Twenty-one escort craft — dragons, war-beetles, UFOs and two Transformers — that fly beside your fighter and pulse buffs on a timer. Internally the system is called "Guard"; economically it is the Sentinel Chip funnel: a refresh-gacha with a 0.2% jackpot, a 40-star ladder, and a four-slot passive recast.
Sentinels are the third companion collection, unlocked late — player level 58 (drones come at 45). A deployed Sentinel escorts your fighter and fires its Talent Skill on a fixed cycle: "After battle starts, release a skill every {n} seconds…". Every kit is that same sentence with a different payload — damage, healing, or a stacking buff.
The build guides in the game pair one Sentinel with a trio of drones per formation, and the Sentinel kits are transparently designed as the drone layer's support column: their skills and passives raise electric / burning / reflect enhancement — the exact three damage channels the drone elements deal — plus healing, crit and dodge.
A Sentinel's Talent Skill has 5 levels, gated by the Sentinel's own level: unlocked at Lv 1, upgraded at Lv 20 / 40 / 60 / 80.
Five SSR (including the Transformers collab units Onslaught and Ironhide), six SR, five R, three N and two G. The UI sorts them into five type families — Avian, Insecta, Special, Draconic, Wing Plane — which appear to be flavor/filter groupings; the mechanical identity lives in each unit's Talent Skill.
Every Talent Skill is the same chassis: fire every N seconds, deal (or heal) a % of ATK, apply a rider. The grade hierarchy is visible in the riders: SSR riders stack and last until battle ends (permanent snowball), while SR-and-below riders are timed buffs that expire. Healing kits are explicitly PvP-only ("Healing effect works only in PvP battles") — like drones, this layer's sharpest edge is the PvP metagame.
The numbers behind {0}–{4} are in the game data: skill.json carries 105 guard rows (21 Sentinels × 5 skill levels) with the placeholder values, the cooldown, the damage rate and a link into battle_buff — which is what the combat layer actually reads. That makes every kit directly comparable.
Two facts collapse the comparison. First, skill level only scales the damage or heal number — the rider is identical at Lv 1 and Lv 5 on all 21 units, verified across every row. Second, every timed rider lasts exactly 2.1 seconds, so a kit's real buff strength is value × (2.1 ÷ cooldown). A permanent rider is worth roughly three times the same number on a 6-second cycle.
That gives one honest yardstick for damage — damage rate ÷ cooldown, in multiples of ATK per second at max skill level — and one for the rider, its time-averaged value. The mode scalars (pveRate 1.875 / pvpRate 0.025) are identical on all 21 Sentinels, so they cancel out of any comparison between them.
Sustained skill damage at max skill level, in multiples of your ATK per second. Grade tracks this loosely at best: one SR tops the chart, and Savageplate — an N — out-damages every R in the game purely because it fires on a 2-second cycle instead of a 6-second one.
The five healers (skill.type == 3) deal no damage whatsoever. Their output is flagged PvP-only in their own descriptions, which makes them dead weight in campaign and specialists in arena.
Comparing each description's placeholders against the buff the combat layer actually applies turns up two discrepancies. battle_buff is what runs, so behaviour follows the data, not the text:
A Sentinel's passives are the only part of it you choose. Each slot rolls one of 18 families in one of three colours, and Recast lets you re-roll them — so the real question is which families are worth locking. The answer depends on the unit's own kit and, sharply, on the mode.
Three rules do most of the work:
#fleet pool as the matching passive, so Onslaught + Electric Enhancement, Swiftplume + Burning, Ironhide + Reflect all stack additively — and they are the same channel the drone trio in that formation deals.These are starting shortlists based on each kit, not a measured best-in-slot ranking. Switch modes for damage-focused or opponent-focused suggestions. Existing stats, the enemy, drone composition and battle length can change the order; the individual explanations describe those tradeoffs.
A passive contributes a stat while its Sentinel's passives are active in your formation. It has no separate attack, proc chance, cooldown or timed duration in the passive configuration. The Talent Skill is a different system: it fires on a timer and may add a temporary or stacking buff. Keep those two sources separate when calculating a build.
The precise part we can calculate is which stat receives how much. The extracted tables also identify each stat's intended effect. They do not expose the complete runtime damage, healing or opponent-counter formulas. Examples labeled illustrative combat model explain what an assumed formula would do; they are not verified combat predictions. Evidence and limits are listed below.
A stored value of 0.25 means +25 percentage points to that attribute. If your Burning Enhancement pool already contains 30%, an active +25% passive makes it 55%, not 37.5%. That does not automatically mean your final damage rises by 25%: another enhancement, an enemy counter, a cap, or the share of your damage coming from that source changes the result.
Every passive family has 10 levels and a linear ladder within each color: value at level L = that color's L1 value × L. L10 is the total held at level 10, not an extra amount to add to L1–L9. The progression tables in each explanation show both endpoints. A passive's color is separate from the Sentinel's G/N/R/SR/SSR grade; grade controls the number of slots.
Different active sources of the same stat add to its pool. This arithmetic does not prove that duplicate families can occupy every slot, that passives on undeployed copies contribute, or that a timed Talent Skill buff can exceed its own stack limit.
The configured Burning Enhancement stays at +30 percentage points. Sentinel Enhancement raises the stat used for Sentinel damage and healing. Swiftplume's separate buff writes a fixed burnUp#fleet = +0.30; it contains no reference to Sentinel Enhancement, attack, or the damage dealt. An Orange L10 Sentinel Enhancement passive adds guardUp#fleet = +0.75. These are two different attributes.
Molten Core's direct hit: the first skill is scheduled at 4 seconds, then every 4 seconds. Its damage coefficients at skill levels 1–5 are 17.3 / 20.7 / 24.2 / 27.6 / 31.1 × attack. The five skill levels unlock at Sentinel levels 1 / 20 / 40 / 60 / 80. Raising the skill level increases that direct coefficient; all five levels link to exactly the same burning buff.
The burning rider: the buff has one stack, a fixed +30% value, and no expiry configured. The data therefore supports one persistent +30% contribution after activation, not another +30% every four seconds. Its English description omits the permanence wording; the no-expiry structure matches the other persistent Sentinel buffs. Exact event ordering within the skill activation is not exposed.
Worked stat example: start with 20% Burning Enhancement from other sources. Swiftplume's rider brings the pool to 50%. Adding Orange L10 Burning Enhancement brings it to 75%. Adding Orange L10 Sentinel Enhancement instead leaves the burning pool at 50% and adds 75 points to the separate Sentinel Enhancement pool. It does not turn Swiftplume's +30% rider into +52.5%, change the four-second cycle, or unlock a second stack.
Illustrative combat model: let an otherwise resolved direct Sentinel hit be 10,000 damage before Sentinel Enhancement. If its factor is 1 + guardUp, with no existing enhancement, enemy decrease, decay or other changes, +75% gives 17,500. If the existing pool is already +50%, that model moves 15,000 to 22,500: a 50% relative increase, despite adding 75 stat points. The exact runtime factor is unverified; the worked numbers illustrate why raw passive percentages are not universal final-damage percentages.
Which passive helps which part? Sentinel Enhancement targets Swiftplume's direct output. Burning Enhancement targets the damage of Burning drones described by that stat's tooltip. Swiftplume supplies a burning-stat bonus; that does not classify its own direct hit as a Burning-drone attack or prove that the skill applies a burn damage-over-time effect. To support your burning drones, compare Burning Enhancement against their existing pool and the opponent's Burning Decrease. To strengthen Swiftplume's own hit, compare Sentinel Enhancement against its existing pool and the opponent's Sentinel Decrease.
This passive adds atk#tgt70, the Sentinel contribution. It is not an Attack % bonus and does not modify the Sentinel's skill coefficient. The published main-gun, winggun, missile and ultimate attack formulas each finish with an additive + atk#tgt70. The same stat can therefore contribute to more than Swiftplume's own output.
Exact formula example: with Orange L10, that input increases by 1,000. If all other inputs are held fixed, a 50,000 main-gun attack result becomes 51,000 in the published attribute formula. The other three listed attack formulas each receive the same +1,000 contribution. This term sits outside their (1 + atkP#total) multiplication; those particular total-Attack-% terms do not multiply it again.
Limits and use: no trigger or duration is attached to the passive. The exact attack value chosen by a Sentinel's runtime skill calculation is not established by the label “% of attack”; do not automatically translate +1,000 sheet attack into 1,000 × 31.1 extra Swiftplume damage. Treat it as a confirmed flat contribution to the listed attack formulas. Its relative value becomes smaller as those attack totals grow.
This passive writes hpMAX#tgt70. The published maximum-HP formula ends with + hpMAX#tgt70, so this is a flat contribution to your combined maximum HP. It does not describe a separate Sentinel health bar, regenerate HP on a timer, or increase a skill's healing coefficient.
Exact formula example: Orange L10 adds 4,000 to that input. Holding other inputs fixed, 100,000 calculated maximum HP becomes 104,000. This is a 4% relative increase in that example; the same passive adds only 0.4% to a 1,000,000 total. The term is outside the formula's (1 + hpMAXP#total) factor, so that factor does not turn this passive's +4,000 into a larger contribution.
Limits and use: a larger maximum is different from restoring missing HP. Any skill based on a percentage of maximum HP could depend on the larger total, but that must be checked in that skill's own formula. Sentinel heals described as a percentage of attack should not be assumed to scale with this HP passive. Current-health adjustment when a formation changes and mode-specific normalization require runtime evidence.
This adds to critDamage#total. Orange L10 contributes +50 percentage points to the critical-damage stat; it does not add 50 points to Crit Rate. The published stat formula reads this pool directly. The Combat Power formula subtracts a 2.0 baseline, consistent with the guide's 200% base critical-damage interpretation.
Exact stat example: a 2.00 critical-damage input becomes 2.50. Illustrative combat model: if a critical hit is normal damage × that input and there is no opposing reduction, a 10,000 normal hit would crit for 20,000 before the passive and 25,000 afterward. That is 25% more damage on a critical hit, not 50% more total damage. At a fixed 20% crit chance, average damage in this simplified example moves from 12,000 to 13,000, an 8.33% gain.
Limits and use: it needs an attack that can crit and a successful critical roll. Rendclaw's Crit Rate rider can improve its usefulness by making eligible crits more frequent, but the passive data does not establish that every Sentinel skill, burning tick, reflected hit or healing event can crit. Enemy Crit Damage Reduction and the actual combat formula determine the realized gain.
This adds critDmgRed#total. Its tooltip describes reducing damage taken from enemy critical hits. The passive raises your defensive stat; it does not install a timed negative stat on the enemy, remove their chance to crit, or provide a general shield against every hit.
Exact stat example: an existing 20% reduction input plus Orange L10's 50 points gives 70%. Illustrative combat model: if the critical multiplier is enemy critDamage − your critDmgRed, an enemy with 2.50 and your new 0.70 would resolve to 1.80. With the original 0.20 it would resolve to 2.30. That shows why +50 points is not the same as halving all incoming damage.
Limits and use: benefit depends on the enemy's crit frequency and critical-damage magnitude. Crit Resistance addresses frequency; this passive addresses magnitude. The data does not publish the minimum critical multiplier, clamp order or a blanket campaign exclusion for this passive. It is especially relevant against opponents that crit, but a target that never crits gives it no corresponding damage event to reduce.
This adds elecUp#fleet. The in-game tooltip says the stat increases all damage from Electric drones, by up to 100%. It describes the drone's element channel, not an extra lightning attack created by the passive. The published attribute formula combines elecUp#fleet + advantage#fleet; Combat Attribute Boost can therefore contribute to the same displayed offensive result.
Exact pool example: Onslaught's persistent +30-point Electric Enhancement rider and an Orange L10 passive's +25 points contribute 55% to the raw electric pool. A separate +10-point Combat Attribute Boost makes the published combined attribute 65%, assuming no other sources. Neither bonus increases Onslaught's fixed +30-point rider itself.
Limits and use: the tooltip's “up to 100%” describes the effect ceiling; it does not expose the conversion from a high raw stat to the final bonus. Enemy Electric Decrease is the named counter. Do not assume every point converts linearly to final damage, or that this buffs unrelated Sentinel attacks because Onslaught granted the stat. It is useful when Electric drones contribute meaningful damage; without a matching source there is nothing identified by the tooltip to enhance.
This adds your own elecDown#fleet. The tooltip says it resists the enemy's Electric Enhancement. The published defensive attribute combines elecDown#fleet + resistance#fleet; Combat Resistance Boost is another source of that combined result.
Exact stat example: Orange L10 contributes 30 points. With an existing 10-point electric-decrease source and 5-point Combat Resistance Boost, the published combined attribute is 45%. This is not a confirmed 45% reduction to all incoming damage, and the passive does not directly subtract 45 points from an enemy's stored electric stat.
Illustrative combat model: a simple net-stat model against 60% Electric Enhancement would compare 0.60 − 0.45 = 0.15. The real conversion, floors and order are not recovered, so 15% is only the result under that model. Choose this passive to answer an opponent's electric damage channel. It does not establish protection against burning, reflect or Sentinel damage, and has no identified benefit against an enemy that has no electric mechanic.
This contributes burnUp#fleet. Its tooltip identifies all damage from Burning drones, by up to 100%. It does not say “extend burn duration,” “apply a burn,” or “increase the Sentinel's fixed buff.” The published attribute formula reads burnUp#fleet + advantage#fleet.
Exact pool example: Swiftplume's active rider gives +30 points; Orange L10 Burning Enhancement gives +25. Together they contribute 55% to the burning pool. Starting from an additional 20% supplied elsewhere gives 75%. Those are additive stat inputs, not multipliers of 1.30 × 1.25. Swiftplume's own contribution remains exactly +30 points and remains limited to one stack.
Limits and use: enemy Burning Decrease counters this channel. The 100% effect ceiling and any nonlinear conversion mean the final gain cannot be read directly from the raw pool. Individual effects may also reference the burning stat through their own formulas; those need separate verification. This passive does not create a burn damage-over-time duration, tick rate or stacking rule. See the Swiftplume example for the direct-hit versus drone-support decision.
This adds burnDown#fleet, whose stated purpose is to resist enemy Burning Enhancement. The published combined attribute is burnDown#fleet + resistance#fleet. It is a defensive contribution on your side, not a cleanse applied to an existing burn status.
Exact stat example: 15% existing Burning Decrease plus Orange L10's 30 points gives 45% in that pool. If you also have 10 points of Combat Resistance Boost, the published combined attribute becomes 55%. This does not reduce the configured +30% value in an opposing Swiftplume's rider; it provides the counter-stat used when resolving the relevant enemy effect.
Limits and use: it does not promise immunity to burning, fewer ticks, shorter duration, or a reduction to Swiftplume's direct skill hit. Those are separate mechanics. In an illustrative subtraction model, an enemy's 75% combined enhancement versus your 55% combined decrease leaves 20 points; the actual damage conversion and floor are unverified. Its value depends on how much of the opposing formation's threat comes from the burning channel.
This contributes refUp#fleet. The tooltip says it increases all damage from Reflect drones, by up to 100%. The published attribute formula adds advantage#fleet to that pool. The passive itself supplies no event trigger, retaliation probability or damage-reflection percentage.
Exact pool example: Ironhide's persistent rider contributes 30 points and Orange L10 Reflect Enhancement contributes 25, making 55% before other sources. Adding 10 points of Combat Attribute Boost produces a 65% combined attribute in the published formula. As with Swiftplume, the passive adds to the pool; it does not change Ironhide's rider from +30% to +37.5%.
Limits and use: owning this stat does not by itself make your fighter reflect an incoming attack. A drone or skill must supply the relevant damage mechanic. Reflect Decrease is the opponent's counter, and the tooltip's effect cap does not reveal the exact runtime curve. Prefer this family when Reflect drones are a substantial part of the formation's output, rather than treating the word “reflect” as a universal defensive shield.
This adds refDown#fleet. Its tooltip describes resisting the enemy's Reflect Enhancement, and the published attribute formula combines it with resistance#fleet. It does not state that it disables an enemy reflect ability or reduces that ability's activation chance.
Exact stat example: a 20% existing reflect-decrease pool plus Orange L10's 30 points gives 50%. With another 5 points from Combat Resistance Boost, the combined attribute becomes 55%. Against 80% enemy enhancement, a simple subtraction model would compare an effective difference of 25 points. That comparison is illustrative; it is not evidence of a 25% final reflect multiplier.
Limits and use: this is relevant to a matching opponent damage channel. It does not claim to reduce all outgoing damage that might later be reflected, all enemy attacks, or the fixed stat buff granted by Ironhide. Whether a specific retaliation effect uses this channel must be established from that effect. With no enemy reflect-related output, this passive has no matching purpose identified by its tooltip.
This adds crit#total. The published total is crit#total + advantage#fleet. Orange L10 supplies 25 percentage points. The tooltip states that the maximum effect against a player is 80%; the raw stat can still contain contributions beyond that effective cap.
Exact input example: a 10% starting crit pool, Rendclaw's fully stacked +25-point rider, and this passive's +25 points produce 60% before other sources. Add 10 points of Combat Attribute Boost and the published combined input becomes 70%. Do not add a second 10% baseline if the starting stat already includes it.
Illustrative combat model: subtracting an opponent's 20% Crit Resistance from that 70% input gives 50% crit chance before any other processing. The actual counter conversion and cap order are unverified. More crit chance only helps damage events allowed to crit; it does not increase the damage of each crit, skill cadence, or a fixed Sentinel buff. Compare Crit Damage when eligible crits are already frequent, and remember that opponent resistance can change whether an apparent raw-stat excess is useful.
This adds critResist#total; the published combined defensive input is critResist#total + resistance#fleet. The tooltip says it reduces the chance of receiving critical-hit damage. It addresses whether an eligible hit becomes critical, while Crit Damage Reduction addresses its magnitude if it does.
Exact stat example: Orange L10's 30 points plus 10 points of existing Crit Resistance and 5 points of Combat Resistance Boost produce a 45% combined input. Illustrative combat model: if an enemy's effective crit input is 60% and counters subtract, the result would be 15% instead of 45% without this passive. This is a percentage-point difference, not “30% fewer crits” as a relative calculation.
Limits and use: there is no passive-specific timed duration or proc. Benefit requires enemy crit-capable damage. The tables do not establish the runtime lower bound or exact order of the opponent's 80% cap and your resistance. Do not assume that matching the enemy's displayed number guarantees immunity, or that Crit Resistance reduces the size of a normal hit.
This adds dodgeProb#total. The published dodge attribute combines that with advantage#fleet. Its tooltip describes dodging enemy bullets and states a maximum effect of 80% against players. The passive does not move the fighter, increase movement speed, or provide a timed invulnerability window.
Exact stat example: 15% existing dodge, Orange L10's 25 points and 10 points of Combat Attribute Boost make a 50% combined dodge input. Illustrative combat model: using the PvP guide's 85% base hit and an enemy's 20% hit bonus, the modeled hit chance would be 85 + 20 − 50 = 55%. Removing this passive would make it 80% in that model.
Limits and use: the actual miss probability depends on enemy accuracy, caps and attack eligibility. Neither the passive nor its tooltip proves that every Sentinel skill, periodic tick or reflected event can be dodged. A separate skill that triggers on a successful dodge can benefit indirectly when eligible dodges become more frequent; the passive does not change that skill's trigger cooldown, stack cap or fixed reward.
This adds accuracy#total. The published hit-rate attribute is accuracy#total + resistance#fleet. Despite sounding offensive, it therefore receives Combat Resistance Boost in these formulas. Its tooltip promises a greater chance to hit; it does not promise larger damage on an attack that already lands.
Exact stat example: Orange L10's 30 points, an existing 5% accuracy contribution and 10 points of Combat Resistance Boost make a 45% combined accuracy input. Illustrative combat model: against 50% dodge, an 85% base-hit model gives 85 + 45 − 50 = 80% hit chance. Without this passive, the same model gives 50%.
Limits and use: the 85% baseline belongs to the PvP reconstruction; do not add it twice to a figure already presented as a final chance to hit. Actual runtime caps, guaranteed-hit attacks and the stat conversion need separate confirmation. Accuracy is most useful when missed eligible attacks are a real loss of output. Once those attacks already land reliably, raising their damage may offer more than additional raw accuracy.
This adds guardUp#fleet. The tooltip explicitly covers damage dealt and healing provided by Sentinels. It is a shared fleet attribute, not the flat Attack passive and not a multiplier embedded into each Talent Skill's buff. The published attribute formula reads it directly; Combat Attribute Boost is not added to it there.
Exact stat example: 18% from Timesand Tech and Orange L10's 75 points give 93% before other sources. The attack contribution of a Sentinel's levels, stars or Attack passive is a separate input. Swiftplume's +30% burning rider, Rendclaw's fixed crit-rate increments and a skill's cooldown/stack cap are likewise separate configured properties.
Illustrative combat model: under output × (1 + guardUp), 10,000 base output moves from 11,800 at +18% to 19,300 at +93%, a 63.56% relative gain. This is not a verified final-damage or final-healing prediction: enemy Sentinel Decrease, possible stat conversion and other combat factors remain to be resolved. The data does not publish how Sentinel Enhancement and Healing Enhancement combine on a Sentinel heal.
Limits and use: this passive does not create healing on a damage-only Sentinel. The healing skills' PvP restriction remains in force. There is no numeric effect cap stated in the tooltip, so the L10 +75% is a per-slot contribution cap, not proof of a +75% total-stat cap or unbounded linear damage. Read the Swiftplume breakdown and the source list for the two different calculations.
This adds guardDown#fleet. The tooltip explicitly describes reducing damage dealt and healing provided by enemy Sentinels. The published attribute formula reads this pool directly; Combat Resistance Boost is not included. It is your counter-stat, not a timed debuff written into the enemy's Sentinel configuration.
Exact stat example: 18% from another source plus Orange L10's 75 points gives 93%. It does not rewrite an opposing Swiftplume's fixed +30% Burning Enhancement buff to +7.5%, shorten its duration or remove its stack. The intended target is the enemy Sentinel's direct damage/healing output.
Illustrative combat model: if the factor were 1 + enemy guardUp − your guardDown, an enemy at +150% and you at 93% would give a factor of 1.57. Without the passive, your 18% would give 2.32. The actual formula and lower bounds are not recovered, so +75 points does not establish “75% less final damage.” Its value depends on enemy Sentinel output; an enemy with no Sentinel damage or healing offers no matching output to suppress.
This adds cureUp#fleet. Its tooltip covers all healing provided by you, a broader description than Sentinel Enhancement's Sentinel-only scope. The published attribute formula reads the pool directly. This passive itself has no healing interval, target-selection rule, base heal amount or conversion of damage into healing.
Exact stat example: Orange L10 adds 100 points. Doomblast's rider supplies 10 points per stack, up to five stacks; at its full +50%, the two contributions make 150% before other sources. The passive does not raise Doomblast's per-stack 10% to 20%, change its stack limit or accelerate the Talent Skill's cycle.
Illustrative combat model: if a 10,000 eligible heal uses 1 + cureUp, +50% gives 15,000 and +150% gives 25,000, a 66.67% relative increase. This assumes no enemy Healing Decrease, missing-HP limit or other multiplier. A Sentinel heal may also be affected by Sentinel Enhancement, but whether these factors add, multiply or undergo another conversion is unverified.
Limits and use: no healing source means no corresponding gain. Sentinel healing skills specifically state PvP-only healing, and this passive does not enable them in campaign. “All healing” also does not prove how every lifesteal, revival, shield or pickup is categorized; those effects need their own evidence. Do not assume a stronger heal can exceed maximum HP or select a different recipient.
This adds cureDown#fleet. Its tooltip says it reduces all healing provided by enemies, rather than only Sentinel healing. The published attribute reads this pool directly. It does not deal damage, remove maximum HP, rewind a heal already received or apply a separate timed healing lock.
Exact stat example: an existing 20% contribution plus Orange L10's 100 points gives 120%. This is a raw stat value; it does not establish 120% less healing or turn a heal into damage. No runtime healing floor or complete Enhancement-versus-Decrease expression is published in these tables.
Illustrative combat model: under heal × (1 + enemy cureUp − your cureDown), an enemy with +150% Healing Enhancement and your 120% Decrease would turn a 10,000 base heal into 13,000. Without the passive, your 20% Decrease would yield 23,000. The enemy would still heal in that model despite your passive reading +100%. Sentinel Decrease could also apply to a Sentinel-origin heal; the interaction and calculation order are unverified.
Limits and use: choose this to answer meaningful enemy sustain. Without an eligible enemy heal, it has no matching effect. The stat's PvP-enabled flag alone does not establish a universal PvE ban or prove that every enemy recovery mechanic uses the same healing channel.
The September 26 configuration snapshot contains 54 passive definitions and 540 level rows. Names and exact contributions come from guard_passive_skill and guard_passive_skill_level. All 540 values are represented in the passive ladder table. The progression summaries above reuse that same extracted data. Attribute identities come from attrs_base / attrs_combo; scope and stated caps come from the game's English stat tooltips.
Attribute formulas: power_calculation rows 1–4 and 10 show the flat Attack/HP terms; 14–15, 54–63 and 66–69 show the combined accuracy, dodge, elemental, Sentinel, healing and critical stats. Its attr_algorithm describes a stat calculation. Its power_algorithm scores Combat Power; the attack/HP clamps in that score are not damage caps. The isDecay flag does not supply a usable runtime decay curve.
Swiftplume: guard_unlock_skill and skill rows 100401–100405 give the unlock levels, coefficients and four-second cadence. battle_buff 1000401 links to 10004012, which writes burnUp#fleet: 0.3, with maxStack: 1 and no duration. Sentinel Enhancement is a separate write in passive level group 15. This supports the fixed +30-point rider and separate output stat; the complete runtime combat function is not recovered.
All 18 passive attributes are marked as enabled in PvP. That flag is not proof of PvP exclusivity. Unknowns remain around opponent-counter conversion, runtime caps and decay, damage-event eligibility for hit/crit rolls, exact attack inputs for skills, healing recipient/overheal handling, and Enhancement/Decrease calculation order. The PvP calculator uses a reconstructed combat model; treat it as a comparison tool with those limits, not a replacement for measured combat evidence.
Owning specific Sentinels activates permanent account-wide set bonuses, each themed around the members' shared family — Fetters are the Sentinel equivalent of the drones' Combination sets. The premium set is Finale Overture: all four non-healer SSRs for +3% PvP damage — a collection tax squarely aimed at arena players.
Levels 1→150 cost Overclock Chips only, and the bill scales hard with grade: maxing a G costs 191,398 chips; an SSR costs 3,826,519 — exactly 20× — for 3× the stats (SSR caps at 18,656 HP / 4,664 ATK vs the G's 6,220 / 1,555).
A wrong investment can be undone: Reset refunds all materials and drops the Sentinel back to level 1, priced by grade (100 → 300 Diamonds, G → SSR). Star level survives the reset.
Reclaiming (selling) a Sentinel returns 80% of everything spent — and the game even dangles "a chance to obtain extra special items when reclaiming."
The full 150-level curve, every grade side by side — HP / ATK held at each level, and the Overclock Chip cost to advance from it:
Evolution consumes one same-name Sentinel per star, 40 stars total — so a maxed unit is 41 copies. Unlike drones there is no escalation: always one feeder per star. Each star adds a flat HP/ATK block, and every evolution also randomly upgrades a passive skill level or grants a new passive (up to the unit's slot cap) — evolution and the passive lottery are the same button.
Passive slots are the real grade gap: G/N/R hold 1–2 passives, SR holds 3, SSR holds 4. Each passive rolls from 18 families in three grades — Blue Purple Orange — at published odds of 73% / 18% / 9%, and levels 1–10 within each grade. The library is two mirrored halves: enhancements for your fleet (electric/burning/reflect/healing up, crit, dodge, Sentinel damage) and suppressions against the enemy's (the same list, downward) — pure PvP arms-race design.
The new Universal Recast Core item and English instructions describe replacing an existing passive with a chosen Blue, Purple or Orange skill while retaining its current level. Higher-grade selections cost more cores. The text says each Recast Core or Protection Core spent on recasting grants one Universal Recast Core.
A new compensation-mail template refers to previously consumed cores. The inspected tables and recovered code do not establish the selection prices, eligibility, compensation calculation or live rollout. The random-recast costs below remain unchanged in the supplied data.
Recast rerolls all unlocked passive slots (levels are retained) for 20 Recast Cores; results can be saved or discarded. Locking slots you want to keep costs Protection Cores, escalating steeply: 20 / 40 / 100 / 170 / 250 for 1–5 locks.
It's the drone recast loop re-skinned — reroll, preview, lock the keepers, pay more for each protected slot — pointed at a 4-slot SSR endgame: four Orange passives at level 10 each.
Levels, stars and passives grow one Sentinel. Sentinel Enhancement grows all of them at once — and almost none of it is bought on the Sentinel screen. It is an ordinary fleet attribute (guardUp, attrs_base #133) whose in-game tooltip reads, in full:
The tooltip covers Sentinel damage and healing. Fixed attribute buffs are separate: Swiftplume still contributes +30 percentage points of Burning Enhancement when Sentinel Enhancement rises. It also does not rewrite the flat HP/ATK contributed through #tgt70, skill cooldowns, or stack limits. Read the individual passive explanation before treating this stat as a universal multiplier.
Sentinel healing skills explicitly restrict their healing to PvP; Enhancement does not remove that restriction. Each Sentinel also carries pveRate: 1.875 and pvpRate: 0.025, whose values are 75× apart. Their precise position in the runtime formula is unverified, so that ratio alone is not a prediction that a passive produces 75× more final damage in campaign.
Sentinel Decrease (guardDown, #134) is the named counter for enemy Sentinel damage and healing. Its practical value requires an enemy Sentinel output to reduce. The exact Enhancement-versus-Decrease expression, including any conversion or floor, is not supplied by the stat or Combat Power formulas.
guardUp#fleet / guardDown#fleet — a single fleet-wide pool. Sources are written into the same key from six different systems and read once, so they are additive, not multiplicative: a 26.4% Mech Spirit line and an 18% Super Tech line make 44.4%, not 49.2%.guardUp bonuses come from the six routes below. The October snapshot adds two Enhancement and two Decrease Super Technologies. These are configured sources; the data does not establish when all of them become available.Two of the 18 passive families are Sentinel Enhancement and Sentinel Decrease. In the random-recast system, a slot rolls a family and a colour together, so the odds of a given slot landing Orange Sentinel Enhancement are 9% ÷ 18 = 0.5%, and the level within that colour is a second lottery. The newly described selection feature offers a different route; its prices and availability are not established here.
Eight of the 45 Mech Spirits carry Sentinel Enhancement as one of the four permanent attribute lines they grow from Lv 1→60, and eight more carry the Decrease mirror. Because all owned Spirits' bonuses are active simultaneously — this is a stat checklist, not a party-picker — these do not compete: collect all eight and you hold every line at once, for 117.6% total at Lv 60 across the board. That is by far the largest permanent block, and it is the reason Sentinel Enhancement is really a Mech Spirit stat wearing a Sentinel name.
One Mech Spirit Combination pays out in Sentinel Enhancement on top of the individual lines — 22 levels, gated on every member's Spirit level:
Separate from the attribute lines, three Spirits' battle skills hand out Sentinel Enhancement during the fight itself, and all three last until battle ends — they ramp rather than expire. All values below are at skill Lv 13 (the cap, reached at Spirit Lv 60):
Timesand Tech (unlocked by owning 10 max-level Camp Technologies) runs to Lv 180, with a Sentinel Enhancement milestone every 20 levels — +2 percentage points each, +18% at the cap. Mirage Realm Tech and Magma Core Tech have the same Enhancement ladder, bringing the configured technology subtotal to +54%. Magma Core Tech receives its name in the November downloaded data; live availability remains server-controlled.
Bionic Tech (unlocked by completing 50 daily missions) supplies the Sentinel Decrease mirror. Cosmic Tech and Froststar Tech each add another +18% Decrease. All six ladders are shown below.
Three Pilots grant Sentinel Enhancement. Kerron Joy uses an accumulator on research spending. Kerron Joy's stage-up counter does not read a level, a star or a formation slot — it reads how much research material you have consumed, and every 300 units advances one stage, eleven stages deep. Nothing about it requires the Pilot to be deployed. Cinder now adds +0.75% for every 30 successful Collection Center transports, while the new SSR Rosalind adds +1% for every 5 Bombardings completed. Their Tactics caps are +8.25% and +11%, bringing the three Enhancement accumulators to +27.5%. Yuelo now supplies the Decrease mirror through Pilot recruits.
St Rena, the SSR Pilot #6006 introduced as an unnamed definition in October, gains her name and skill text in the November downloaded data. She supplies +1% Sentinel Decrease per five Trial Against Foes completions, capped at +11%. She remains outside the standard recruit pools; her live availability is not established by the supplied files.
camp_tech_lv, i.e. ordinary Camp Technology. Fina Star's is the Supercomputing Chip, the research-point item for Super Tech. So the Decrease side double-dips — levelling Bionic Tech pays Sentinel Decrease twice, once from its own milestones and again through Fina Star — while the Enhancement side does not: Timesand Tech and Kerron Joy are funded from two separate currencies.Sentinel Enhancement is scored as an offensive stat, weighted against your three attack channels; Sentinel Decrease is scored as a defensive one, weighted against max HP. Both are clamped, so the score stops responding once your attack (or HP) leaves the band:
guardUp#fleet; the elemental attributes separately include advantage#fleet. Both Sentinel stats carry isDecay: 1, but no usable decay expression is provided here. That flag alone cannot establish the runtime curve or where it is evaluated.Stacking every configured permanent source at maximum gives before a single Sentinel passive is rolled. That requires eight Mech Spirits at Lv 60, a complete Overload Protocol, all three Enhancement technologies at Lv 180, and Kerron Joy, Cinder and Rosalind fully accumulated. This is a configured upper bound; live availability for the new technologies and every contributing Pilot is unconfirmed.
One Orange L10 slot contributes +75%, slightly more than four times Timesand Tech's +18% in raw stat points. A recast costs 20 Recast Cores before lock costs, and the advertised Orange-specific-family rate is 0.5% per slot. That chance does not guarantee a particular passive level or establish that four duplicate families are allowed. The total below is a conditional sum of configured sources, not a verified attainable loadout or final-output multiplier.
Everything runs on Sentinel Chips. Path one is the familiar refresh shop (same engine as drones): 10 chips per refresh, 4 free refreshes daily, buy what appears at list price (30 chips for a G up to 1,000 for an SSR). The published odds are the harshest in the game: SSR 0.2% — with a guaranteed R-or-better every 20 refreshes, and the auto-refresh robot ("Stop when SSR appears") happy to drain your stack.
Path two is the Sentinel Supply chest: 1 chip per draw. Only 20% of draws yield an actual unit; the other 80% pay out chest-shards that assemble into selectable chests. Every 100 supplies guarantees SR-or-SSR value, and the very first supply hands you a scripted SR-band unit to set the hook.
Duplicates also assemble from shards (20 → 100 by grade), and the hangar itself is monetized: 10 bag slots free, expandable to 50 at 500 Diamonds per slot.
The Sentinel Chip lane mirrors the drone store beat for beat: a free weekly trickle, then five escalating weekly packs. Chips also drip from Hypercoin and Wormhole-Badge exchange shops and daily-mission chests — but the arithmetic below is why the packs exist.
guard_main (roster, rates, passive roll odds), guard_level (750 rows), guard_star (205), guard_passive_skill(+_level) (54 + 540), guard_fetters, guard_unlock_skill, guard_type, guard_piece, game_base_param (guard_* economy constants), Supply chest via open_chest_config 109/112 → reward_new 223/1021–1023, packs from cfg_store.attrs_base #133–134 and attrs_combo #203–204 (definitions), power_calculation #60–61 (CP weights, isDecay), guard_passive_skill_level levelGroup 15/16, camp_tech + camp_high_tech_lv 999002/999003/999011/999012/999015/999016 (six Sentinel-stat technologies, 181 rows each), aviator_base (six contributing Pilots) + aviator_count (stage counters; the currency each counter reads was traced through camp_tech_lv vs camp_tech type-5 expItem), intelligence_level (Spirit attribute lines), intelligence_fetters(+_level) group 7, battle_buff (in-combat ladders). The claim that these six are the complete source list is a search for guardUp/guardDown across all 288 game data tables.guard_ui_*, guard_skill_name/desc_*, evolution/recast/reclaim notices).skill.json (105 guard rows = 21 × 5 levels: show_value1, skillCD, skillDamRate, type, buffs) joined through guard_unlock_skill; the riders resolved by walking each buffs chain into battle_buff (attribute writes are cmdType 1, enemy-side when camp = 1; Glimshade's lifesteal is cmdType 53 gated on EnemyInBurnStatus>0). DPS is skillDamRate ÷ skillCD at Lv 5; uptime is the buff's 2.1s duration over the cooldown.skill.json. guard_guide looks like a build-recommendation table but ships as a stub: all 22 rows carry the same three passives and the same four attribute keys, and none of its guide_title_* strings are translated, so it is not used as a source here.