The Class Nobody Had Solved Yet

A build guide can give you the answer. The more useful skill is learning how to find the answer when no guide exists.

A Diablo IV class launched without a guide. The players who solved it discovered something more valuable than the answer: how to learn when nobody has written the instructions yet.

Lord of Hatred released on April 28, 2026, with two new classes in it. I rolled a Warlock that night.
The pieces were public. Blizzard had previewed the class ahead of launch — two resources instead of one, Wrath for skills and Dominance for the demons, the skill clusters, the hex, the skills by name. You could read all of it in advance, and I had.
What didn’t exist yet was the sentence you make out of those nouns. Which cluster to commit to. Which tag to stack. Which of the interactions the game shows you are worth building around, and which ones are decoration.
I took Doom at level one because it was an Abyss skill and applied a hex, and the hex increased Abyss damage, and there was a Core skill two levels up carrying the same tag. Dread Claws at three. Then I read upgrade text for an hour, which is most of what building actually is.
Blizzard was patching the game while I was still learning it. The rules had already started moving under me.
I made four or five Warlocks that season. The first one was mine. I ran Cascading Dread instead of the circular upgrade because I wanted damage stacking on a single target more than I wanted the screen cleared, and I built the whole thing around standing still, which turned out to be wrong.
Within a week, nearly every Warlock guide I saw had converged on Dread Claws. There was a slope in the class, and once somebody had identified the strong interaction, the answer was obvious enough to copy.
But somebody has to find the slope first.

II. Somebody Measured It

By the second week the guides existed. They always do.
Here is the kind of thing in them.
Attack speed has breakpoints. You can add attack speed, watch the number on your character sheet go up, and find that the attack doesn’t execute any faster than it did before. The stat rose. The behavior didn’t. Then you cross a threshold and it jumps.
None of that is in the game. The character sheet shows a number rising smoothly. The behavior underneath it is a staircase.
Somebody sat down and measured all of it.
That’s the part the guide doesn’t contain. Somebody changed one variable and left the others alone. Recorded what happened. Ran it again, because one sample isn’t enough. Compared what they observed against what the tooltip claimed and found the two describing different things. Found where the steps in the staircase are. Wrote it down. Then a patch landed, some of it stopped being true, and they did it again.
Players like MacroBioBoi do this in public. Patch notes drop, and he opens them and does the arithmetic on camera, in the gap between the notes going up and the patch going onto the servers.
The guide is the answer sheet. The expertise happened before the answer sheet existed.
At work, there was no Maxroll — no site where somebody had already tested the system and published the build.

III. The Other File

Before this, I worked in a call center for six years. Agent, then floor support, then team leader.
The company published everything. Average handle time. Hold time. After-call work. Transfer rate. Schedule adherence, and the variance on schedule adherence. Sales. Sales opportunities. Close rate. Survey scores, and the response rate on the surveys. Refreshed every morning, every agent, every number, going back as far as you cared to scroll.
Forty-one columns. Unweighted. Just there.
Everyone watched customer service, sales, and handle time. Those were the three numbers everyone watched. Nobody could have told you their transfer rate. I couldn’t have told you mine.
The weights were in a different file. I eventually found them in a compensation document in the shared drive — the scorecard categories with a percentage next to each one.
Transfer rate counted twice as much as handle time.
I built something in Excel that pulled the daily export and threw away everything not on that list. It was not sophisticated. It was a filter. What it did was reduce forty-one columns to six, in the order they were worth.
Then I gave it to my team, and my team started transferring differently, and within two months the leaders on either side of me were asking where the numbers came from.
It took about a week. I thought I had built a dashboard. What I had built was a model of what the system rewarded.

IV. What the Sheet Doesn’t Show

Transferring a call lowers your handle time. The call ends on your line and continues on somebody else’s.
So the number everyone was protecting was being protected by the behavior that damaged the number worth twice as much. Not a subtle interaction. Arithmetic. It was invisible because the two figures lived in separate columns and neither one carried a weight.
The floor was not wrong to watch handle time. For agents, handle time could disqualify a payout outright, and a disqualified payout is worth more attention than a percentage. They learned that at agent level and it was correct at agent level. Then some of them got promoted, the weights underneath them changed, and nobody sent a note.
Sales opportunities were worth nothing on the scorecard. No points for logging one. But you couldn’t record a sale without logging an opportunity first, and sales carried real weight — so the number worth zero determined how much of the number worth something was available to you at all. There was no place on any dashboard for a thing like that to appear, because dashboards display contribution.
And handle time was not a target. It was a window. Too long disqualified you and too short disqualified you, and in the months it appeared on the payout sheet — which was some months and not others — people lost multipliers on both ends.
A short handle time meant easy calls. Cancellations, balance checks, four minutes and gone, nothing there to expand. A long one meant real troubleshooting. You don’t choose your calls. The queue does, and call mix moves the number in ways an agent can’t fully control.
In Diablo, a single hit lands somewhere inside a range, and the ranges compound. One hit tells you almost nothing about your build. You have to watch a great many hits before the number means anything.
The payout was monthly.
Those are five different ways a sheet of accurate numbers can mislead you. Weight. Interaction. Gates. Thresholds. Noise. Which figures actually count. Which ones move each other. Which contribute nothing directly and permit everything. Which improve continuously but only pay at a boundary. Which are mostly variance wearing the costume of a result.
None of them are printed anywhere. Finding them is what somebody has to do before a guide can exist.
Once somebody else has done it, of course, you can skip nearly all of this.

V. Blood Wave

I played a Necromancer in Season 6. Blood Wave looked slow and I did not want it.
I opened a guide, took the skill allocation in the order it was given, farmed the two pieces it said to farm, and by the second week the thing was clearing everything I put in front of it. One of the best characters I have ever had. I made two more builds on it that season because I liked the class that much.
I never learned the class. I got the outcome and skipped the part where you find out why it works.
I got the build. I didn’t get the skill that produced the build.
This is not an argument against guides. A guide optimizes for performance and it is very good at it. Discovery optimizes for understanding, and it is slow, and it produces worse characters. Most of the time you want the first thing. The mistake is assuming it hands you the second.

VI. Two Examples

I didn’t have language for why those two kinds of learning felt so different until I came across an old study on analogical transfer.
Gick and Holyoak, 1983, Schema Induction and Analogical Transfer. They gave people one story containing a solution and then tried to help them carry that solution to a different problem. They summarized the story. They stated the underlying principle out loud. They drew it as a diagram. None of it reliably got people there.
Then they gave people two analogous examples and asked them to describe what the two had in common. People produced the underlying structure themselves, as a byproduct of the comparison — and the quality of the description they produced predicted how well they solved the later problem.
The researchers were careful about what that does and doesn’t mean. It isn’t that a principle can’t be taught by being said. It’s that being told the rule and being able to see the rule are not always the same event. Sometimes the abstraction becomes visible only once you have a second case to hold it against.
Diablo 2, Diablo 3, Diablo 4, and a call center in North Carolina.
For years I thought I was learning six unrelated things. Three damage systems, a scorecard, Excel, and how to run a floor. What I was doing, over and over, was one procedure: walk into a system full of visible numbers, work out which ones matter, find the relationships that aren’t displayed, build something small that shows only those, test it against what the system actually pays, and rebuild it when the rules move.
I couldn’t see it while either system was the only thing in front of me.

VII. Account Bound

The first Warlock wasn’t my best Warlock. It may have been the most useful one.
It was built around standing still, which was wrong, and I found that out the way you find out anything in that game, repeatedly and at speed. The Blood Wave Necromancer was a far better character and taught me almost nothing.
In Diablo, some things belong to the character and some things belong to the account. The season ends and the character stops being the problem you’re solving. The build can stop being optimal at the next patch. The specific numbers keep changing because changing them is part of how the game stays alive.
The call center job ended. The spreadsheet is irrelevant. I couldn’t tell you today what the weights were.
What survived wasn’t a lesson. It was a reflex. When I’m handed a system with numbers in it now, I assume the sheet is incomplete, and I go looking for the file with the weights in it.
So when you borrow an answer, ask what somebody had to figure out in order to produce it.
What did they measure? What turned out not to matter? What relationship isn’t visible in the final recommendation?
You’ll still follow the guide. Follow it. Just don’t confuse having the build with knowing how builds get found.
That part is account bound.