PoE2 Crafting Calculator Mistakes: 3 Bugs We Caught
When you work out PoE2 crafting costs by calculation, a small slip in how the calculation is built turns straight into a wrong number. In our ExileDesk test calculation, a biased random generator gave 68 div (div = Divine Orb), using the Exalted Orb's minimum modifier levels for Transmutation and Augmentation doubled the magic-item cost, and the way we tracked a desecration marker put us 24% off the engine.
Zundamon
Once you build the formula, an expected-cost calculation gives the right number, right?
Metan
We thought so too. But every time we replayed it on the engine, the numbers were off.
Zundamon
So the problem was in the calculation!?
Metan
Yes. All three were mistakes you'd never notice from the answer alone.
Mistake 1: A Biased Random Generator Gave 68 div
On 2026-10-20 and 21 we tested working out, backward, the expected cost to finish a pair of Intelligence gloves (suffix: +2 to Level of all Projectile Skills). The first thing that tripped us up was the random generator.
Zundamon
Why does an expected-cost calculation need random numbers?
Metan
Evaluated as is, rare paths that go round in circles inflated the values.
Metan
So inside the calculation we follow crafts one run at a time, with a cap on how often the same move can repeat.
Zundamon
And those runs use random numbers.
Metan
Right. With our first, simple generator (a linear congruential generator), the top order came out at 68 div.
| Counted with | Cost of the top order | What happened |
|---|---|---|
| Calculation (simple generator) | 68 div | Biased numbers, too high |
| Calculation (mulberry32) | 61.9 div | Dropped just by swapping the generator |
| Engine (1,000 runs) | 58.7 div | The played-out value |
Zundamon
Same calculation, and the generator alone moved the number!
Metan
Our calculation rules now say: use mulberry32, because linear congruential generators are biased.
Mistake 2: Transmutation and Augmentation Floors Aren't the Exalted Orb's
Starting from a white base, the first half makes a magic item with Orbs of Transmutation and Augmentation. Their Greater and Perfect versions have a minimum modifier level for the mods they add.
Zundamon
Isn't the floor the same for every orb?
Metan
Our first calculation used the Exalted Orb's Greater 35 and Perfect 50. But the real values are different.
| Orb | Greater | Perfect | What's different |
|---|---|---|---|
| Exalted Orb | 35 | 50 | Our first calculation used these for Transmutation and Augmentation too |
| Orb of Transmutation / Augmentation | 44 | 70 | The real floors, higher than the Exalted Orb's |
Zundamon
What goes wrong if the floor is too low?
Metan
Low tiers of the target family count as possible, so hits look rarer in the calculation.
Metan
So we estimated the magic-item cost at double. The earlier "about 580 div per restart" is too high by that much.
Zundamon
How did you fix it?
Metan
The engine keeps the floors in CURRENCY_FLOOR, with separate values for Transmutation and Augmentation. We matched the calculation to them.
Mistake 3: A Desecration Marker Put Us 24% Off
The test kept "is there a mod added by Desecration" as a marker in each state. Only one marker was allowed, and with a marker present, bones couldn't be used.
Zundamon
What was wrong with the marker?
Metan
After a desecrated target was removed with an Orb of Annulment, we kept the marker.
Metan
Then bones counted as unusable, and it chose a path of Exalted Orbs and Annulments over and over.
| Times +2 Projectile was removed (per run) | Times | What happened |
|---|---|---|
| Calculation | 1.02 | The leftover marker caused repeated Annulments |
| Engine | 0.45 | The played-out value |
Zundamon
The calculation removed the target more than twice as often!
Metan
So the marker now also records which target came from Desecration, and disappears when that target is removed.
For one of the target orders, here is the calculation next to the engine.
| Cost per item, one order | Calculation | Engine | Gap | Result |
|---|---|---|---|---|
| Before the fix | 109 div | 83 div | 24% | Calculation far too high |
| After the fix | 70.5 div | 65.1 div | 6-8% | Gap shrank |
Zundamon
The costs themselves are very different before and after.
Metan
Runes are handled differently. The "after" figures exclude 10.8 div of runes, so compare the gap, not the totals.
Metan
Even after the fix, a 6-8% gap remains, always with the engine cheaper.
The same kind of mistake, once more
Handing the calculated answer to the engine tripped on the marker too.
Zundamon
It's just handing it over. What happened?
Metan
The engine can't see which target came from Desecration. So we told it "if Speed is there, the desecrated mod is Speed."
Metan
It then kept removing a Speed mod that had come from a Chaos Orb, looping 280 times per run.
Zundamon
How did you fix that?
Metan
We now pass the marker answer that the calculation's own runs went through most often.
How We Caught Them: Calculation Next to Engine
On their own, all three gave plausible numbers. We caught them because we replayed the calculation's best moves on the engine, 1,000 runs at a time, and lined the numbers up.
Zundamon
Did playing it on the engine show anything else?
Metan
Yes, more mistakes on the calculation side. Omens that pick a side, such as Omen of Sinistral / Dextral Crystallisation, can't be used unless that side has a removable mod.
Metan
A Perfect Essence also needs a removable mod. Only a Rare with zero mods takes it without losing one.
Zundamon
Do you do anything to stop endless loops?
Metan
At the start of the test, the owner told us automated expected-cost calculations bug out easily and should have caps.
Metan
So re-adding a mod is capped at 3,000 times, and everything else at 300.
Zundamon
So now you never stop at the calculation.
Metan
Right. Rule number one of our calculation work: show every calculated number next to the engine's result, with the gap in %.
Not yet verified (4)
- This is about a calculation experiment that decides crafting moves automatically. Deciding moves automatically is not a feature of the site (there are too many branches; the owner decided against it on 2026-10-21).
- We don't yet know what causes the 6-8% gap left after fixing Mistake 3.
- The "double" in Mistake 2 is the recorded estimate. We haven't recomputed the restart cost after the fix.
- The subject was one pair of gloves. We haven't checked whether other item types show gaps of the same size.
Expected-cost calculations drift from small details like the random generator, the floors and a marker. Lining them up with played-out numbers is the quickest way to catch the mistakes.
Where the Numbers Come From
Show sources (6)
- 68 and 61.9 div (random generator), 109 and 70.5 div and 1.02 removals (desecration marker): values from the ExileDesk calculation experiment, 2026-10-20 to 21 (calculation).
- 58.7 div (random generator), 83 and 65.1 div and 0.45 removals (desecration marker), the 280-time loop: 1,000 engine runs of the calculation's answers (engine). The 24% and 6-8% figures are the gaps between the two.
- Prices from the 2026-10-21 market (1 div = 718 Exalted Orbs).
- Transmutation / Augmentation floors (Greater 44, Perfect 70) and Exalted Orb floors (35, 50): the engine's
CURRENCY_FLOOR(engine). - Magic-item cost doubled, restart about 580 div: the experiment's notes (estimate; not recomputed after the fix).
- The mulberry32 rule and the "line it up with the engine" rule: ExileDesk's calculation rules (docs/org/math/README.md).
Try It Yourself
Play a craft one step at a time on the crafting bench, or compare Chaos, Augmentation and Exalted spam in the spam simulator, opened from the button by the Currency heading on the crafting bench.
Related posts
- PoE2 crafting expected cost: the cheapest move skips Omens
- PoE2 crafting cost variance: average, median and unlucky 10%
- PoE2 Greater vs Perfect orb difference: min mod levels
VOICEVOX: Zundamon / Shikoku Metan
Character art: Shikoku Metan / Zundamon standing art by Sakamoto Ahiru