
  
• the beginning and the end of each graph is cut a bit short, because the end points are averaged out using several of the surrounding data points

The dates on the charts are old because that's when this research was done. However, it's not conclusive and I had all the intentions of resuming the research but Real Life has come ahead and occupied me since, so I wanted to release what I had so far and would be happy if anyone found this useful and even would be thrilled if people would like to refine the research from here.

So, if you do train up smithy with this method, please post your STR, starting smithy skill level, mining skill level and ingots used. Total ingots is useful input already, but separate ingots used for 50-70, 70-94 and 94-100 would be preferable if possible.

How to record the data for the charts:

1. Set up a remote house with no neighbors within a screen-or-so and with a forge & anvil. Place a locked down container next to the toon, fill with ingots and tongs, and set up the appropriate restock agents for both.
   
2. Build a macro to keep making the last item, and smelt the produced item type after each smithy attempt. After each skill gain, the macro would move all ingots into one stack in the restock container and use Item ID to read the stack size into the journal. It is very important to ensure that there is only one ingot stack anywhere nearby, as the macro targets the stack by type and the results will fail if it targets the wrong stack. This is also why point 1. mentioned a remote house with no neighbors nearby.
   
3. Within the UO directory, the Desktop/<account>/UO Renaissance/<toon>/uo.cfg file has to be modified to add a line "JournalSaveFile=journal-<MyToonName>.log" for the toon that will be doing the smithing. This is very important, as the data is taken from the produced journal file and if you don't make the file toon-specific, you can't play UO in parallel while the smithy macro is running or you risk getting bogus results!
   
4. Before starting a new item run, you log out, delete (or move) the old journal file to ensure you start up with a fresh journal, and log back in. This will also ensure your UO client knows not of any other ingot stacks.
   
5. After the item run, store the produced journal in a safe place and make sure the produced file name represents the item that was made
6. Throw all the item-specific journal files at a custom Perl program that produces the chart in the original post
7. If restocking is necessary in the middle of a run, stop the macro, restock, and say "restock X" (with X being the number of added ingots) before resuming the macro. The Perl program should interpret such "say" lines to track ingot usage through restocks properly. 10k ingots appearing in between crafted plate gloves does not suddenly make them the most economic option.

This was a whole lot of fun to work on, and the reason the charts have a timestamp is that I was eager to look at the data from work while my smith was macroing away at home. I had an URL that updated once a minute showing the live progress of the smith, drawing up a line gradually as his skill gain progressed. Fun times!

Still waiting for the correlation measurement for oak art type trees vs. oak harvesting.


Thanks Canis, cool work in the OP. However it reminds me way too much on my research work that I still have to do...

No testing? What kind of study is this?

I have a ton of extra ingots, and the standing best Blacksmith training method; I will run it once or twice

Maybe a theoretical study is more effective here? You know the exact difficulty of the items and you know the exact number of necessary ingots to create. That should in theory be enough.

Two questions though to answer beforehand:
- How is the impact of the difficulty on the gain rate. (One concept for that is, that the gain rate is equal to the fail chance... or at least linearily correlated to it)
- Is there a difference in gain rates for succesful smithy attempts and failures.

That was the basis of the BS in under 40k guide, keeping the difficulty within a certain range while using the best ingot item. I would support such a data driven study. Mine was loose, as it started only for my use

Also, keep in mind that not all guides are created equal. So, the data is inherently dirty if you’re using other people’s work (even my own).

I would gladly help test this, if someone would kindly supply the ingots

Yes, there is some variance which is why Canis published two runs of short spears and gorgets. You can see the trend there.

He did smooth out that line a fair bit more during tests which shows a much cleaner graph but it also meant losing a chunk of line at the start and finish of the run.

I can assure you it was a thorough examination of the gains

This is the sort of dedication to investigation that I respect and appreciate. I don't even have a smith but I love this.

I would, but they somehow ended up in BlackEye's possession along with all my other resourcers ;-)

FB

Testing was done, but it's incomplete.

An incomplete one, and I thought I made that clear in the OP. I've had the data for 1.5 years with no progress so I thought I'd drop what I have to the community.

There are several brilliant people who have put research into server mechanics and who I thought may have use for this. People such as yourself.

"This must have been when I was younger..."

I'm glad you asked!

Because in theory, practice and theory are the same. However, in practice they're not.

How would you be sure that your gain model is the same used by the server? I'd expect you need more testing to verify the theoretical gain model than you would to just test the relevant items. I'd be glad to be proven wrong, however.

I'm just giving you shit, man. It's cool what you've done.

I am running it right now It's definitely lean on the ingots for now, but the success % near the 49.6 and 95.1 marks is really high so we'll see how it is then.

Update 1: NPC trained to 50 took ~2k ingots (vs 6k previously)

Starting the short spears, 8.6% success chance

Thanks for doing this study and sharing your results with us Canis!

It's been a while since I looked at these, but from what I remember those limits would have been at or very near the limit of 100% success and hence no more gains. It certainly felt like the results were more in favor of low ingot use per item than targeting a certain difficulty level, as long as there are gains to be made. Sounds like your tests will soon either confirm of falsify this view