Engraving enhancement: what to enhance first, when to stop

✍️ Alice・愛麗沙

The order comes from a coefficient, not a feeling

“Shards and gold are finite — which piece first?” is the single most repeated sub-question inside combat-power threads. The answer is sitting in the displayed combat-power formula that Korean players reverse-engineered: attack counts 1 per point, defence 0.5, HP 0.25. Our own gap model records those three weights at medium, medium and high confidence respectively, and — the part that matters — both independent sources we hold agree on all three. (The same model records a genuine conflict on the sub-stat weight, 0.3 versus 0.6. That one we do not use here, and you should distrust anyone who quotes it as settled.)

Weapon-rune engraving enhancement adds attack. Armour-rune enhancement adds defence. So one stage on the weapon is worth, on the displayed number, two stages of armour. Stack a second effect on top: every enhancement level also raises the rune’s own equipped bonus by 5%, and that part is slot-blind — it treats every piece identically. Add the two together and the priority sorts itself: weapon > accessories > armour and emblem.

Put plainly: one weapon stage equals two armour stages. When money is short, that exchange rate is the entire answer and everything else is a footnote. If you want to see which of your own slots is furthest behind, the gap diagnostic flags it per slot; the wider upgrade-order question — gold line versus trace line versus ticket line — is the combat-power priority page’s job. This article only drills into the ordering inside the enhancement line.

The official ladder: +11 is a cliff, pity is an instalment plan

Draw the whole ladder and you do not need to read it to see where it breaks:

Taiwan’s published figures in full, starting at +5 — the first stage where failure is possible:

+1 to +4 always succeed (omitted). When banked pity reaches 100%, the next attempt succeeds outright and consumes no materials; overflow carries to the next stage. Probabilities are NEXON NOW official Taiwan disclosures; tables update site-wide when the dataset does.Odds DB

Three things to take from that table.

+1 to +10 is the warm-up. The first four stages cannot fail. From +5 to +10, every failure banks 15% to 39.2% of a pity bar, and when that bar hits 100% the next attempt succeeds outright and costs no materials — which caps any single stage in this band at a handful of failures before a free success is guaranteed (the table’s last column does the division for you). There is no decision to make here. Take all six slots to +10.

+11 is the cliff, and it is a double cliff. Success drops 30% → 15%. The pity award drops 15% → 7.5% in the same step. Two halvings at once: the expected number of attempts doubles and the worst-case number of failures before the guarantee doubles with it — from at most 7 at +10 to at most 14 at +11, then 20 at +12, 29 at +13, 40 at +14, 72 at +15, and 200 at every stage from +16 up. That last number is not a typo, and it is the real reason “just keep going” stops being a strategy. Your weapon-first policy is what gets cashed here: the weapon crosses the cliff alone while everything else waits at +10.

Failure never destroys or downgrades anything. The gear and the rune both survive (Korean-community consolidation; no Taiwan counterexample recorded). The one live hazard is the curse — community reports say a failed low-stage enhancement can inflict it, but the sources disagree on whether that starts at +4 or +5, and neither is official, so we keep both readings. A cursed piece is cleared by a curse-removal potion, a blacksmith, or simply by your next successful enhancement.

Chart the pity column on its own and the same inflection appears a second time — the stage where your odds are halved is the stage where your consolation prize is halved too:

That is what “+11 costs more than one step” means in concrete terms. It is not one step. It is a different game.

The catalyst is never a trap — and the off-by-one that says it is

From +10 you can burn a re-burn catalyst. Taiwan does not publish it as “success rate up” — it publishes a joint distribution over two rolls. Working that distribution out is easy. Lining it up against the right row of the success ladder is the part that trips people, and getting that wrong flips the verdict completely.

Take the square root of the both-succeed cell and you recover the per-roll rate: 15%, 10%, 7%, 5%, 3%, 1% for the six published rows. Those are not arbitrary numbers — each one is a rate that already appears on the success ladder. The question is which row.

The two tables index differently, and the dataset says so. The success ladder’s note is explicit: stage N is the attempt that takes you TO +N. The catalyst table’s rows count the other way — stage N is the attempt you make from +N. The cross-check takes one line: the catalyst’s +15 row has a both-succeed cell of 0.01%, which is 1% squared, and 1% is the ladder’s +16 entry. All six rows shift by exactly one in the same direction.

Align them and the catalyst’s two rolls turn out to be at precisely the native rate of the very attempt you are making — not a discounted rate. Two independent rolls at p succeed with probability 2p − p², which is greater than p for every p below 1. So:

• +10 → +11 — 27.75% with the catalyst against 15% bare. About 1.85×.
• +11 → +12 — 19.00% against 10%. About 1.90×.
• +12 → +13 — 13.51% against 7%. +13 → +14 — 9.75% against 5%.
• +14 → +15 — 5.91% against 3%. +15 → +16 — 1.99% against 1%. About 1.99×.

There is no band where the catalyst is a losing trade. It nearly doubles the per-attempt chance everywhere, and the multiplier creeps closer to a clean 2× as the base rate falls, because the both-succeed overlap you “waste” shrinks with p.

If you have seen the other verdict — that 27.75% loses to 30% at +10 — the 30% in that comparison is the rate of the previous attempt, the one that took you from +9 to +10. It is a real number on a real official table, just not the number that attempt is racing against. This site’s own Chinese edition of this article made exactly that comparison and reached the opposite conclusion until 2026-09-03; the alignment above is why this edition never followed it, and is what the Chinese edition now says too.

One honest limit on the size of the win: the disclosure covers probabilities only, and the catalyst’s own acquisition cost is not published anywhere. The multipliers above are per-attempt, not per-gold. If you have a reason to ration catalysts, it is a supply question, not a probability one.

Two caveats that keep this honest. The published joint table stops at +15, so above that we have no disclosed catalyst distribution at all and this analysis does not extend there. And a catalyst attempt where both rolls succeed still advances only one level — but it banks a much larger pity award, which gives the item a legitimate second job: even a failed push leaves you measurably closer to the free guaranteed success. If you are deliberately farming pity rather than chasing the level, that changes the calculus.

Where the shards come from — and why gold is the real constraint

Engraving enhancement eats two things: gold and rune shards. Shards arrive, ranked by reliability, from dismantling surplus runes (the bulk of it, and the logistical basis for the whole “break them down and feed enhancement” consensus), from the daily-capped routine drops in the black rifts and the abyss, and from events.

Because the shard supply is a daily-attempt system, its accumulation rate is essentially fixed. That inverts the usual strategy question: shard stock is not the variable you should be optimising — gold is. Clear your daily runs and shards will carry you to +10 without thought. What needs real budgeting is the gold behind every stage past the cliff.

And here the honest answer is that we cannot give you one. Taiwan’s disclosure covers probabilities only — the official page itself lists gold and shard cost as not published, because the legal requirement is rates, not costs. Every cost table circulating for this game, including the estimator on our own enhance-odds page, is community-sourced. Ours was once flagged for its provenance; on 2026-10-02 we checked it against the Taiwan client, and the +15 row matches two Taiwanese players’ enhancement screens to the coin while +1 to +10 are exactly twice a Taiwanese player’s measured armour costs — so it is a real ★6 legendary weapon cost table (armour costs about half). The other stages from +11 have not been checked row by row yet, so treat a high-stage gold-per-attempt number — ours included — as an order of magnitude until they are, and trust the number on your own enhancement screen over any table. The success rates and pity awards on this page are a different matter entirely: those are archived from the official pull and verified line by line.

Your stop-loss line: three budgets, three answers

• Free-to-play. The board-reported ceiling configuration is weapon +15 with all armour at +9, at roughly 29,000 combat power (player self-report) — but that is a long grind, not a plan. The executable target is weapon +13, everything else +10. Past +13, each stage’s gold requirement climbs into a bracket where the honest question stops being “will it succeed” and becomes “would this gold move me further on a different line?”
• Light spender. Once the weapon clears +15, take the accessories next rather than the armour — accessory-rune sub-stats feed the damage-increase bucket directly, so their real-combat return outruns their contribution to the displayed number.
• Monthly pass and above. While pushing the weapon toward +17, check that your off-hand proficiency link has kept pace. That slot frequently returns more than another forced stage on the main weapon.

One principle covers all three, and it deserves its own line:

Move on the side where shards are not the bottleneck. Shards come in steadily from dismantling and from daily rift and abyss runs; gold is the genuinely scarce resource. So “should I enhance this” is always, underneath, “does this gold have a better home right now” — and the combat-power source ranking is where you check the answer.

A last external reference point for when to stop: you are not enhancing for a prettier number, you are enhancing to clear a content threshold. How far your current wall actually is, and whether one more weapon stage closes it, is what the gap tool computes. Once the threshold is cleared, this line yields priority to the trace and ticket lines. Do not spend a season proving yourself against a 1% stage.

So is “enhance everything evenly” right or not?

The board line — “dismantle spares, feed the shards into even enhancement, it beats fusion for combat power” — comes from players who actually ran the experiment, and we agree with its first half: shards into enhancement beats gold into fusion. It is the word evenly that needs an asterisk. It holds below +10 and nowhere above, because in that band every stage is cheap and the equipped bonus is slot-blind. Past +10, treating a stage that costs four times as much as a warm-up stage is not egalitarianism, it is a category error.

The correct mental model is two phases: take everything to +10 (even), then send the weapon alone over the cliff (concentrated), and come back for the accessories when you can afford to. Even first, then concentrated — the order of those two words is the whole article.

What we do not know

Three gaps, stated plainly. Our cost table is only checked at +1 to +10 and +15, so we cannot tell you the gold-optimal stopping point in gold terms — only in stages. The curse trigger stage is unresolved (+4 versus +5, two conflicting community relays, no official statement), and the size of the curse penalty is undisclosed entirely. And the catalyst distribution is published only to the +15 departure row, so the analysis stops where Nexon’s table does; pushes above +16 are unmodelled.

Our verdict

Three lines: take every slot to +10, send the weapon alone over the +11 cliff and up to +13, and use a catalyst on every attempt past the cliff that you can supply one for. That is one turning point more than the board’s “enhance evenly” and roughly half the gold of “dump everything into the weapon” — and every turning point matches a number Nexon published, not a feeling.

A claim you can overturn: we assert that even-to-+10 then weapon-only-past-the-cliff is the highest-expected-value gold allocation in Taiwan S0. If anyone computes — from the same official table — or measures a different allocation that yields more combat power for the same gold, concentrating earlier or later, bring the working and we will recompute and rewrite this section. The catalyst finding is even easier to falsify, and it rests entirely on one claim: that the two tables index stages one apart. Show that they do not — an in-client screenshot of a catalyst attempt at a known stage would settle it — and this section is wrong from its first line.