Master Duel crafting is rigged*
The N/R Numbers#
If you click on the "odds" button displayed next to a pack, and scroll down to the bottom, you can find this table of glossy/royal chances:
Tentatively, we may assume these to be accurate for cards pulled from packs. There would likely be legal trouble otherwise, and I don't particularly care to crawl through masses of pack opening videos. But what about crafting? As far as I can tell, rates there are not disclosed anywhere in-game.
We might assume this to be the same as the rate from packs - that would be reasonable enough. But why assume when you can test? While SR/URs are too expensive to spam craft, I've been collecting my own data recently when trying to get glossy playsets of some random N/R cards. It looks something like this:
| Card | Rarity | Start count | End count | Glossy rate |
|---|---|---|---|---|
| Lifeless Leaffish | R | 0/0/- | 33/3/- | 3/33 |
| Shif, Fairy of the Ghoti | R | 3/0/- | 97/2/- | 2/94 |
| Gold Pride - Roller Baller | N | 2/0/- | 146/3/- | 3/144 |
... and so on.
In total, 1196 craft attempts resulted in 53 glossy N/R cards - a 4.4% rate. The true glossy rate is almost surely (with 99% confidence) at least 3.1% but no more than 6.3%. Expect ~20 crafts per glossy, or ~60 for a playset - ~1200 N/R dust after refunds.
The SR/UR Numbers#
It takes an incredible amount of data to get a solid read on royal rates (as they're expected to be in the 1% range, I'd feel more comfortable with 2-5000 samples), but glossy SR/URs are a bit more doable. From my data so far, 968 SR/UR crafts resulted in 49 glossy cards (5.1%), and 9 royals (0.9%). We can't say much about the royal rate yet - it may be anywhere around 0.4-2.2%. But this is enough to conclude the SR/UR glossy rate does not match the stated 9% odds from packs (p<0.0001). With 99% confidence, the true SR/UR glossy rate is between 3.5% and 7.3%.
If I had to take a wild guess, I'd say the overall probabilities currently look like:
Be careful when spending your dust, and be even more careful when deciding whether or not to spend your dollars.
Notes#
Some people, when it comes to probability, like to come up with increasingly bizarre justifications on what influences the result. Maybe you get better odds if you craft exactly on the hour, with a left-hand mouse. Maybe you get worse odds if you have at least 3 royal rare cards in the archetype. Without evidence these are equally worthless, but I will head off one variation - the batch size does not appear to matter. It's a somewhat smaller sample (and more annoying to collect), but my experimental glossy rate for N/R cards is about the same whether 3 are crafted at once or only 1 (10/212 glossy with batch size 1, reject a 10% glossy rate with p<0.02).
I have also started collecting data on UR/SR crafts, but that will take much longer to get a respectable sample. If you would like to contribute, feel free to message me. I'll periodically update this page as my sample size improves. Most critically, any submitted data must be decided as "data to submit" before the crafting takes place.