Genshin Impact Artifact Substats Guide: Priority Ranking, Roll Values & Keep-or-Feed

Substats are the 4 lines under the main stat on every piece — this is where RNG truly decides if a piece is "good," since main stats only have a handful of fixed options per slot. Genshin has exactly 10 substat types: 3 flat stats (Flat HP, Flat ATK, Flat DEF), 3 percentage stats (HP%, ATK%, DEF%), plus Elemental Mastery (EM), Energy Recharge% (ER), Crit Rate% (CR) and Crit DMG% (CD). A piece never rolls a substat identical to its own main stat, and never repeats the same substat twice before reaching 4 lines.

• A 5★ piece drops with 3 or 4 substat lines already rolled at random.
• Every +4 levels (up to +20) it gets a value added to an existing line (or unlocks a new one if under 4 lines) — 5 upgrades total over its lifetime.
• For a piece that starts with all 4 lines: total value-additions = 4 (initial) + 5 (upgrades) = 9 rolls spread randomly — averaging ~2.25 rolls per line, but luck can pile most of them onto just 1-2 lines.

Since it's pure RNG, locking the right main stat still matters more than obsessing over substats first — see the artifact overview page.

Every time a substat gets a value added, it lands on one of 4 tiers: 70% / 80% / 90% / 100% of its max possible value ("high roll" = near 100%). Per-roll ranges on a 5★ piece (cross-checked KQM + BitTopup):

• Crit Rate%: 2.7% – 3.9% (top roll 3.89%)
• Crit DMG%: 5.4% – 7.8% (top roll 7.77%)
• ATK% / HP%: 4.1% – 5.8% each (same tier group, so they roll identically)
• Energy Recharge%: 4.5% – 6.5%
• Elemental Mastery: 16 – 23 points
• Flat HP: 209 – 299
• DEF%: ~5.1% – 7.3% (flagged: single clear source — DEF% always rolls higher than ATK%/HP% due to a different main-stat coefficient)
• Flat ATK / Flat DEF: ~14-19 and ~16-23 (flagged: single source — few guides bother listing these precisely since it's the substat worth dumping first)

A theoretically perfect 5★ piece (every roll hitting 100%) reaches a Roll Value of ~900% — a purely theoretical number almost nobody actually hits (single KQM source, not fully cross-checked). To collapse Crit Rate/Crit DMG into one comparable score, use the Crit Value formula on the Crit Rate/Crit DMG ratio page.

There's no single priority chart that fits EVERY character, but the community consensus (genshin-builds, KQM, BitTopup) settles on this framework for most main DPS:

• S — always keep: Crit Rate%, Crit DMG% — a piece rolling ≥20% Crit DMG or ≥10% Crit Rate from substats alone is already solid.
• A — high priority: ATK% — still contributes near-linear damage, don't discard it just because it's not crit.
• B — team-dependent: Energy Recharge% (only needed if your ult isn't charging in time each rotation) and Elemental Mastery (only worth it if the character actually triggers or buffs a reaction — see the block below).
• F — nearly worthless: Flat HP/ATK/DEF — genuinely useful during AR1-25 when base ATK is only ~200, but past AR30 they're basically dead substats; 3 flat lines combined still lose to a single Crit Rate line.

A practical stopping rule: a piece is "good enough" once it has the correct main stat plus at least 2 of the 4 substats you actually need (usually Crit Rate, Crit DMG, ATK% and one flex slot) at decent rolls — you don't need to wait for all 4 to be perfect.

This is the most argued-over substat because its value depends ENTIRELY on the reaction type a character actually uses — having reactions in your kit doesn't automatically make EM good:

• Amplifying reactions (Vaporize, Melt) still scale with ATK, Crit and DMG Bonus like normal damage — EM is only a bonus on top, it can't replace ATK%/Crit.
• Pure Transformative reactions (Superconduct, Overloaded, Swirl, Bloom/Hyperbloom) do NOT scale with ATK or DMG Bonus and can't crit — damage depends only on character level and EM. This is where EM becomes the #1 DPS substat, outclassing even Crit.
• Catalyze reactions (Aggravate/Spread) are NOT the same bucket as pure Transformative even though they also scale hard with EM: this is bonus damage added on top of the triggering hit, so its base value (from level + EM) still CAN crit and still gets the DMG Bonus of that hit — Raiden/Cyno/Keqing aggravate builds still need Crit, not just EM.

Practical rule: pure Transformative builds (e.g. a Bloom/Hyperbloom sub-DPS, an EM-stacking team) → prioritize EM on par with Crit; Aggravate/Spread builds (Raiden, Cyno, Keqing...) → still prioritize Crit Rate/Crit DMG like a normal DPS (the bonus damage can crit), then stack EM on top to boost the base — don't dump Crit entirely for EM; normal DPS builds (Vaporize/Melt/physical) → EM is dead weight, prioritize Crit/ATK% first. See how each reaction scales in detail on the elemental reactions page.

The community often says "dump everything into Crit, ignore ATK%" — but a mathematical optimization model (Lagrange multipliers) for a standard DPS build (ATK% Sands, DMG Bonus Goblet, Crit Rate Circlet) actually shows: out of all available rolls, the optimal split is ~7 rolls into ATK%, ~14 into Crit Rate, ~17 into Crit DMG — meaning ATK% is still worth ~19% of total rolls, not 0% (flagged: single source — one author's specific model, assuming a standard build, not an absolute rule for every character).

• The Crit DMG:Crit Rate ratio from this model comes out to ~182.6% : 90.7%, still roughly 2:1 — matching the CD + CR×2 Crit Value formula the community commonly uses.
• Practical takeaway: don't go to the extreme of discarding every ATK% line chasing pure Crit — a piece with the correct main stat + 2 Crit lines + 1 ATK% line + 1 flex line is already a good build; you don't need to wait for all 4 lines to be Crit.

To understand why the 2:1 ratio is optimal and how to apply it per role, see the detailed Crit Rate/Crit DMG page.

"Off-piece" means using a piece that does NOT belong to your main set (say, a Circlet from a different set) purely because its main + substats are excellent — a 4pc set bonus doesn't always beat a piece with perfect substats, especially on Sands/Goblet/Circlet slots which already have many main stat options.

• Feed it immediately: a piece with ≥2 Flat HP/ATK/DEF lines out of its 4 — it's almost guaranteed worse than a fresh roll.
• Keep even if imperfect: a piece with the correct main stat + ≥2 of 4 substats in your needed group (usually Crit Rate, Crit DMG, ATK%/EM depending on build) even at non-max rolls — this is the “good enough” bar to equip right away instead of chasing further.
• Wait on Strongbox/Transmuter: if you're missing exactly ONE of the two sets a domain drops, the Strongbox (3x 5★ into 1) or the newer Artifact Transmuter/Reshaping lets you target the right set or reshape main/substats without re-farming from zero — see full mechanics and costs on the efficient artifact farming page.

Anti-regret tip: when torn between keeping or feeding, ask “is this piece blocking a slot another build needs?” — if not, just let it sit until a clearly better replacement shows up instead of feeding it hastily.

Most new (and even veteran) players waste resources by repeating a few basic mistakes:

• Forcing 100% Crit Rate: anything past 100% Crit Rate is pure waste — the overflow does nothing, so stop around the point where CR:CD sits near 1:2 instead of maxing one stat alone.
• Treating EM as a “universally good” substat: applying one character's logic to every build — EM is only strong for Transformative reactions, and near-useless otherwise (see block 4).
• Keeping a piece “out of pity” despite 2+ flat lines: sunk-cost feelings don't change the damage math — flat lines don't scale with %, so their relative value shrinks the more your character levels up.
• Comparing two pieces by raw summed % without converting: 1% Crit Rate ≠ the value of 1% ATK% or 1 point of EM — use the Crit Value formula or a damage calculator instead of adding numbers by feel.
• Waiting for a “perfect” 4/4 roll before using a piece: per the optimal distribution in block 5, a piece just needs to hit the “good enough” bar to be equipped right away — save your resin for farming other characters instead.

To quickly check what your current build is missing, use the damage calculator on the damage calculator page to compare before/after a substat swap.