The final week of Version 3.2 Their Secret Histories closes on Oct 20, and Version 3.3 Journey to the Sky arrives Oct 21 with Phoenix, a Fire Anomaly agent, and Severian, a Wind Attack agent. For players still farming talent nodes, the most common mistake is not a lack of Dennies but a misallocated Battery Charge. Dumping Charge into Dodge or Assist on a Stun unit while the Special Attack that builds Daze sits at rank 4 is a wasted week of farming, not a build. The fix is a role-based upgrade order that respects how each kit actually produces damage, stun, or support value.
⚙️ The Universal Order: Level, Core, W-Engine Before Any Talent Node
Agent Level and Ascension come first because they unlock base stats and raise the skill cap. Anything else is capped progress. Core Skill, Rank A through F, is the first talent node players should touch; it scales the unique passive and team-wide buffs that no Basic Attack rank can replace. W-Engine Level is third for Base ATK and specialist stats such as Impact or Anomaly Proficiency. Drive Discs should stay at 12 until the set is optimized, then push to 15. This sequence prevents the classic trap of spreading Battery Charge across low-impact nodes while the kit's engine remains underpowered.
⚔️ Attack and Anomaly: Damage Windows vs. Buildup Math
For Attack agents such as Severian and Hoshimi Miyabi, the order is Basic Attack ≥ Special Attack > Chain/Ultimate > Dodge > Assist. Damage is proactive, so reactive nodes wait. Anomaly agents flip the script. Phoenix needs Core Skill as a mandatory stop for Disorder and Anomaly CRIT multipliers, then Special Attack for the highest Buildup per hit. A Phoenix at Core F with rank-6 Special will out-Disorder a Basic-maxed build every rotation. Players should time Disorder windows around the stun burst, not raw Basic uptime.
🛡️ Stun, Support, and Defense: Where the Core Skill Carries the Kit
Stunners such as Nangong Yu and Lycaon prioritize Special Attack > Basic Attack > Dodge; the goal is to get Stun state online fast, so players chase the highest Daze multipliers. Supports such as Astra Yao and Rina lead with Core Skill for buff potency, then EX Special for energy generation. Defense units follow the same Core-first rule because shield scaling often lives on the Core passive, with EX Special next for invulnerable frames. This is stun-lock math: Daze multipliers before reactive safety.
| Role | Upgrade order |
|---|---|
| Attack | Basic ≥ Special > Chain/Ultimate > Dodge > Assist |
| Anomaly | Core > Special > Basic > Chain/Ultimate > Dodge |
| Stun | Special > Basic > Dodge > Assist > Chain/Ultimate |
| Support | Core > Special > Assist > Basic > Chain/Ultimate |
| Defense | Core > Special > Assist > Dodge > Basic |
| Rupture | Core > Basic > Special > Chain/Ultimate > Dodge |
💥 Rupture: The Sheer Force Exception
Rupture agents such as Yixuan do not follow the Attack or Anomaly template. They scale off HP and Sheer Force, so Core Skill converts those stats into defense-ignoring damage—it is the whole kit. Basic Attack edges out Special Attack because the Sheer Force conversion rides on sustained strings, not single nukes. Dodge stays last; defensive stats do not replace the HP-to-Sheer Force engine.
🗺️ 3.3 and 3.4 Roadmap: Spend Mats Where the Meta Is Heading
According to the HoYoverse Zenless Zone Zero 3.3 Special Program, the Version 3.4 Honkai: Star Rail crossover is confirmed with Kafka as an Anomaly agent and Robin as a Support. That crossover is scheduled for December 2026, so players should bank Anomaly and Support materials now. Battery Charge is the bottleneck, not Dennies. Some players weigh zenless top up options before a new S-Rank drops, but the smarter play is to reserve Charge for Core and role-critical active skills instead of speculative nodes.
Before patch-week testing lands, players should hold off over-investing in Assist nodes because the new Mech Combat system in 3.3 may shift value toward Assist stats. Spend Battery Charge on Core and role-critical active skills first, not on M0-versus-M6 vanity ranks. Patch-week data will tell the community whether Assist deserves a bump; until then, the universal order still wins.