PoE2 Crafting Expected Cost: The Cheapest Move Skips Omens
Working PoE2 crafting expected cost backward from the finished item, the cheapest route barely used expensive Omens and removed misses with a plain Orb of Annulment. On one pair of gloves, a route close to intuition cost about 3,400 div (div = Divine Orb), while the best route cost 105 div, and 94 div over 1,000 engine runs. That's about one thirtieth.
Zundamon
When a miss rolls, the best play is an expensive Omen that removes only the miss's side, right?
Metan
That's what most people think. But the calculation picked a plain Orb of Annulment, at 0.44 div each.
Zundamon
It might hit a target, and it's still cheaper?!
Metan
Yes. It came to about one thirtieth of the intuitive route. Let's go through why.
What We Calculated
The item is a pair of Intelligence gloves: one Fractured prefix, plus +2 to Level of all Projectile Skills on the suffix side.
Zundamon
What gloves did you start from?
Metan
A Rare with one Fractured prefix and +2 to Level of all Projectile Skills as a suffix.
Zundamon
And what are you aiming for from there?
Metan
Maximum Energy Shield or maximum Mana at T1, plus the mods from the Essence of Hysteria and the Essence of Horror.
Zundamon
How do you even work out "average div to finish"?
Metan
You start from the finished item and step backward, one craft at a time.
Metan
For each state, you try every possible craft and keep the one with the lowest expected cost afterward. It's value iteration on a Markov decision process.
Zundamon
Sounds like there would be endless states.
Metan
Summarize a state as "which targets are present + how many misses on each side," and there are only a few hundred.
Metan
So nothing heavy like machine learning is needed. Starting over from a white base was also one of the allowed moves.
Result: About 1/30 of the Intuitive Route
Here is the expected cost to finish for each route.
| Route | Expected cost to finish | Source | Versus the cheapest route |
|---|---|---|---|
| Cheapest route (calculated) | 105 div | backward calculation | baseline |
| Same route, 1,000 runs on the engine | 94 div | engine | about the same as the calculation |
| Rule: "if +2 Projectile is lost, start over" | 775 div (engine 788 div) | calculation + engine | much higher |
| A route close to intuition (Chaos + Omen of Sinistral Erasure) | about 3,400 div | calculation only | an order of magnitude higher |
| Annul misses + Omen of Sinistral Annulment | about 3,900 div | calculation only | even higher |
The closer a route is to intuition, the more it costs, by an order of magnitude.
Zundamon
105 div versus about 3,400 div?! That's a different order of magnitude.
Metan
And the cheapest route came to 94 div over 1,000 engine runs.
Zundamon
But with bad luck, it costs more, right?
Metan
Half of the runs finished within 64 div, and one in ten took 223 div or more.
Inside the Cheapest Route
The moves the calculation picked, by situation:
| Situation | Move chosen | Why |
|---|---|---|
| A miss rolled | Plain Orb of Annulment (0.44 div) | The expected loss from hitting a target is smaller than a side-picking Omen (14 to 21 div each) |
| Aiming for maximum ES / maximum Mana T1 | Greater Exalted Orb + Omen of Sinistral Exaltation | Adds only to the target's side |
| Adding an Essence | Omen of Sinistral/Dextral Crystallisation (0.4 div) | Adds it while removing from the side with a miss |
| +2 Projectile was lost | Don't start over: put it back with Chaos Orbs | About 31 div on average. Starting over from white costs about 580 div |
Plain Annulment for misses, Chaos for a lost target. Both are cheap moves.
Zundamon
So you don't use Omen of Sinistral Erasure or the Omens of Sinistral/Dextral Annulment?
Metan
They weren't chosen in most situations. They cost 14 to 21 div each.
Zundamon
But a plain Annulment can hit a target.
Metan
It can. Still, it's only 0.44 div.
Metan
If "chance of hitting a target × cost to put it back" is less than the Omen's price, the plain Annulment wins.
Zundamon
Surely you start over if +2 Projectile is lost?
Metan
You put it back with Chaos Orbs. Each one lands it 0.32% of the time (measured on the engine).
Zundamon
0.32%? That almost never lands...
Metan
Even so, it averages about 31 div. Starting over from white costs about 580 div, so it's no contest.
Metan
On these gloves, putting it back is cheap, so hitting it doesn't hurt much. That's why the plain Annulment wins.
Where Omens Pay Off
So when should you use the expensive Omens? Splitting another pair of gloves into 54 situations, the rule was simple.
| Where the miss and targets sit | Move to pick | Why |
|---|---|---|
| Targets only on the same side as the miss | Plain Annulment | Picking a side protects no target, so you just lose the Omen's price |
| A target sits on the side with no miss | A move that picks the side (Omen of Sinistral Annulment, Chaos + Omen of Sinistral Erasure, a Bone + Omen of Dextral Necromancy, and so on) | It protects the side with the target |
Picking a side only matters when a target sits on the side with no miss.
Zundamon
Can the break-even point be written as a formula?
Metan
It can. It's in the box below.
On the same 54-situation gloves, we also measured what happens when moves are banned.
| Restriction | Versus the cheapest route | What happens |
|---|---|---|
| Ban side-picking Omens and Erasure (all targets T1) | about 36 times (about 190,000 div) | The target's side can't be protected, so losses from hitting targets pile up |
| Ban side-picking Omens and Erasure (targets T3 or better) | about 11 times | Even with looser targets, the loss is large |
| Ban plain Annulment | +17 to 18% | You end up using expensive Omens even where a cheap move works |
| Starting over (69 div per base) | never chosen in any situation (always at least 67 div worse) | Continuing from what you have is cheaper |
Always using Omens and never using them both lose.
Zundamon
Banning Omens costs about 36 times more?!
Metan
That's when all targets are T1. About 190,000 div.
Zundamon
Then just use Omens every time.
Metan
That loses too. Banning plain Annulment costs 17 to 18% more. Switching by situation is cheapest.
Zundamon
When misses pile up, I just want to throw it away and start over.
Metan
The calculation never chose starting over. It was always at least 67 div worse.
Metan
The Gold Ring in Don't Hoard Your Misses follows the same logic.
Metan
Annulling while holding misses gives more chances to hit a target.
Not yet verified (3)
- "About 3,400 div" and "about 3,900 div" are calculation only, not checked on the engine.
- The break-even formula matching all 54 situations is a comparison inside the calculation, not checked against in-game results.
- The price of buying the base is not included. Nothing here has been compared with Craft of Exile.
Remove misses with a plain Annulment, and use a side-picking Omen only when a target sits on the side with no miss. That is the cheapest route the calculation found.
Where the Numbers Come From
Show sources (3)
- The calculation works backward using the engine's weights and tiers and market prices from 2026-10-21 (1 div = 718 Exalted Orbs). It uses approximations (a miss's mod group doesn't shrink the pool, only T1 is considered, and so on), and still matched engine runs within 10%.
- "About 3,400 div" and "about 3,900 div" are calculation only, not checked on the engine.
- The price of buying the base is not included. Nothing here has been compared with Craft of Exile.
Try It Yourself
You can apply the moves one by one on the crafting bench.
Related posts
VOICEVOX: Zundamon / Shikoku Metan
Character art: Shikoku Metan / Zundamon standing art by Sakamoto Ahiru