# Fortress Defense research audit

Snapshot: 2026-09-26; reviewed 2026-10-03. Exact client version unconfirmed. These notes document the extracted client, not a live-server test.


---

# Fortress Defense defenders: runtime audit


Refreshed local data and shipped managed runtime recovered in September 2026; client version not independently established.


The 52 hero rows represent eight deployable archetypes at six levels, plus four hidden Quantum helpers. Raw area is a shape code. Actual cells, extracted from fresh bundle prefab localCoords, are 1, 3, 4, 2, 2, 1, 3 and 1 for types 1–8.


## Scope and calculation rules


- Nominal direct DPS is sustained damage to one stationary, sufficiently durable target in range, with every counted projectile landing, at ordinary battle speed and no held-energy boost. It averages the full rotating attack cycle.

- No talents, attack auras, vulnerability, shields, damage reduction, heavy hits or critical outcomes are included. The raw hero crit field is not imported into the attributes used by the recovered damage path.

- DPS tables use real-number coefficients before per-hit integer truncation and fixed-step timing. Actual damage can be slightly lower due to rounding, target acquisition, bullet travel, timer quantization or overkill.

- Explosive includes the primary target’s strong-shot self-splash. Quantum excludes hidden-field damage. Expulsion counts one central projectile. Fission counts one strip on the primary target even at level 5 or 6.

- Occupied cells are read from the actual fresh-OBB UI block prefab localCoords, not fortress_hero.area, which is a shape code.

- Current total attack = int(base attack × max(1 + summed attack modifiers, 0)). Ordinary hit = int(total attack × skill coefficient × (1 + source damage increase) × (1 + target dmgRed)). Positive dmgRed in this mode increases damage taken.

- Cooldown = base cooldown / max(1 + atkSpeedP, 0.1) / max(1 + atkSpeedPP, 0.1). Held energy doubles attack-timer progress for deployed types 2–8; Crystal and hidden Quantum helpers are excluded. The same multiplier also ages the affected unit’s buffs faster.


## Wish Crystal


A one-cell Reforge Core generator with the highest base HP and aggro values. Its production is independent of normal attack-speed and held-energy acceleration.


Type 1; 1 occupied cells; raw area code 1; footprint [{'x': 0, 'y': 0}].


| Level | ID | ATK | HP | CD seconds | Raw crit | Aggro | Nominal direct DPS | DPS/cell | Lv1 copies |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 101 | 0 | 240 | 20 | None | 500 | 0.0 | 0.0 | 1 |
| 2 | 102 | 0 | 360 | 18 | None | 600 | 0.0 | 0.0 | 2 |
| 3 | 103 | 0 | 480 | 16 | None | 700 | 0.0 | 0.0 | 4 |
| 4 | 104 | 0 | 600 | 14 | None | 800 | 0.0 | 0.0 | 8 |
| 5 | 105 | 0 | 720 | 12 | None | 900 | 0.0 | 0.0 | 16 |
| 6 | 106 | 0 | 840 | 10 | None | 1000 | 0.0 | 0.0 | 32 |


### Skill and production


Skill IDs 10101–10106 self-target and add one Reforge Core (item 501) per activation. Cooldowns by level are 20, 18, 16, 14, 12 and 10 seconds; nominal production is 3, 3.33, 3.75, 4.29, 5 and 6 cores per minute of battle time.

HeroUnit selects itself, so an enemy is not required for production. BattleUnit.GetFinalAtkSpeed ignores both attack-speed attribute groups for this unit, and holding the energy ball does not double its timer.


### Worked example and merge choice


At level 1, one Crystal makes 60/20 = 3 cores per minute. Two make 6 per minute; merging them into one level 2 makes 60/18 = 3.33 per minute while freeing one cell. Production retained is 55.56%, and combined HP falls from 480 to 360. Merge when space or a later level is worth that immediate loss.

At level 6 it has 840 base HP in one cell and authored aggro 1000. These values support a durable utility role, but the runtime follows a supplied target queue; a front-row placement does not guarantee it will absorb every enemy attack.


### Merge retention versus two previous-level pieces


| Merge | Cells freed | Direct DPS retained | HP retained |
|---|---|---|---|
| 1→2 | 1 | not a damage unit | 75.00% |
| 2→3 | 1 | not a damage unit | 66.67% |
| 3→4 | 1 | not a damage unit | 62.50% |
| 4→5 | 1 | not a damage unit | 60.00% |
| 5→6 | 1 | not a damage unit | 58.33% |


## Stasis Missile


A three-cell control unit. One 100% ATK missile every 1.5 seconds combines damage with a slow, then unlocks freeze and a surrounding slow.


Type 2; 3 occupied cells; raw area code 3; footprint [{'x': 1, 'y': 0}, {'x': 0, 'y': 1}, {'x': 1, 'y': 1}].


| Level | ID | ATK | HP | CD seconds | Raw crit | Aggro | Nominal direct DPS | DPS/cell | Lv1 copies |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 201 | 150 | 160 | 1.5 | None | 375 | 100.0 | 33.333333 | 1 |
| 2 | 202 | 300 | 240 | 1.5 | None | 425 | 200.0 | 66.666667 | 2 |
| 3 | 203 | 390 | 320 | 1.5 | None | 475 | 260.0 | 86.666667 | 4 |
| 4 | 204 | 520 | 400 | 1.5 | None | 525 | 346.666667 | 115.555556 | 8 |
| 5 | 205 | 840 | 480 | 1.5 | None | 575 | 560.0 | 186.666667 | 16 |
| 6 | 206 | 1680 | 560 | 1.5 | None | 625 | 1120.0 | 373.333333 | 32 |


### Projectile and control


One projectile per 1.5-second activation deals 1.0 × current total ATK. Release range is 8 authored logic units, projectile speed 10, lifetime 3 seconds and hit radius 0.2. The bullet ends on its first enemy collision. Range is multiplied by the battle logic scale.

At every level, slow buff 10201 applies −30% move speed for 2 seconds. The authored proc rate is 50%. The recovered roll compares integer Range(0,100) with <= rate × 100; with the usual exclusive upper bound this is 51 successful outcomes, not 50.

Levels 3–6 add a separate nominal 15% freeze roll for buff 10202, lasting 1 second. The slow attempt runs first. Talent 22 adds 10 percentage points to the freeze threshold if the selected target has slow buff 10201, including a slow just applied by this same projectile. Levels 5–6 additionally slow other living enemies within radius 1 for 1 second after a successful freeze; this secondary effect excludes the selected primary target.

Slow and freeze use replacement rules: repeat applications refresh the same buff ID rather than multiplying slow strength. Talent 12 adds 0.4 second per rank to buff 10201, up to 3.2 seconds at rank 3. The short surrounding slow 10203 has no duration-talent link.


### Worked example


A level 6 missile has 1680 ATK. With no damage modifiers, each successful ordinary hit deals 1680 damage and nominal direct DPS is 1680/1.5 = 1120, or 373.33 per occupied cell. The control benefit is additional to that number; it cannot be converted into a universal damage multiplier.


### Merge retention versus two previous-level pieces


| Merge | Cells freed | Direct DPS retained | HP retained |
|---|---|---|---|
| 1→2 | 3 | 100.00% | 75.00% |
| 2→3 | 3 | 65.00% | 66.67% |
| 3→4 | 3 | 66.67% | 62.50% |
| 4→5 | 3 | 80.77% | 60.00% |
| 5→6 | 3 | 100.00% | 58.33% |


Freeze buff 10202 is checked by the movement agent, not by MonsterUpdate’s attack branch. It stops movement, but an enemy already in attack range can still attack in the recovered runtime.

## Quantum Missile


A four-cell 2×2 unit whose 100% ATK circular hit can create one hidden damage field per enemy hit from level 3. Clustering matters far more than its low direct DPS per cell.


Type 3; 4 occupied cells; raw area code 2; footprint [{'x': 0, 'y': 0}, {'x': 1, 'y': 0}, {'x': 0, 'y': 1}, {'x': 1, 'y': 1}].


| Level | ID | ATK | HP | CD seconds | Raw crit | Aggro | Nominal direct DPS | DPS/cell | Lv1 copies |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 301 | 120 | 170 | 2 | None | 380 | 60.0 | 15.0 | 1 |
| 2 | 302 | 240 | 255 | 2 | None | 430 | 120.0 | 30.0 | 2 |
| 3 | 303 | 270 | 340 | 2 | None | 480 | 135.0 | 33.75 | 4 |
| 4 | 304 | 360 | 425 | 2 | None | 530 | 180.0 | 45.0 | 8 |
| 5 | 305 | 480 | 510 | 2 | None | 580 | 240.0 | 60.0 | 16 |
| 6 | 306 | 960 | 595 | 2 | None | 630 | 480.0 | 120.0 | 32 |


### Circular strike and hidden helpers


Every 2 seconds, active skill 10301–10306 deals 1.0 × total ATK to every enemy within radius 1.5 of the chosen strike position. Release range is 8 authored units. This is an active circular hit rather than a traveling projectile.

From level 3, OnAoeTarget calls its after-hit effect once for each enemy processed in the circle. Each call creates a hidden type-99 helper at the same strike position: IDs 9903, 9904, 9905 or 9906. There is no one-helper-per-cast cap in that spawn chain. Hitting several clustered enemies can therefore create overlapping fields.

Each helper attacks all living enemies inside radius 1.5 for 0.48 × its own total ATK per pulse. Its ATK is loaded from its own row: 270, 360, 480 or 960. The summoner argument is not used to copy the parent weapon’s attack or buffs. Helper cooldown is 1 second at levels 3–4 and 0.5 second at levels 5–6; base lifetime is 2 seconds.

Lifetime is reduced before the pulse check, and a helper dies when it reaches zero. The pooled unit’s attack timer is not reset by the inspected SetData, OnSpawn or OnRecycle methods. Consequently a guaranteed exact pulse count per field cannot be read simply as lifetime/cooldown. Tables below intentionally count the direct strike only.

Talent 13 adds 8% field damage per rank, up to 24%; talent 23 extends field lifetime by 1 second. Helpers explicitly skip other talent applications, and neither ordinary speed talents nor held energy accelerate their pulse timers. Held energy can still create new fields faster by accelerating the parent missile.


### Worked example with explicit hit and pulse counts


Assume a level 6 cast hits three enemies, all three remain inside all three fields for three pulses, and there are no modifiers or critical outcomes. Direct damage is 3 × 960 = 2880. Each pulse deals int(960 × 0.48) = 460 damage. Field damage is 3 helpers × 3 targets × 3 pulses × 460 = 12420. Total damage across those enemies is 15300. Three pulses is an explicit scenario assumption, not a guaranteed lifetime count.

For one enemy without counting fields, level 6 direct DPS is 960/2 = 480, or 120 per cell. For k helpers and n landed pulses on one surviving target, add k × n × int(0.48 × helper ATK) per cast. This makes Quantum a cluster specialist rather than a universal low-value unit.


### Merge retention versus two previous-level pieces


| Merge | Cells freed | Direct DPS retained | HP retained |
|---|---|---|---|
| 1→2 | 4 | 100.00% | 75.00% |
| 2→3 | 4 | 56.25% | 66.67% |
| 3→4 | 4 | 66.67% | 62.50% |
| 4→5 | 4 | 66.67% | 60.00% |
| 5→6 | 4 | 100.00% | 58.33% |


## Explosive Missile


A two-cell horizontal damage unit. From level 3, the stronger shot also splashes its surviving primary target, making it substantially stronger than a tooltip-only 180% calculation.


Type 4; 2 occupied cells; raw area code 6; footprint [{'x': 0, 'y': 0}, {'x': 1, 'y': 0}].


| Level | ID | ATK | HP | CD seconds | Raw crit | Aggro | Nominal direct DPS | DPS/cell | Lv1 copies |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 401 | 125 | 160 | 1.5 | 0.1 | 385 | 100.0 | 50.0 | 1 |
| 2 | 402 | 250 | 240 | 1.5 | 0.1 | 435 | 200.0 | 100.0 | 2 |
| 3 | 403 | 270 | 320 | 1.5 | 0.1 | 485 | 312.0 | 156.0 | 4 |
| 4 | 404 | 360 | 400 | 1.5 | 0.1 | 535 | 416.0 | 208.0 | 8 |
| 5 | 405 | 666 | 480 | 1.5 | 0.1 | 585 | 888.0 | 444.0 | 16 |
| 6 | 406 | 1200 | 560 | 1.5 | 0.1 | 635 | 1600.0 | 800.0 | 32 |


### Rotating shots and primary-target splash


Levels 1–2 fire a 1.2 × ATK shot every 1.5 seconds. At levels 3–4, the executor cycles normal, normal, strong; at levels 5–6 it alternates normal, strong. Duplicate skill IDs represent repeated turns in this cycle, not simultaneous shots.

Normal shot 10403 deals 1.2 × ATK. Strong shot 10407 first deals 1.8 × ATK to the collision target, then independently deals 1.0 × ATK to every living enemy within radius 1 of that target. The splash loop does not exclude the primary target, so a primary that survives the first hit receives 2.8 × ATK in total. A primary killed by the first hit is skipped by the living-target check.

The strong splash uses its authored radius directly, while release range uses the battle scale factor. Projectile speed is 10, lifetime 3 seconds and hit radius 0.25. These are the measured config/runtime conventions, not a conversion to map cells.

Raw hero rows contain crit = 0.10, but GetHeroUnitAttr does not copy that field into the battle attributes used by damage. The base comparison therefore does not award a 10% critical chance. Pulse Blast II separately adds heavy-hit weight to Explosive Missile; the selectable talent set does not establish a general critical-chance upgrade.


### Worked example and progression


At level 6, ATK is 1200. Normal damage is 1440. A strong shot deals 2160 direct plus 1200 self-splash to a surviving primary, for 3360. Over two 1.5-second activations: (1440 + 3360)/3 = 1600 nominal single-target DPS, or 800 per cell. Nearby enemies receive another 1200 per strong shot.

At levels 3–4, the long-run primary coefficient is (1.2 + 1.2 + 2.8)/3 = 1.7333 ATK per activation. At levels 5–6 it is (1.2 + 2.8)/2 = 2 ATK. These values assume the same sufficiently durable target remains available and survives the initial strong hit.


### Merge retention versus two previous-level pieces


| Merge | Cells freed | Direct DPS retained | HP retained |
|---|---|---|---|
| 1→2 | 2 | 100.00% | 75.00% |
| 2→3 | 2 | 78.00% | 66.67% |
| 3→4 | 2 | 66.67% | 62.50% |
| 4→5 | 2 | 106.73% | 60.00% |
| 5→6 | 2 | 90.09% | 58.33% |


Pulse Blast I adds 12% attack per rank to every Explosive damage component, up to 36%. Pulse Blast II adds 0.05 heavy-hit weight and uses a 4.05× selected multiplier. With no other critical or heavy-hit weights, its expected multiplier is 0.95 + 0.05 × 4.05 = 1.1525. Direct and splash roll separately. A level-6 Explosive with rank-3 I and II therefore has approximately 1600 × 1.36 × 1.1525 = 2507.84 nominal expected primary DPS before truncation, or 1253.92 per cell.

## Expulsion Missile


A two-cell vertical spread attacker. It sends three 100% ATK projectiles, but side hits and fragment hits depend on geometry. The recovered split implementation does not support every advertised recursive split.


Type 5; 2 occupied cells; raw area code 5; footprint [{'x': 0, 'y': 0}, {'x': 0, 'y': 1}].


| Level | ID | ATK | HP | CD seconds | Raw crit | Aggro | Nominal direct DPS | DPS/cell | Lv1 copies |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 501 | 80 | 180 | 1.8 | None | 390 | 44.444444 | 22.222222 | 1 |
| 2 | 502 | 160 | 270 | 1.8 | None | 440 | 88.888889 | 44.444444 | 2 |
| 3 | 503 | 210 | 360 | 1.8 | None | 490 | 116.666667 | 58.333333 | 4 |
| 4 | 504 | 280 | 440 | 1.8 | None | 540 | 155.555556 | 77.777778 | 8 |
| 5 | 505 | 360 | 520 | 1.8 | None | 590 | 200.0 | 100.0 | 16 |
| 6 | 506 | 720 | 600 | 1.8 | None | 640 | 400.0 | 200.0 | 32 |


### Three lanes and fragmentation limits


Each 1.8-second activation spawns a center projectile and two side projectiles at +10° and −10°. Each deals 1.0 × ATK, has speed 10, lifetime 3 seconds and hit radius 0.5. The center receives a target reference; the side bullets are spawned with no target reference. Each ends at its first collision.

From level 3, the center projectile’s hit can spawn two fragments at ±15° around its owner-to-target direction. Each fragment deals 0.5 × owner ATK, travels at speed 10, lives 3 seconds and has hit radius 0.25. Fragments spawn forward of the collision position by three parent hit radii.

The inspected SplitBullet immediately returns when its stored target is null. Side bullets and child fragments are spawned with null targets, and OnHitTarget does not replace that stored target with the collision victim. Thus the recovered ordinary path allows the center to split once, but not side-bullet splitting or further child splitting.

Levels 5–6 author fragment 10510 with a 50% chance to create 10509. SplitBullet does not read that rate, and the null-target return prevents that recursive split in the inspected spawn path. This is a client implementation discrepancy; do not count an extra expected 50% wave in a DPS estimate.


### Worked example


At level 6, ATK is 720. One central hit per activation gives 720/1.8 = 400 nominal DPS, or 200 per cell. If all three initial projectiles hit something, the aggregate is 3 × 720/1.8 = 1200 DPS. If the two permitted center fragments also hit, add 2 × 360/1.8 = 400, giving 1600 aggregate DPS across the affected enemies.

The 1600 figure is a five-hit coverage scenario, not guaranteed damage against one target. Narrow or distant targets can miss the side lanes, and forward-spawned fragments need enemies in their paths.


### Merge retention versus two previous-level pieces


| Merge | Cells freed | Direct DPS retained | HP retained |
|---|---|---|---|
| 1→2 | 2 | 100.00% | 75.00% |
| 2→3 | 2 | 65.62% | 66.67% |
| 3→4 | 2 | 66.67% | 61.11% |
| 4→5 | 2 | 64.29% | 59.09% |
| 5→6 | 2 | 100.00% | 57.69% |


## Focused Cannon


A one-cell single-target attacker and an efficient way to fill remaining space. Its configured 5% crit is not imported by the recovered battle-stat path, so the baseline uses ordinary hits.


Type 6; 1 occupied cells; raw area code 1; footprint [{'x': 0, 'y': 0}].


| Level | ID | ATK | HP | CD seconds | Raw crit | Aggro | Nominal direct DPS | DPS/cell | Lv1 copies |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 601 | 60 | 100 | 1.2 | 0.05 | 400 | 50.0 | 50.0 | 1 |
| 2 | 602 | 120 | 150 | 1.2 | 0.05 | 450 | 100.0 | 100.0 | 2 |
| 3 | 603 | 180 | 200 | 1.2 | 0.05 | 500 | 150.0 | 150.0 | 4 |
| 4 | 604 | 240 | 250 | 1.2 | 0.05 | 550 | 200.0 | 200.0 | 8 |
| 5 | 605 | 480 | 300 | 1.2 | 0.05 | 600 | 400.0 | 400.0 | 16 |
| 6 | 606 | 960 | 350 | 1.2 | 0.05 | 650 | 800.0 | 800.0 | 32 |


### Compact direct fire


One 1.0 × ATK projectile every 1.2 seconds. Release range is 8 authored units, speed 10, lifetime 3 seconds and hit radius 0.2. It ends on its first collision. This simple baseline makes comparisons and one-cell packing straightforward.

The raw hero row contains crit = 0.05. The recovered GetHeroUnitAttr does not import that value into the attributes read by CalcJudgeRate, so the baseline uses crit 0. Pulse Blast II is specific to Explosive Missile, so it does not add heavy-hit weight to Focused Cannon.


### Worked example and weak-point caveat


At level 6, 960 ATK gives 960/1.2 = 800 nominal DPS in one cell. With an independently verified +10% ATK aura and no other modifiers, total ATK is int(960 × 1.10) = 1056 and nominal DPS is 1056/1.2 = 880.

Weak Point I authors +50% damage against targets below its 30% HP threshold; Weak Point II authors an execution attempt against eligible normal monsters. The II branch suppresses I rather than combining with it. The condition reads target.attrSet.hpMax instead of target.FinalAttr.hpMax; the inspected initialization does not populate that backing field. Their practical trigger rate is unresolved, so neither is included in the baseline or used to claim a guaranteed execution build.


### Merge retention versus two previous-level pieces


| Merge | Cells freed | Direct DPS retained | HP retained |
|---|---|---|---|
| 1→2 | 1 | 100.00% | 75.00% |
| 2→3 | 1 | 75.00% | 66.67% |
| 3→4 | 1 | 66.67% | 62.50% |
| 4→5 | 1 | 100.00% | 60.00% |
| 5→6 | 1 | 100.00% | 58.33% |


## Fission Laser


A three-cell L-shaped strip attacker. From level 3 it makes hit enemies take 10% more incoming damage; level 5 adds a second offset strip for coverage.


Type 7; 3 occupied cells; raw area code 4; footprint [{'x': 0, 'y': 0}, {'x': 1, 'y': 0}, {'x': 0, 'y': 1}].


| Level | ID | ATK | HP | CD seconds | Raw crit | Aggro | Nominal direct DPS | DPS/cell | Lv1 copies |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 701 | 100 | 160 | 2 | None | 395 | 50.0 | 16.666667 | 1 |
| 2 | 702 | 200 | 240 | 2 | None | 445 | 100.0 | 33.333333 | 2 |
| 3 | 703 | 270 | 320 | 2 | None | 495 | 135.0 | 45.0 | 4 |
| 4 | 704 | 360 | 400 | 2 | None | 545 | 180.0 | 60.0 | 8 |
| 5 | 705 | 420 | 480 | 2 | None | 595 | 210.0 | 70.0 | 16 |
| 6 | 706 | 840 | 560 | 2 | None | 645 | 420.0 | 140.0 | 32 |


### Strip geometry and vulnerability


Active skills 10701–10706 hit enemies in a vertical rectangle of width 1.5 and length 5, dealing 1.0 × total ATK per strip. The rectangle extends forward from its anchor, rather than five units centered around it. Release range is 8. The third target parameter, 27, is a talent ID, not another distance.

The first strip’s horizontal center follows the selected target; its starting y is the target’s y, capped at the attacker’s y + 1.5. Levels 1–4 emit one strip. Levels 5–6 emit two strips immediately in the same activation, with the second centered 1–2 units horizontally toward the board’s center from the first.

At base width 1.5, the original stationary target lies outside the offset second strip. Therefore releaseCount 2 doubles coverage opportunities, not guaranteed primary-target damage. Other enemies in an overlap can be struck by both strips. Talent 27 changes the size to width 2 and length 6.

From level 3, each hit applies buff 10703 for 5 seconds. Its +0.10 dmgRed is multiplied as 1 + dmgRed by the common damage formula, making the victim take 10% more damage from all sources. Same-ID reapplication refreshes duration rather than adding another 10%. Talent 17 adds 0.04 per rank, reaching 22% vulnerability at rank 3. The hit that first applies it is calculated before the buff is added.


### Worked example and support value


At level 6, one primary strip deals 840 and nominal raw primary DPS is 840/2 = 420, or 140 per cell. Once the enemy remains debuffed, that strip’s own output becomes 462 DPS before rounding, but the rest of the formation also benefits. Do not add a second guaranteed 840 hit to this primary calculation.

If the rest of a formation supplies 4000 raw DPS to enemies kept marked, base vulnerability adds about 400 DPS; rank-3 talent 17 adds about 880. This is why one level-3-or-higher Fission Laser can be more valuable than its direct DPS table suggests. Additional lasers improve coverage and uptime, but their identical debuff does not stack strength.


### Merge retention versus two previous-level pieces


| Merge | Cells freed | Direct DPS retained | HP retained |
|---|---|---|---|
| 1→2 | 3 | 100.00% | 75.00% |
| 2→3 | 3 | 67.50% | 66.67% |
| 3→4 | 3 | 66.67% | 62.50% |
| 4→5 | 3 | 58.33% | 60.00% |
| 5→6 | 3 | 100.00% | 58.33% |


## Energy Barrier


A one-cell support unit that shields orthogonally adjacent allies every 3 seconds. From level 3 its configured survival aura adds 10% ATK, and level 6 adds 10% kill-energy gain.


Type 8; 1 occupied cells; raw area code 1; footprint [{'x': 0, 'y': 0}].


| Level | ID | ATK | HP | CD seconds | Raw crit | Aggro | Nominal direct DPS | DPS/cell | Lv1 copies |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 801 | 0 | 100 | 3 | None | 430 | 0.0 | 0.0 | 1 |
| 2 | 802 | 0 | 150 | 3 | None | 480 | 0.0 | 0.0 | 2 |
| 3 | 803 | 0 | 200 | 3 | None | 530 | 0.0 | 0.0 | 4 |
| 4 | 804 | 0 | 250 | 3 | None | 580 | 0.0 | 0.0 | 8 |
| 5 | 805 | 0 | 300 | 3 | None | 630 | 0.0 | 0.0 | 16 |
| 6 | 806 | 0 | 350 | 3 | None | 680 | 0.0 | 0.0 | 32 |


### Adjacent support


Skill 10801–10806 has no damage coefficient. It applies shield buff 10801 every 3 seconds to living orthogonally adjacent allies, excluding the caster. Adjacency is based on occupied shapes, so a large neighboring defender can share an edge with the Barrier.

Each shield application adds int(recipient maximum HP × 0.03 × (1 + provider shield-strength bonus)) to the stored shield, capped at the recipient’s maximum HP. The ShieldAdd rule creates an instance while replacing hpShieldP at 0.03; it does not raise the percentage with each application. The 999 duration is a buff timer, and damage can consume the shield sooner.

Levels 3–6 author survival buff 10803, +10% ATK to adjacent allies; level 6 adds buff 10805, +10% energy gained by their kills. Multiple providers use additive aura instances. Their removal is tied to provider death. One startup route applies these to neighbors correctly; the inspected AddOnceBuff path reverses provider and recipient, so refreshing or duplicate application must be treated cautiously.


### Worked example and placement


A level 6 Focused Cannon has 350 HP. An unmodified 3% shield application adds int(350 × 0.03) = 10 shield, up to a stored shield cap of 350. An unmodified +10% attack aura increases its nominal DPS from 800 to 880.

Place a Barrier where it shares edges with several valuable attackers, while keeping it alive. Its own raw damage is zero and its HP is only 350 at level 6, so it is primarily support rather than a guaranteed tank. Holding energy accelerates its timer along with other non-Crystal deployed units.


### Merge retention versus two previous-level pieces


| Merge | Cells freed | Direct DPS retained | HP retained |
|---|---|---|---|
| 1→2 | 1 | not a damage unit | 75.00% |
| 2→3 | 1 | not a damage unit | 66.67% |
| 3→4 | 1 | not a damage unit | 62.50% |
| 4→5 | 1 | not a damage unit | 60.00% |
| 5→6 | 1 | not a damage unit | 58.33% |


## Level 6 primary-target comparison


| Unit | Cells | Raw nominal DPS | DPS/cell | What is excluded |
|---|---|---|---|---|
| Wish Crystal | 1 | 0.0 | 0.0 | Production |
| Stasis Missile | 3 | 1120.0 | 373.333333 | Control value |
| Quantum Missile | 4 | 480.0 | 120.0 | All field pulses |
| Explosive Missile | 2 | 1600.0 | 800.0 | Splash to other enemies |
| Expulsion Missile | 2 | 400.0 | 200.0 | Side bullets and fragments |
| Focused Cannon | 1 | 800.0 | 800.0 | Conditional weak point |
| Fission Laser | 3 | 420.0 | 140.0 | Second strip and shared vulnerability |
| Energy Barrier | 1 | 0.0 | 0.0 | All support value |


## Verified merge rules


Two distinct pieces with the exact same hero definition ID can merge if neither is at maximum level. The target upgrades to the next row of its type; the other piece is consumed through the placement callback and sent as CostUid in the server payload. Different levels cannot merge in this path.

One level 6 represents 32 level-1 copies under ordinary binary merging. Every merge frees one copy of that archetype’s footprint, but does not preserve the two pieces’ combined HP or necessarily their damage. Per-archetype ratios compare one upgraded piece with two pieces of the previous level.

At level 1→2, every attacking archetype doubles ATK, so ordinary direct damage is preserved while occupied space halves. From level 2→3, several units sacrifice direct throughput for new effects: two level-2 Explosive Missiles deal 400 combined nominal DPS; one level-3 deals 312 but gains strong-shot splash. Two level-2 Focused Cannons deal 200; one level-3 deals 150.

At level 5→6, most attack rows double ATK and preserve direct damage while halving space. Explosive is an exception: two level-5 copies total 1776 nominal DPS; one level-6 totals 1600, retaining 90.09%. Crystal production always falls versus two lower-level copies. Barrier merging trades provider count and adjacency coverage for level thresholds.

No separate currency amount is present in the inspected local merge payload. This does not establish every server-side eligibility or event-economy rule. Upgrade items may provide an alternative progression route and are not priced as ordinary binary merges.


## Early levels and practical thresholds

At levels 1–2, Focused Cannon and Explosive Missile tie at 50 and then 100 nominal direct DPS per cell. Use Focused to fit single holes and Explosive where its horizontal pair fits; keep a Crystal when ongoing Reforge Core income matters. A level-1 Barrier has shielding but no attack aura, and Fission has no vulnerability until level 3.

The first merge preserves direct damage for every attacking type while halving its occupied space, but reduces the two pieces’ combined HP to roughly 75%. Further merges can lose direct throughput: level 2→3 Focused retains 75%, and Explosive retains 78% while gaining splash. Seek skill thresholds or free needed cells rather than merging automatically.

At level 3, Stasis gains freeze, Quantum gains fields, Explosive gains its strong shot, Fission gains vulnerability and Barrier gains its authored attack aura. These effects reward different enemy patterns; the direct-DPS table alone is not the whole choice.

## Compact sustained-damage core

For durable single targets and tight space, use Focused Cannons and Explosive Missiles as the main damage core. At level 6 they tie at 800 nominal direct DPS per cell; Focused fits individual cells, while Explosive also damages nearby enemies. At levels 3–5, Explosive has slightly higher direct DPS per cell under the surviving-primary assumption.

A concrete balanced package is two Explosive Missiles, two Focused Cannons, one level-3-or-higher Fission Laser, one level-3-or-higher Barrier and one Crystal. It uses 11 occupied cells: 4 + 2 + 3 + 1 + 1. At level 6, the six damage-core cells alone supply 4800 nominal raw primary DPS, before Fission damage or support. This is a utility-oriented package, not the maximum possible boss DPS in 11 cells. The shape arrangement must fit your unlocked board.

With a level-6 Fission Laser continuously marking the target, that package’s 5220 raw primary DPS becomes approximately 5742 at base 10% vulnerability, before rounding. Barrier attack buffs depend on actual adjacent coverage and the initialization caveat, so they are not silently added to this number.

For a stationary boss and no survival or economy constraint, 11 level-6 Focused Cannons, or an equivalent Focused/Explosive mixture that fits, supply 8800 nominal primary DPS in 11 cells. The balanced package sacrifices that ceiling for economy, shielding, strip coverage and vulnerability.

## Control and cluster adjustments

Add one Stasis Missile when enemies reach your formation too quickly; level 3 is the freeze threshold, and level 5 adds the nearby slow. Its three-cell cost buys control as well as damage.

Add Quantum Missile from level 3 when enemies arrive in dense groups and remain inside the circular fields. Its four-cell cost is hard to justify from direct DPS alone, but multiple enemies hit can create multiple overlapping helpers. Pairing a field with control can increase realized pulse hits.

Use Expulsion when broad lanes let the side shots and fragments land. Its five-hit aggregate scenario can match strong compact attackers, but its reliable one-lane output is much lower. Choose based on observed enemy geometry rather than the three-projectile tooltip alone.

## Support, survival and merge priorities

Fission level 3 unlocks shared vulnerability, but one copy is not automatically the best use of three cells. At level 6, three Focused Cannons provide 2400 raw primary DPS, while one Fission provides 420 and amplifies its marked target. With 10% vulnerability, the rest of the formation must supply more than 19380 DPS to that target for Fission to beat the three Cannons on pure single-target damage: 1.10 × (D + 420) > D + 2400. At 22% vulnerability the threshold is 8580. Multiple strip targets lower that opportunity cost; identical debuffs do not stack strength.

Place a Barrier where its edges touch several high-value units. A Crystal has unusually high HP for one cell and also produces economy, but aggro numbers do not by themselves prove front-row targeting: the recovered enemy selector follows a supplied hero queue. Keep the entire formation alive rather than assuming one designated tank protects everyone.

Merge aggressively when space prevents deploying stronger pieces, but compare the resulting skill threshold and direct-DPS retention. Keep multiple Crystals only when their extra production is worth the occupied space and survivability tradeoff. Avoid treating every merge as a damage upgrade.

## Bosses and support break-even

Against an isolated durable boss, prioritize high-level Focused and Explosive pieces, count one Expulsion lane and one Fission strip, and do not assume Quantum fields hit more than one target. Stasis can still be useful if its control actually affects that enemy, but a blanket boss-control immunity rule was not established here.

A level-6 Barrier using one cell can add at most 80 DPS to each adjacent unmodified level-6 Focused Cannon through a single +10% ATK aura. Four such neighbors give 320 DPS, less than the 800 DPS of using that cell for another Cannon. Its shielding, energy support, adjacency to stronger multi-cell units or survival benefit must justify the difference.

Against clustered waves, Explosive splash, Quantum’s per-enemy field spawning and Fission’s multi-target strips can overturn the isolated-boss order. Track how many targets each area actually hits; no universal highest-DPS setup follows from the raw roster alone.

## Runtime evidence index


All RVAs below are from recovered Extension_4.dll (Assembly-CSharp). Config IDs are preserved in the JSON level rows.


| Type / method | RVA | Verified use |
|---|---|---|
| HeroExtensions.CheckHeroesCanUpgrade | 0x114c33 | Exact-ID, distinct-piece, not-max merge eligibility |
| HeroBlockEntity.UpgradeLv | 0x114b58 | Advance to next definition of the same type |
| DropUtils.TryMergeWithTargetHero | 0x10f6b0 | Upgrade target, send CostUid and TargetUid, consume source |
| HeroPlacementHandler.ComputeHeroOccupiedCells | 0x10f0d4 | Enumerate prefab shape.localCoords |
| BlockComponentCreateFactory.GetBlockComponent | 0x112720 | Load hero uiPreview prefab |
| ActiveSkillExecutor.Execute / ExecuteOnce | 0x134098 / 0x134194 | Rotate skills; emit immediate strips at offset positions |
| BulletSkillExecutor.Execute / Fire | 0x1343a8 / 0x13445c | Rotate shot IDs; center and ±10° Expulsion shots |
| BattleFieldDataProvider.GetHeroUnitAttr | 0x138008 | Import base stats without raw crit |
| BattleUnit.GetTotalAtk | 0x13b986 | Apply atkP to FinalAttr.atk |
| BattleUnit.Step / HeroUpdate | 0x13bba8 / 0x13bca4 | Timer and energy doubling; one attack per elapsed cooldown |
| BattleUnit.HeroEagleTrapUpdate | 0x13bd90 | Lifetime check before pulse |
| BattleUnit.GetFinalAtkSpeed | 0x13c42c | Speed groups; Crystal/helper exclusion |
| BattleUnit.SetData / OnSpawn / OnRecycle | 0x13b824 / 0x13b95c / 0x13b9a4 | No attackTimer reset in inspected lifecycle |
| SkillLogic.Init / OnAoeTarget | 0x13a5ec / 0x13aa34 | Rectangle dimensions; circular hit and per-enemy effects |
| SkillLogic.SummonPlayer | 0x13aca7 | Spawn helper at strike position |
| UnitFactory.SummonHeroFromConfig | 0x13d6d8 | Helper loads own definition, not parent attack |
| SkillLogic.GetDamageValue / OnHitTarget | 0x13a704 / 0x13a9dc | Coefficient and modifiers before after-hit buff |
| SkillLogic.OnSkillToRectTarget / OnSkillToCross | 0x13af18 / 0x13afe0 | Strip collision and adjacent allied targets |
| Vector2f.PointInRect | 0x13e628 | Centered x, forward-only y rectangle |
| BulletLogic.GetDamageValue / GetExtendDamageValue | 0x13c778 / 0x13c8fc | Main and splash coefficients; conditional Weak Point branch |
| BulletLogic.OnHitTarget / ExtendHitTarget | 0x13cdd4 / 0x13ce44 | First collision, living-primary self-splash |
| BulletLogic.AddBuff / SplitBullet | 0x13cfc0 / 0x13d218 | Ordered integer proc rolls; null-target split gate |
| BulletLogic.Init / GetFirstEnemyOnSegment | 0x13c6c4 / 0x13cb5c | Stored target and collision routing |
| BattleFormula.CalcAtk / CalcSkillCd / CalcSkillDamage | 0x134d28 / 0x134d56 / 0x134e8a | Truncation, cadence and damage multiplication |


Fresh asset evidence: output/fortress-defenders-audit/prefab-shapes.json contains the bundle path, component path ID, prefab name and coordinate list for every deployable footprint. It was read directly with UnityPy.read_typetree from wf.obb, not the stale unpacked obb directory.


## Limitations


- These are recovered-client predictions from the local refreshed snapshot, not instrumented live-match measurements or a confirmed v1.7.820 ruleset.

- Field pulse count depends on fixed-step boundaries and retained pooled attack-timer phase. The field example explicitly assumes a count; no universal burn DPS is asserted.

- Expulsion recursive splitting, raw hero critical chance, Weak Point’s HP backing field and the alternate Barrier aura application path disagree with simple tooltip interpretations. They are surfaced rather than folded into optimistic damage totals.

- Enemy targeting uses a supplied queue, so authored aggro and board position alone cannot guarantee a frontline tank.

- Talent-specific effects are cross-referenced to docs_src/fortress-powerups-notes.json; direct damage comparisons do not monetize control, economy or shielding.


## Source files


- data/fortress_hero.json

- data/fortress_skill_active.json

- data/fortress_skill_bullet.json

- data/fortress_buff.json

- data/fortress_skill_talent.json

- data/_loc_en.json

- output/fortress-defenders-audit/

- output/fortress-rules-audit/formula.il.txt

- output/fortress-rules-audit/data.il.txt

Additional linked evidence: docs_src/fortress-powerups-notes.json; BaseBuff.OnAdd 0x134868 and BattleUnit.UpdateShield 0x13c378 confirm additive shield grants capped at maximum HP.


---

# Fortress Defense powerup and buff runtime audit

Refreshed local configuration and managed runtime recovered from the September 26 snapshot; build version not established.

This audit covers every row of `fortress_skill_talent` and `fortress_buff`: 20 selectable-powerup definitions, two enemy passives, and seven combat buffs. It follows the managed runtime in `Extension_4.dll`, using the current decrypted tables rather than the old unpacked asset directory. The structured companion is `docs_src/fortress-powerups-notes.json`.

## Verified findings that affect decisions

- Choices are remote-service results. Configured weights are not exact client-computed offer probabilities; the server’s filters are unavailable.
- Swift bypasses Wish Crystal and Quantum burn helpers. The actual helper summon path reads its own configured stats; it does not call the separately defined `OverrideHeroUnitAttr` method.
- Laser Beam I is a target vulnerability upgrade, affecting other turrets’ normal incoming damage as well as Fission. Reapplication refreshes a single mark.
- Weak Point II skips the entire Tier I damage branch. Both tests read `target.attrSet.hpMax`, which differs from the field used for ordinary maximum HP; the initialization discrepancy remains an explicit limitation.
- Area Protection II is retaliation based on the recipient’s attack. The runtime does not require shield to remain positive when the hit occurs.
- Freeze is checked by the movement agent only. A frozen enemy already in attack range can still attack.
- Stasis/execute/knockback use an inclusive integer-roll comparison. Nominal 15%, 25%, 10%, and 4% thresholds accept 16, 26, 11, and 5 outcomes of `Range(0,100)` respectively. Whack uses a different continuous weighted calculation.

## Selection and allocation evidence

`FortressServiceRmt.RandomTalent` (0x194130) sends the battle EXP level and receives talent IDs. `SelectTalent` (0x194270) sends selected ID and battle level; `OnSelectTalentSuccess` (0x194210) applies the returned ID. `FortressTalentSelectView.RefreshTalentItems` (0x23490c) displays the supplied array and does not generate local choices.

`FortressModel.UpdateSelectedTalent` (0x192794) increments the permanent rank dictionary, then broadcasts a delta of **one**. `BattleController.OnTalentChanged` (0x13678c) forwards that delta to all battle units. `BattleFieldDataProvider.UpdateTalent` (0x13838c) adds `value × delta`; new-unit initialization uses accumulated ranks. This is additive rank stacking, not repeated application of the total rank on every selection.

The same model method branches on type 2 for Supply Drop: add item 501 × 15, broadcast the item grant, and return without inserting it into the talent dictionary. No client-side max-rank enforcement was found in the selection path. A scan of the recovered client did not find gameplay reads of `FortressTalentDef.maxLv`, `needTalent`, `correlation`, `weight`, or `change`; those fields remain configured metadata rather than proven local offer logic.

## Formulas

For nonnegative ordinary attacks, `trunc` denotes the managed integer conversion after the float calculation. Rounding examples isolate neutral hit quality and no vulnerability unless stated.

```text
A = trunc(baseATK × max(1 + atkP, 0))
hit = trunc(A × skillRate × judge × (1 + dmgInc) × (1 + target.dmgRed))
CD = baseCD / max(1 + atkSpeedP, 0.1) / max(1 + atkSpeedPP, 0.1)
moveSpeed = baseSpeed × max(1 + moveSpeedP, 0)
shieldGrant = trunc(recipient.maxHP × max(0.03 × (1 + provider.hpShieldPP), 0))
retaliation = trunc(recipient.currentTotalATK × refAtkP)
```

`BattleFormula.CalcJudgeRate` (0x134ddc) samples `System.Random.NextDouble() × (whack + crit + max(0.3, 1 − whack − crit))`. The selected whack multiplier is `4 + whack`, and selected crit multiplier is `1.5 + crit`. For Pulse Blast II alone, whack=0.05 and crit=0 produce 5% whacks at 4.05×. Separate direct and splash damage evaluations roll independently. This does not use the integer probability path of Stasis.

## Complete powerup catalog

### How choices and ranks work

Level-up choices are supplied by the server. The client sends the battle EXP level, displays the returned talent IDs, then applies the server-confirmed choice. Listed weights, caps, turret correlations and prerequisite IDs describe the shipped configuration; they do not establish the exact chance of a card appearing.

Selecting a permanent powerup adds one rank and applies only that new rank to units already in battle. Newly created units receive their accumulated ranks. Bonuses using the same attribute add together: two Warmaster Soul ranks mean +10% attack, not 1.05 × 1.05. Separate attack and firing-speed factors multiply in sustained firing.

Tier II records point to their matching Tier I record. The client does not implement the offer filter, so the precise prerequisite rank and whether a correlated turret must be deployed cannot be confirmed from this runtime. Treat the named prerequisite as configured, without assuming it requires a maxed Tier I.

Supply Drop is an immediate item grant. It bypasses the permanent-rank dictionary, despite its configured max rank of 1. This explains its different behavior, but does not prove the server will offer it repeatedly.

### What the percentages multiply

Attack A = trunc(base attack × max(1 + summed attack bonuses, 0)). Ordinary hit damage = trunc(A × skill coefficient × hit-quality multiplier × (1 + attacker damage bonus) × (1 + target vulnerability)). Damage calculations use the skill owner’s current attributes when the hit resolves.

Cooldown = base cooldown ÷ max(1 + attack-speed bonus, 0.1) ÷ max(1 + second attack-speed factor, 0.1). Swift adds to the first factor. Wish Crystal and Quantum burn helpers explicitly bypass both factors; they also bypass the energy hold’s doubled attack-timer progress.

For a normal attacking turret with 1,000 base attack and a 2-second cooldown, ten Warmaster Soul ranks and ten Swift ranks give 1,500 attack every 1.333 seconds: 2.25 times the original sustained attack throughput, before targeting, skill patterns and integer damage rounding.

Movement = base speed × max(1 + summed movement modifiers, 0). Different slow IDs and the global Slow talent add together. Reapplying the same slow ID refreshes its timer instead of adding another copy.

### 1: Swift

**Maximum rank:** 10. **Configured weight:** 10. **Prerequisite ID:** none. **Turret correlation:** none. **Quality code:** 2.

+5% firing rate per rank; +50% at rank 10.

**Effect and stacking**

Adds 0.05 per rank to atkSpeedP for friendly turrets. The resulting cooldown divides by 1 + the combined bonus. It improves the cadence of damage turrets and Energy Barrier shield applications.

Wish Crystal and Quantum’s hidden burn helpers use their base cooldown directly, so Swift does not increase core production or burn pulse frequency.

**Worked example**

A 2-second cooldown becomes 2/1.05 = 1.905 seconds at rank 1 and 2/1.5 = 1.333 seconds at rank 10. The tenth rank improves throughput by 1.50/1.45 − 1 = 3.45% relative to rank 9.

**Priority:** Strong general choice for a mixed attacking formation; especially useful when faster control or shield application also matters.

**Evidence:** fortress_skill_talent 1; BattleFieldDataProvider.UpdateTalent 0x13838c; BattleUnit.GetFinalAtkSpeed 0x13c42c; BattleFormula.CalcSkillCd 0x134d56.

### 2: Slow

**Maximum rank:** 10. **Configured weight:** 10. **Prerequisite ID:** none. **Turret correlation:** none. **Quality code:** 2.

−5 percentage points of enemy movement speed per rank; −50% at rank 10.

**Effect and stacking**

Adds −0.05 per rank to every enemy’s moveSpeedP. The target check uses enemy camp, so it includes bosses and summoned enemies.

Adds to the direct Stasis slow and its separate splash slow. The final movement factor has a zero floor. This talent alone does not satisfy Frost Slow II’s requirement for the specific direct-slow buff 10201.

**Worked example**

A speed-2 enemy moves at 1 with rank 10. With direct Stasis slow it moves at 2 × (1 − 0.50 − 0.30) = 0.4. If the separate splash slow is also present, 1 − 0.50 − 0.30 − 0.30 is clamped to zero.

**Priority:** High priority when enemies reach the formation before it can kill them. Its value is extra firing time, not a direct damage multiplier.

**Evidence:** fortress_skill_talent 2; BattleFieldDataProvider.UpdateTalent 0x13838c; BattleFormula.CalcMoveSpeed 0x134d7d.

### 3: Warmaster Soul

**Maximum rank:** 10. **Configured weight:** 10. **Prerequisite ID:** none. **Turret correlation:** none. **Quality code:** 2.

+5% attack per rank; +50% at rank 10.

**Effect and stacking**

Adds 0.05 per rank to friendly turret atkP. This shares an additive pool with Pulse Blast I, Gravity Shock I and applicable Barrier attack auras.

Quantum burn helpers deliberately accept only talents 13 and 23. Warmaster improves Quantum’s direct attack but not its burn helper. Helpers use their own configured attack of 270/360/480/960 at helper levels 3/4/5/6; the actual summon path does not inherit the parent’s buffed attack.

**Worked example**

With 1,000 base attack, ten ranks give 1,500 attack. If Pulse Blast I is also rank 3, an Explosive Missile has 1,000 × (1 + 0.50 + 0.36) = 1,860 attack; multiplying 1.50 × 1.36 would overstate it.

**Priority:** Strong general damage choice. Favor Swift instead when shield or control cadence is valuable, and dedicated burn bonuses when Quantum helper damage dominates.

**Evidence:** fortress_skill_talent 3; BattleUnit.InitTalentAttr 0x13ba48; BattleUnit.OnTalentChanged 0x13b9f8; UnitFactory.SummonHeroFromConfig 0x13d6d8; BattleFieldDataProvider.GetHeroUnitAttr 0x138008.

### 4: Tactical Prowess

**Maximum rank:** 5. **Configured weight:** 5. **Prerequisite ID:** none. **Turret correlation:** none. **Quality code:** 3.

+8% kill EXP per rank; +40% at rank 5.

**Effect and stacking**

Applies expP to enemy units, whose death award uses their own bonus. The EXP award is converted to an integer before entering the battle model.

EXP earned beyond the next talent threshold is discarded by the recovered level-up path, so a larger award can waste some of its benefit. Summoned enemies with zero base EXP remain worth zero.

**Worked example**

A base award of 100 becomes 108 at rank 1 and 140 at rank 5. If only 30 EXP is needed for the next choice, that kill contributes 30 toward progression; the excess does not carry to the next threshold.

**Priority:** Consider early when the defense is stable and many kills remain. Immediate damage or control is more useful when survival is already tight.

**Evidence:** fortress_skill_talent 4; BattleFieldDataProvider.UpdateTalent 0x13838c; BattleUnit.GetKillExp 0x13c48a; BattleWorld.OnUnitDead 0x138c14; FortressModel.AddBattlingExp 0x192614.

### 12: Frost Slow I

**Maximum rank:** 3. **Configured weight:** 10. **Prerequisite ID:** none. **Turret correlation:** 2. **Quality code:** 3.

Direct Stasis slow lasts +0.4 seconds per rank, up to 3.2 seconds.

**Effect and stacking**

Extends buff 10201 from 2 seconds to 2 + 0.4 × rank seconds. It increases duration only: the movement penalty remains 30%.

Reapplying this same buff resets its timer to the current upgraded duration. It does not add another 30% slow. The separate one-second splash slow 10203 and one-second freeze 10202 are not extended.

**Worked example**

At rank 3, a direct slow lasts 3.2 seconds instead of 2. A reapplication 1 second later restarts a full 3.2-second timer, rather than leaving 5.2 seconds or applying a second −30%.

**Priority:** Useful with Stasis when slow coverage has gaps; less valuable when repeated hits already maintain the direct slow continuously.

**Evidence:** fortress_skill_talent 12; fortress_buff 10201/10202/10203; BuffContainer.AddBuff 0x13495c.

### 13: Matrix Bombardment I

**Maximum rank:** 3. **Configured weight:** 10. **Prerequisite ID:** none. **Turret correlation:** 3. **Quality code:** 3.

+8% Quantum burn damage per rank; +24% at rank 3.

**Effect and stacking**

Adds dmgInc only to type-99 Quantum burn helpers. It does not increase Quantum’s initial area hit. The helpers appear from level-3 and higher Quantum skills.

Each burn pulse uses the helper’s own attack, a 0.48 skill coefficient, this damage bonus and the target’s vulnerability. Swift and Warmaster do not modify helper attack or pulse cadence.

**Worked example**

A level-3 helper with 270 attack deals trunc(270 × 0.48) = 129 per neutral pulse. Rank 3 gives trunc(270 × 0.48 × 1.24) = 160. This is per hit on a target inside the helper’s area, not total damage per cast.

**Priority:** Good when level-3+ Quantum helpers contribute a substantial share of damage. Skip as an immediate damage upgrade for a formation with only level-1/2 Quantum.

**Evidence:** fortress_skill_talent 13; BattleUnit.InitTalentAttr 0x13ba48; fortress_hero 9903–9906; fortress_skill_active 19903–19906; SkillLogic.GetDamageValue 0x13a704.

### 14: Pulse Blast I

**Maximum rank:** 3. **Configured weight:** 10. **Prerequisite ID:** none. **Turret correlation:** 4. **Quality code:** 3.

+12% Explosive Missile attack per rank; +36% at rank 3.

**Effect and stacking**

The tooltip says normal attack damage, but the code adds atkP to turret type 4. That attack feeds normal, empowered direct and splash damage calculations.

Its bonus adds to Warmaster and attack auras. It does not replace the skill’s 1.2 normal, 1.8 empowered direct or 1.0 splash coefficient.

**Worked example**

With 1,000 base attack and no other bonuses, the normal 1.2 hit rises from 1,200 to 1,344 at rank 1, or 1,632 at rank 3. The same rank-3 attack value produces a 2,448 empowered direct hit and 1,360 splash hit.

**Priority:** Efficient for an Explosive-heavy formation: its first rank is 12% to that turret’s attack versus Warmaster’s 5% across the formation.

**Evidence:** fortress_skill_talent 14; BattleFieldDataProvider.UpdateTalent 0x13838c; BulletLogic.GetDamageValue 0x13c778; BulletLogic.GetExtendDamageValue 0x13c8fc; fortress_skill_bullet 10401–10407.

### 15: Gravity Shock I

**Maximum rank:** 3. **Configured weight:** 10. **Prerequisite ID:** none. **Turret correlation:** 5. **Quality code:** 3.

+10% Expulsion Missile attack per rank; +30% at rank 3.

**Effect and stacking**

Adds atkP to turret type 5. Main and fragment shots use that owner’s modified attack, so the gain is not limited to the central projectile.

Adds to Warmaster and other atkP bonuses; the projectile’s separate skill coefficient still applies.

**Worked example**

A 1,000-attack Expulsion main shot becomes 1,300 at rank 3. A 0.5-coefficient fragment becomes 650 instead of 500, with normal hit quality and no target vulnerability.

**Priority:** Good for an Expulsion-heavy formation, especially where multiple projectiles connect. The actual number of hits depends on geometry.

**Evidence:** fortress_skill_talent 15; BattleFieldDataProvider.UpdateTalent 0x13838c; BulletLogic.GetDamageValue 0x13c778; fortress_skill_bullet 10501–10510.

### 16: Weak Point Snipe I

**Maximum rank:** 1. **Configured weight:** 10. **Prerequisite ID:** none. **Turret correlation:** 6. **Quality code:** 3.

Configured +50% Focused Cannon damage against targets at or below the 30% HP threshold.

**Effect and stacking**

This is a conditional +0.5 dmgInc term evaluated when a Focused Cannon bullet hits. It is separate from attack bonuses. The comparison is inclusive: HP equal to the threshold qualifies.

The actual threshold reads target.attrSet.hpMax × 0.30, while ordinary maximum HP lives in FinalAttr.hpMax. The traced spawn path clears attrSet.hpMax and does not populate it. The expected 30% behavior therefore needs a live check.

When Weak Point Snipe II is owned, this Tier I branch is skipped completely, including failed execute rolls and boss targets.

**Worked example**

If the threshold backing field contains 1,000, a target at 300 HP qualifies and a neutral 100-damage shot becomes 150. If that backing field remains zero, a living positive-HP target cannot pass the condition.

**Priority:** Treat as an uncertain pick in this snapshot. Do not build a damage estimate around its advertised bonus without confirming the HP condition in play.

**Evidence:** fortress_skill_talent 16; BulletLogic.GetDamageValue 0x13c778; BattleUnit.SetData 0x13b824; AttributeSet.Clear 0x13d84c; UnitFactory.CreateEnemyFromConfig 0x13d578.

**Limits:** The HP backing-field discrepancy is unresolved by static client tracing; no categorical claim of live ineffectiveness is made.

### 17: Laser Beam I

**Maximum rank:** 3. **Configured weight:** 10. **Prerequisite ID:** none. **Turret correlation:** 7. **Quality code:** 3.

Raises Fission’s enemy vulnerability from 10% to 14%/18%/22%.

**Effect and stacking**

The upgrade modifies buff 10703, which level-3+ Fission applies to enemies for 5 seconds. It adds 0.04 per rank to the buff’s base dmgRed value of +0.10.

Despite the internal name dmgRed, positive values multiply damage taken by 1 + dmgRed. This is target vulnerability, so ordinary incoming damage from other turrets and Quantum helpers also benefits. Repeated Fission hits refresh the same debuff instead of stacking another copy.

The debuff is applied after the hit. The first hit on an unmarked enemy does not gain that newly applied vulnerability.

**Worked example**

A neutral 1,000-damage hit against a newly established rank-3 mark becomes 1,220 rather than 1,100 with the base mark. The upgrade is 1.22/1.10 − 1 = 10.91% more damage while the mark is present, not 12% more relative to the already-marked target.

**Priority:** High value with level-3+ Fission when the rest of the formation attacks the same marked enemies. No immediate benefit before Fission unlocks the debuff.

**Evidence:** fortress_skill_talent 17; fortress_buff 10703; fortress_skill_active 10703–10706; BattleFieldDataProvider.AddBuffAttr 0x138808; BattleFormula.CalcSkillDamage 0x134e8a.

**Limits:** Existing refresh-only marks are not immediately recomputed when this talent ranks up; examples assume a newly applied mark.

### 18: Area Protection I

**Maximum rank:** 3. **Configured weight:** 10. **Prerequisite ID:** none. **Turret correlation:** 8. **Quality code:** 3.

+8% shield amount per rank; +24% at rank 3.

**Effect and stacking**

Adds hpShieldPP to Energy Barrier. Each shield grant is trunc(recipient maximum HP × 0.03 × (1 + provider hpShieldPP)). The percentage is a multiplier on the existing 3% shield, not +8 percentage points of HP.

Shield grants add to the recipient’s current shield, capped at the recipient’s maximum HP. Multiple applications therefore build shield, while the hpShieldP attribute itself is replaced at 0.03 rather than stacking per cast.

**Worked example**

For a 1,000-HP recipient, one grant is 30 shield normally and trunc(1,000 × 0.03 × 1.24) = 37 at rank 3. Three uninterrupted grants add 111, up to the 1,000 shield cap.

**Priority:** Useful when Barrier’s recipients are under sustained pressure. Swift also increases grant frequency and helps the attacking formation, so compare both options.

**Evidence:** fortress_skill_talent 18; BaseBuff.OnAdd 0x134868; BattleFormula.CalcHpShield 0x134d92; BattleUnit.UpdateShield 0x13c378.

### 19: Supply Drop

**Maximum rank:** 1. **Configured weight:** 5. **Prerequisite ID:** none. **Turret correlation:** none. **Quality code:** 3.

Immediately grants 15 Reforge Cores.

**Effect and stacking**

Adds 15 of item 501 and emits the item-update event. It is an immediate gain rather than a persistent combat attribute.

The client does not store this choice in the permanent talent-rank dictionary. Its configured max rank is 1; future offer eligibility remains server-controlled.

**Worked example**

The configured pre-war random-hero refresh costs 15 cores, so this grant equals one such refresh at the shipped cost.

**Priority:** Choose when an extra refresh now can fix a weak formation or complete a merge. Its future value depends on what the refresh offers.

**Evidence:** fortress_skill_talent 19; FortressModel.UpdateSelectedTalent 0x192794; game_base_param fortress_random_hero_cost=15.

### 22: Frost Slow II

**Maximum rank:** 1. **Configured weight:** 10. **Prerequisite ID:** 12. **Turret correlation:** 2. **Quality code:** 4.

Adds 10 percentage points to Stasis freeze chance when the target has direct Stasis slow.

**Effect and stacking**

Level-3+ Stasis already has a nominal 15% one-second freeze. With this talent and buff 10201 present, the configured rate becomes 25%. The test specifically checks that buff ID; global Slow and the separate splash slow do not qualify.

The direct slow roll executes before the freeze roll, so a slow applied by the same hit can enable the bonus. The runtime rolls an integer from 0 through 99 and succeeds when roll ≤ rate × 100: the nominal 15% and 25% thresholds accept 16 and 26 outcomes respectively.

Level-5/6 Stasis also applies the separate one-second, −30% splash slow to other living enemies within radius 1 after a successful freeze. This talent increases how often that success branch runs; it does not lengthen the freeze or splash.

**Worked example**

Against an already directly slowed target, freeze rises from 16/100 to 26/100 accepted rolls. Against a fresh target, the direct slow accepts 51/100 rolls first, so the combined freeze chance is approximately 0.51 × 0.26 + 0.49 × 0.16 = 21.1%, assuming independent uniform rolls.

**Priority:** A useful control upgrade with level-3+ Stasis, strongest when direct slow is already maintained. It adds no freeze to a level-1/2 Stasis skill that lacks the freeze effect.

**Evidence:** fortress_skill_talent 22; BulletLogic.AddBuff 0x13cfc0; RandomUtility.Range 0x135637; XorShiftRandom.Next 0x135444; fortress_skill_bullet 10203–10206.

### 23: Matrix Bombardment II

**Maximum rank:** 1. **Configured weight:** 10. **Prerequisite ID:** 13. **Turret correlation:** 3. **Quality code:** 4.

Extends each newly created Quantum burn helper from 2 to 3 seconds.

**Effect and stacking**

Adds one second to the helper’s configured lifetime. It does not increase pulse damage or change the helper’s base interval: 1 second at levels 3/4, 0.5 seconds at levels 5/6.

This increases the opportunity to hit enemies that remain in the burn area. Helper expiry is checked before its attack branch; the exact pulse count depends on its timer initialization and simulation boundaries, so duration should not be converted blindly into a guaranteed 50% damage gain.

The talent-change handler updates FinalAttr.lifeTime, whereas the running helper decrements a separate lifeTime field. The reliable benefit is to helpers created after the selection, rather than an assumed immediate extension of every existing burn.

**Worked example**

A new level-5 helper stays available for 3 seconds instead of 2, giving one extra second at its unchanged 0.5-second base cadence. Per-hit damage remains trunc(480 × 0.48) = 230 without other bonuses.

**Priority:** Good for level-3+ Quantum against enemies that stay inside its burn zones; less useful when enemies leave quickly.

**Evidence:** fortress_skill_talent 23; BattleFieldDataProvider.GetLifeTime 0x138544; BattleUnit.UpdateLifeTime 0x13bb21; BattleUnit.HeroEagleTrapUpdate 0x13bd90; fortress_hero 9903–9906.

### 24: Pulse Blast II

**Maximum rank:** 1. **Configured weight:** 10. **Prerequisite ID:** 14. **Turret correlation:** 4. **Quality code:** 4.

Adds a 5% whack weight to Explosive Missile; the selected multiplier is 4.05× in this runtime.

**Effect and stacking**

Adds 0.05 whack to turret type 4. The shared hit-quality calculation chooses between whack, critical and normal outcomes. With no other crit or whack values, the resulting whack probability is 5%.

The runtime passes the whack value itself to 4 + whack, producing a 4.05 multiplier here rather than exactly 4.00. Direct and splash damage evaluations roll separately. Its attack source is still modified by Pulse Blast I and Warmaster.

**Worked example**

A neutral 1,200 normal hit becomes 4,860 when this whack result is selected. With only this 5% whack weight, the expected multiplier is 0.95 × 1 + 0.05 × 4.05 = 1.1525, a 15.25% gain before integer rounding and overkill.

**Priority:** Strong offensive choice for Explosive Missile once its prerequisite is available; valuable across both its direct and splash calculations.

**Evidence:** fortress_skill_talent 24; BattleFormula.CalcJudgeRate 0x134ddc; BattleFormula.CalcWhackDamageRate 0x134e38; BulletLogic.GetDamageValue 0x13c778; BulletLogic.GetExtendDamageValue 0x13c8fc.

### 25: Gravity Shock II

**Maximum rank:** 1. **Configured weight:** 10. **Prerequisite ID:** 15. **Turret correlation:** 5. **Quality code:** 4.

Adds a nominal 4% chance for Expulsion hits to knock an enemy upward by 0.5 logic units.

**Effect and stacking**

The projectile hit path requires an Expulsion owner, this talent and a living hit target. It does not exclude bosses. The integer roll succeeds at ≤ 4, accepting 5 of the 100 integer outcomes in the current generator.

Base displacement is 0.5 plus the repelDistance attribute, scaled to the scene. The target’s KnockBackPosition method imposes a 3-second per-target time gate; a successful proc roll is not a guarantee that every closely spaced hit moves the target.

**Worked example**

With no extra repel distance, an accepted displacement moves a target upward by 0.5 logic units. At 5 accepted rolls per 100 eligible hits on average, useful displacement still depends on the per-target time gate and subsequent movement.

**Priority:** A situational control choice for an Expulsion-heavy formation. Global Slow is more predictable; this provides occasional displacement.

**Evidence:** fortress_skill_talent 25; BulletLogic.KnockBackUnit 0x13c9a4; BattleFormula.CalcRepelDistance 0x134dbb; BattleUnit.KnockBackPosition 0x13babd; BattleUnit..ctor 0x13b784; XorShiftRandom.Next 0x135444.

### 26: Weak Point Snipe II

**Maximum rank:** 1. **Configured weight:** 10. **Prerequisite ID:** 16. **Turret correlation:** 6. **Quality code:** 4.

Configured execute chance against low-HP normal enemies; replaces the Tier I damage branch.

**Effect and stacking**

For Focused Cannon, the runtime first checks this talent. Only IsNormalMonster targets, with base type 1, can enter the execute test. Bosses and other enemy types do not qualify.

If the inclusive 30% threshold passes, it rolls an integer 0–99 and accepts ≤ 10: 11 accepted outcomes for the nominal 10% rate. A successful result returns current HP + current shield as damage, bypassing ordinary attack and damage multipliers.

Owning this talent suppresses Weak Point Snipe I completely, even if the target is a boss, fails the HP condition, or survives a failed execute roll. The same unresolved attrSet.hpMax threshold backing field affects this upgrade.

**Worked example**

Conditionally, if the threshold field is 1,000, an eligible normal enemy at 250 HP with 40 shield is dealt 290 on a successful execute. On a failed roll the ordinary shot is used, without Tier I’s +50% term.

**Priority:** Do not assume this is a strict upgrade over Tier I. It trades away the earlier conditional bonus, excludes bosses and has the unresolved HP-field issue.

**Evidence:** fortress_skill_talent 26; BulletLogic.GetDamageValue 0x13c778; BattleUnit.IsNormalMonster 0x13b93c.

**Limits:** Execution is conditional on the HP backing field being populated correctly; live behavior was not measured.

### 27: Laser Beam II

**Maximum rank:** 1. **Configured weight:** 10. **Prerequisite ID:** 17. **Turret correlation:** 7. **Quality code:** 4.

Widens Fission’s damage rectangle from 1.5 × 5 to 2 × 6 logic units.

**Effect and stacking**

Changes the rectangle used by the Fission area-hit test. Width increases from 1.5 to 2, length from 5 to 6; the skill’s release-distance value remains 8.

It does not add a damage multiplier or another beam. More enemies can be hit and marked if their positions fall inside the larger rectangle.

**Worked example**

The configured rectangle area rises from 7.5 to 12 square logic units, a 60% increase. That is extra coverage, not a guaranteed 60% damage increase.

**Priority:** Good when enemies spread beyond the beam’s normal strip, especially with Fission vulnerability supporting other turrets.

**Evidence:** fortress_skill_talent 27; SkillLogic.Init 0x13a5ec; BattleFieldDataProvider.GetMonkeyTalentRect 0x1385d0; SkillLogic.OnSkillToRectTarget 0x13af18; fortress_skill_active 10701–10706.

### 28: Area Protection II

**Maximum rank:** 1. **Configured weight:** 10. **Prerequisite ID:** 18. **Turret correlation:** 8. **Quality code:** 4.

A Barrier shield application grants retaliation equal to 30% of the recipient’s attack.

**Effect and stacking**

When shield buff 10801 is applied with this talent owned, it sets refAtkP to 0.30 on the recipient. Repeated applications replace this value, rather than stacking it.

On a positive incoming hit, a friendly recipient with refAtkP > 0.1 retaliates against the damage source for trunc(recipient’s current total attack × refAtkP). This uses the recipient’s attack, not the incoming damage or Barrier’s zero attack.

The recovered retaliation check runs before shield absorption and never checks remaining shield. The buff-removal path does not remove its extra refAtkP field. Thus the traced implementation can keep retaliating after shield depletion; do not describe the shield as a runtime requirement for every retaliation.

**Worked example**

A shielded damage turret with 1,000 current attack retaliates for 300 when hit for either 10 or 500. A Barrier turret with zero attack returns zero. An attack bonus raising the recipient to 1,500 also raises retaliation to 450.

**Priority:** Useful when high-attack turrets are regularly being hit and receive Barrier shields. It contributes little when enemies die before attacking or only a zero-attack Barrier takes damage.

**Evidence:** fortress_skill_talent 28; fortress_buff 10801; BattleFieldDataProvider.AddBuffAttr 0x138808; BattleFieldDataProvider.RemoveBuffAttr 0x138a04; BattleUnit.TakeDamage 0x13c008; BattleUnit.CalcRefAtkV 0x13c298.

### 31: Mechanical Evolution

**Maximum rank:** 5. **Configured weight:** 1. **Prerequisite ID:** none. **Turret correlation:** none. **Quality code:** 4.

Configured level-3 turret draw weight rises by 1 percentage point per rank, from 1.8% to 6.8%.

**Effect and stacking**

The change list links rank states 0–5 to random tables 100–105. Each table totals 1,000 weight. The branch leading to level-3 turrets rises from 18 to 68, while the level-1 branch falls from 792 to 742.

The level-2 branch stays at 72, the Wish Crystal branch at 18, and the block branch at 100. The level-3 branch table 2 lists seven equally weighted turret types, excluding Wish Crystal.

These are configured branch weights. The actual refresh and choice results are supplied by the server, so the client cannot prove exactly how those tables are applied to an offered refresh.

**Worked example**

On the configured mapping, rank 1 changes the level-3 branch from 18/1,000 to 28/1,000, and rank 5 to 68/1,000. Relative to the base 1.8% branch, 6.8% is about 3.78 times as frequent, but only 5 percentage points higher in absolute terms.

**Configured branch weights**

| Rank | Lv.3 turret | Lv.2 turret | Lv.1 turret | Wish Crystal | Block |
| --- | --- | --- | --- | --- | --- |
| 0 | 1.8% | 7.2% | 79.2% | 1.8% | 10.0% |
| 1 | 2.8% | 7.2% | 78.2% | 1.8% | 10.0% |
| 2 | 3.8% | 7.2% | 77.2% | 1.8% | 10.0% |
| 3 | 4.8% | 7.2% | 76.2% | 1.8% | 10.0% |
| 4 | 5.8% | 7.2% | 75.2% | 1.8% | 10.0% |
| 5 | 6.8% | 7.2% | 74.2% | 1.8% | 10.0% |

**Priority:** An early investment when many future refreshes remain; it offers no immediate combat increase. Supply Drop is the more immediate formation-building option.

**Evidence:** fortress_skill_talent 31; fortress_random 2/100–105; FortressServiceRmt.RandomTalent 0x194130.

**Limits:** Rank-to-table mapping is configuration evidence; server execution and offer filtering are not recovered.

## Buff behavior and ownership

`BuffContainer.AddBuff` (0x13495c) implements the rules as follows: rule 0 finds an existing same-ID buff and resets its duration without calling `OnAdd`; rule 1 creates a separate instance and adds attributes; rule 2 also creates a separate instance, but the provider replaces the configured attributes. `BaseBuff.OnProviderDead` only expires IDs listed in that provider’s survival-buff array. Ordinary Stasis and shield grants are therefore not automatically removed simply because their source dies.

`BattleUnit.Step` (0x13bba8) advances a friendly damage turret’s buff timer at twice the normal rate while the energy control is held; Wish Crystal and type-99 helpers use the normal factor. Enemy debuffs advance at the normal battle-time rate. Durations here are battle-timer values rather than UI wall-clock guarantees.

### Buff 10201: Direct Stasis slow

**Source:** Stasis Missile, all turret levels. **Duration:** 2 + 0.4 × Frost Slow I rank seconds.

moveSpeedP −0.30. Different slow IDs and global Slow add to it. This exact buff ID enables Frost Slow II’s higher freeze rate.

Same ID refreshes duration; does not stack with itself.

Slow rank 4 and direct slow yield movement factor 1 − 0.20 − 0.30 = 0.50.

**Evidence:** BuffContainer.AddBuff 0x13495c; BulletLogic.AddBuff 0x13cfc0.

### Buff 10202: Stasis freeze

**Source:** Stasis Missile level 3+. **Duration:** 1 second.

BattleUnit.Freeze detects this ID. Only the movement agent checks this freeze state. Enemies already in attack range can still attack; this is a movement freeze, separate from a movement-speed reduction. Frost Slow I does not extend it.

Same ID refreshes duration.

A second successful freeze after 0.6 seconds resets another full one-second timer.

**Evidence:** BattleUnit.Freeze 0x13c5c5; MoveAgent.MoveTo 0x139bec; BattleUnit.MonsterUpdate 0x13bf14; BuffContainer.AddBuff 0x13495c.

### Buff 10203: Stasis splash slow

**Source:** Successful level-5/6 Stasis freeze, other living enemies within radius 1. **Duration:** 1 second.

moveSpeedP −0.30. Does not enable Frost Slow II’s special check by itself.

Refreshes itself; stacks additively with direct slow because the IDs differ.

An enemy under both direct and splash slow moves at 40% speed before global Slow.

**Evidence:** BulletLogic.AddBuff 0x13cfc0; fortress_skill_bullet 10205/10206.

### Buff 10703: Fission vulnerability

**Source:** Fission Laser level 3+. **Duration:** 5 seconds.

Target dmgRed +0.10, plus 0.04 per Laser Beam I rank. The ordinary damage formula multiplies by 1+dmgRed.

Repeated applications refresh one instance; does not stack per beam or turret.

A fresh rank-3 mark changes a neutral 1,000 incoming hit to 1,220.

**Evidence:** BattleFieldDataProvider.AddBuffAttr 0x138808; BattleFormula.CalcSkillDamage 0x134e8a.

### Buff 10801: Barrier shield grant

**Source:** Energy Barrier active skill, all turret levels. **Duration:** 999 buff-timer seconds; actual shield can be consumed earlier.

Shield grant = trunc(recipient max HP × 0.03 × (1 + provider Area Protection I bonus)); adds refAtkP 0.30 when Area Protection II is owned.

Creates another buff instance but replaces hpShieldP at 0.03. Each application adds shield, capped at recipient max HP.

At rank 3, a 1,000-HP recipient gains 37 per application, up to 1,000 stored shield.

**Evidence:** BaseBuff.OnAdd 0x134868; BattleUnit.UpdateShield 0x13c378.

### Buff 10803: Barrier attack aura

**Source:** Energy Barrier survivalBuff at turret levels 3–6. **Duration:** 999 buff-timer seconds; marked for expiry when its survival-buff provider dies.

The formation controller applies the buff to cross-neighbors. A separate first-step AddOnceBuff path reverses owner/provider arguments; exact combined behavior for dynamically added/reused units needs a live check.

Separate instances add +0.10 atkP; different providers can contribute.

Two correctly applied aura instances contribute +20% to the shared attack-bonus pool, not 1.1 × 1.1.

**Evidence:** BattleController.InitFoxBornBuff 0x1367e0; BattleUnit.AddOnceBuff 0x13c4d8; BaseBuff.OnProviderDead 0x134818.

### Buff 10805: Barrier kill-energy aura

**Source:** Energy Barrier survivalBuff at turret level 6 in this configuration. **Duration:** 999 buff-timer seconds; marked for expiry when its survival-buff provider dies.

Increases the energy returned when the buffed turret is the damage source that kills an enemy. The level-5 tooltip mentions this bonus, but level-5 survivalBuff contains only the attack aura; the energy aura appears in level 6.

Separate instances add +0.10 energyP.

One valid aura raises a normal 10-energy kill to 11, and two to 12, before the 1,500 energy cap.

**Evidence:** fortress_hero 805/806; BattleUnit.TakeDamage 0x13c008; BattleController.OnEnergyChanged 0x136964.

## Internal enemy passives

### 30701: Blitz Armor

This is an enemy passive, not a selectable player powerup. Enemy type 7 adds +2% attack each time its timer exceeds 0.5 seconds, resetting the timer after each addition. Maximum 50 additions, or +100% attack.

After 20 additions, base attack 1,000 becomes 1,400; after 50 it becomes 2,000. The threshold is strictly greater than 0.5 seconds, so exact wall-clock milestones depend on the simulation step.

**Evidence:** fortress_npc_main 7; BattleUnit.MonsterAtkAddPerSec_30701 0x13be54.

### 30901: Fissioner

This is an enemy passive, not a selectable player powerup. On death, enemy type 9 summons two enemy-type-5 units. Each inherits 50% of the parent BaseAttr attack, 60% of its BaseAttr maximum/current HP values, and its base movement speed. The summoned units award zero kill EXP.

A parent with 1,000 base attack and 5,000 base HP produces two summons with 500 attack and 3,000 maximum HP each. These use base values, not the parent’s accumulated attack bonuses.

**Evidence:** fortress_npc_main 9; MonsterUnit.Die 0x13d4b8; BattleFieldDataProvider.GetSummonMonsterUnitAttr 0x137ed8.

## Narrow unresolved runtime discrepancies

### Weak Point HP backing field

`BulletLogic.GetDamageValue` (0x13c778) uses `target.GetHp() <= target.attrSet.hpMax × 0.3` in both talent 16 and talent 26 checks. `GetHp` uses `FinalAttr.hp`; `GetMaxHp` (0x13c304) uses `FinalAttr.hpMax`. `BattleUnit..ctor` (0x13b784) creates a new `AttributeSet` and clones the supplied `UnitAttr` into BaseAttr/FinalAttr. `SetData` (0x13b824) clones the config unit and calls `AttributeSet.Clear` (0x13d84c), which sets `attrSet.hpMax` to zero. `UnitFactory.CreateEnemyFromConfig` (0x13d578) calls SetData and component initialization. No direct gameplay population of that attribute field was found; direct field references were limited to the two threshold reads, Clear and CloneTo. Reflective attribute changes are sourced from configs, and the audited talent/buff rows do not grant hpMax.

This strongly suggests a disconnected condition in the recovered static path, but it is not presented as a tested live no-effect claim. The guide gives the intended threshold conditionally and keeps this field discrepancy visible. The Tier II branch suppressing Tier I is directly proven regardless of this uncertainty.

### Refresh-only marks across a talent upgrade

Rule-0 reapplication calls `Reset` without reapplying attributes. `BattleFieldDataProvider.RemoveBuffAttr` reads the **current** talent rank when subtracting a talent-modified attribute. Thus a mark installed at one Laser Beam I rank and removed after a rank change may not remove the same value it originally added. The guide’s 14/18/22% examples assume a newly applied mark with a matching add/remove rank; prolonged refresh through rank changes needs observation.

### Barrier aura owner/provider direction

`BattleController.InitFoxBornBuff` (0x1367e0) applies a Barrier-provided aura to cross-neighbors. `BattleUnit.AddOnceBuff` (0x13c4d8) instead invokes `this.AddBuff(neighbor, id)` while this unit is the Barrier. Both methods exist in the current client; they should not be conflated into an uncomplicated universal adjacency claim for every spawn/setup path. The guide states the expected formation-controller effect and flags the reversed first-step path, while retaining the verified additive buff rule and source-death handling.

### Current helper lifetime versus new lifetime

`GetLifeTime` (0x138544) reads the two-second config and adds talent 23’s one second. `SetData` copies GetLifeTime into `BattleUnit.lifeTime`, which `HeroEagleTrapUpdate` (0x13bd90) decrements. `UpdateLifeTime` (0x13bb21), used by the talent-change handler, only updates `FinalAttr.lifeTime`. This supports the guide’s guarantee for newly spawned helpers while withholding a claim that current helpers gain a full second immediately.

## Practical priorities

### Reliable general picks

Swift and Warmaster are useful across a mixed damage formation. Compare the current additive pools: the next rank’s relative gain is 0.05/(1 + current bonus), so a less-invested pool often has the better marginal damage return. Swift additionally improves Stasis and Barrier cadence, while neither talent improves the hidden Quantum helper.

Slow deserves priority when contact pressure is the failure point. At rank 10, direct Stasis slow and global Slow leave an enemy at 20% of base movement speed, buying substantially more approach time without requiring a damage roll.

### Pick the upgrade that matches deployed damage

Pulse Blast I is efficient for Explosive-heavy damage; Gravity Shock I for Expulsion-heavy damage. A first +12% type-specific attack pick beats a first +5% global attack pick for total immediate damage only if Explosive supplies more than 5/12 = 41.7% of the affected attack-based damage, assuming neutral additive pools and equal uptime. For Gravity’s +10%, the analogous threshold is 50%.

With level-3+ Fission, Laser Beam I can improve damage from the whole formation against marked targets. Rank 3 gives 10.91% more damage relative to an existing base mark, subject to debuff coverage. Laser Beam II adds coverage rather than a flat damage percentage.

With level-3+ Quantum, Matrix Bombardment I directly improves helper hits; II gives them longer to hit targets. These are more relevant to the burn component than Warmaster or Swift.

### Economy early, survival when needed

Tactical Prowess and Mechanical Evolution need remaining kills or refreshes to pay back. Supply Drop gives immediate refresh currency and may be better when a current merge opportunity matters. EXP truncation and lost overflow reduce the realized value of Tactical Prowess.

Area Protection I improves the 3%-of-recipient-HP shield grant multiplicatively. Area Protection II is most useful when a high-attack recipient regularly takes hits; it is not based on incoming damage. Frost Slow I/II are useful when slow or freeze coverage has gaps.

### Upgrades to assess cautiously

Weak Point Snipe II is not automatically better than I: it disables the Tier I conditional damage branch and excludes bosses from execution. Both tiers have the unresolved HP-field check in this snapshot. Avoid treating their tooltip values as guaranteed damage gains.

Gravity Shock II has a small accepted proc rate and a target time gate. Mechanical Evolution’s configured branch weights do not prove live refresh probabilities. Neither should be presented as an unconditional best pick.

## Evidence index

| Artifact | Runtime scope |
| --- | --- |
| `output/fortress-powerups-audit/model.il.txt` | rank application and immediate item reward |
| `output/fortress-powerups-audit/service.il.txt` | server choice and confirmation requests |
| `output/fortress-powerups-audit/select-view.il.txt` | display of server-provided options |
| `output/fortress-powerups-audit/provider.il.txt` | attribute targeting, helper configs, buff application/removal |
| `output/fortress-powerups-audit/unit.il.txt` | attack, damage, retaliation, helper exceptions |
| `output/fortress-powerups-audit/unit-more.il.txt` | shield cap, timers, lifetime updates, enemy growth passive |
| `output/fortress-powerups-audit/unit-init.il.txt` | spawn initialization, normal-enemy predicate, knockback, aura first-step path |
| `output/fortress-powerups-audit/monster-update.il.txt` | constructor and enemy attack behavior |
| `output/fortress-powerups-audit/buff-container.il.txt` | refresh/add/replace stack dispatch |
| `output/fortress-powerups-audit/buff.il.txt` | buff lifetime, shield grant and provider-death handling |
| `output/fortress-powerups-audit/formula.il.txt` | shared arithmetic and whack probabilities |
| `output/fortress-powerups-audit/attribute-helper.il.txt` | additive attribute changes |
| `output/fortress-powerups-audit/attribute-clear.il.txt` | HP backing-field reset |
| `output/fortress-powerups-audit/bullet.il.txt` | conditional effects, execute, Stasis and knockback |
| `output/fortress-powerups-audit/skill.il.txt` | skill damage and rectangle targeting |
| `output/fortress-powerups-audit/skill-init.il.txt` | Fission area upgrade |
| `output/fortress-powerups-audit/factory.il.txt` | actual helper summon path and enemy initialization |
| `output/fortress-powerups-audit/spawn-controller.il.txt` | summon call chain and deployed units |
| `output/fortress-powerups-audit/controller.il.txt` | talent propagation and aura setup |
| `output/fortress-powerups-audit/monster.il.txt` | enemy death summons |
| `output/fortress-powerups-audit/random.il.txt` | integer-roll dispatch |
| `output/fortress-powerups-audit/xorshift.il.txt` | exclusive upper bound of integer rolls |
| `output/fortress-powerups-audit/move-agent.il.txt` | freeze check in movement |
| `output/fortress-powerups-audit/hpmax-xrefs.txt` | references to the threshold backing field |
| `output/fortress-powerups-audit/source-hashes.json` | SHA-256 of input tables/runtime |

## Source integrity

| Source | SHA-256 |
| --- | --- |
| `data/fortress_skill_talent.json` | `6e922324783159e7059631a8c405109f420611d1fdd9e74d6f4d31fd1f84ec47` |
| `data/fortress_buff.json` | `63f4fa21399631ef80aa6adf7fdde9faebfddb49fd79d9234addf4181fd953a6` |
| `data/fortress_attr.json` | `b712d99f106134e7b43ef0ed228fe083a81cf01df7d335f4fb1c2ce4abc053a2` |
| `data/fortress_random.json` | `11fc762d72146d800a673ea4c4c2180514cbc15c58d61c29f68b24db5f7347bc` |
| `data/fortress_hero.json` | `14af310e5b95e0d0a1b242ea3434658a3db948dd6bc798d680d1d282efacc134` |
| `data/fortress_npc_main.json` | `13df3f19c22881b02513464c5452c3d2f37d2de339ca69665dfdcf21146a579f` |
| `data/fortress_skill_active.json` | `8f29c72c5ec0b17bdbc6f99a902614964dd4ba7435b820f424d1a5ae15a8b47b` |
| `data/fortress_skill_bullet.json` | `d2d8d4a5a5d4b0b5c5d6774bd4e15554cd27c03afd800db52393934350b6084d` |
| `data/game_base_param.json` | `dc246858da1a2fd4ae366b3edc11838a9f32d4db89c0947068826b0faa9271f2` |
| `data/_loc_en.json` | `ef622aec6c25190751e7c2ceaf688831840148ffb071307ad05968fe0c1dba0a` |
| `output/crit-cap-audit-2026-09-27/Extension_4.dll` | `97bcc98eb7cce32de4b13464786b5e39542b9488a66c576d948044cce3c9f973` |

## Validation and limits

- Offer filtering, exact card-offer probabilities, prerequisite rank requirements and the server’s use of Mechanical Evolution are not available in the recovered client.
- Integer truncation, finite target uptime, projectile travel, overkill and attack-timer boundaries can change realized results. Numerical examples isolate the stated effect; they are not measured run DPS.
- Weak Point reads a separate attribute HP field in its threshold test. The recovered initialization clears that field; no population path was found. The intended 30% behavior is documented conditionally, not claimed as verified live behavior.
- Existing refresh-only buffs are not recalculated when their talent rank changes. The static add/remove paths also use the current rank during removal; sustained refresh across a rank change needs a live check before assigning an exact result.

The JSON was parsed and checked for all 20 selectable IDs, all seven buff IDs and both enemy passive IDs. Every explicit runtime method/RVA citation was compared against the managed assembly’s method table. The audit does not assert live-server choice rates or measured DPS.


---

# Fortress Defense rules audit

Audited 2026-10-03 against the refreshed extracted tables and recovered managed assemblies. The refreshed build version is unknown. This audit does not treat the older unpacked asset tree or the old method-name dump as behavioral proof. No site files were edited by this research task.

Player copy is in `docs_src/fortress-rules-notes.json`. Its root object has `summary`, `sections`, `sources`, and `limitations`. Each section has `id`, `group`, `title`, and plain-text `paragraphs`; optional `table` contains `columns` and same-width `rows`, all strings. The groups are `rules`, `resources`, `enemies`, `progression`, and `calculations`.

## Confirmed corrections and boundaries

- The campaign has 99 configured stages, 27 boss stages, 50 scheduled enemies per stage, and a 180-second battle timer. All 99 rows and all scheduled enemy IDs were joined and checked. The scheduled count excludes summons.
- Victory requires an exhausted spawn queue and no surviving enemy, including summons. Loss occurs when the regular deployed defender count reaches zero or time expires. Temporary view-99 helpers do not sustain that count.
- The three named bosses cycle Blitz Armor, Crimson Wraith, and Fissioner. The 27 encounters are not 27 different boss definitions. The eight enemy definitions include six visible enemies and two summoned types.
- The 75 EXP rows are outgoing thresholds from level 0 through level 74, unlocking levels 1 through 75. Unlike some other progression tables, the last row is paid. Reaching level 75 costs 575,230 retained EXP.
- Kill EXP is truncated before being added. A level-up kill adds only the missing amount for that level and discards the remainder; it cannot complete several levels. Additional EXP is ignored after level 75.
- Nitrogen is a doubled attack-timer branch, not an ATK stat increase. It also doubles the eligible unit's own buff aging. It excludes Wish Crystals and temporary Quantum helpers. Drain is 75 energy per simulated battle second, starting capacity is configured as 750, meter cap is 1,500, and base kill recovery is 10.
- Positive target `dmgRed` increases received skill damage. The damage function multiplies by `1 + dmgRed`; it is not automatically positive armor.
- Critical and heavy-hit attributes serve as both random-outcome weights and additions to their respective multipliers. The normal outcome has a minimum weight of 0.30, not an unconditional probability floor after normalization.
- The raw hero definition's critical field is not copied by the recovered normal hero-attribute constructor. Combat hit evaluation reads the active attribute set, so a displayed 5%/10% value must not silently be added to baseline crit.
- Shield formula output is a grant. Grants accumulate through `BattleUnit.UpdateShield`, capped at recipient maximum HP. Calling each formula output the total shield capacity would be incorrect.

## Runtime evidence

Unless otherwise stated, all RVAs below refer to `output/crit-cap-audit-2026-09-27/Extension_4.dll`. Combat types use namespace `FlightShooting.Fortress`; the model and remote-service types use `FlightShooting.Data`. The prebattle controller uses `FSCore.Fortress.PreBattle.Controller`.

| Claim | Type and method | RVA | Saved evidence |
| --- | --- | --- | --- |
| ATK/HP additive bucket then integer conversion | BattleFormula.CalcAtk / CalcHpMax | 0x134d28 / 0x134d3f | formula.il.txt |
| Two separate multiplicative speed denominators, each clamped at 0.1 | BattleFormula.CalcSkillCd | 0x134d56 | formula.il.txt |
| Move speed clamps at zero | BattleFormula.CalcMoveSpeed | 0x134d7d | formula.il.txt |
| Shield-grant formula | BattleFormula.CalcHpShield | 0x134d92 | formula.il.txt |
| Skill coefficient: A + B + talentRate × layers | BattleFormula.CalcSkillRate | 0x134dab | formula.il.txt |
| Weighted hit outcomes | BattleFormula.CalcJudgeRate (with out judgeType) | 0x134ddc | formula.il.txt |
| Heavy multiplier 4 + W, crit multiplier 1.5 + C | BattleFormula.CalcWhackDamageRate / CalcCritDamageRate | 0x134e38 / 0x134e44 | formula.il.txt |
| Normal weight max(0.3, 1 − W − C) | BattleFormula.CalcNormalRate | 0x134e50 | formula.il.txt |
| Skill damage and signed target modifier | BattleFormula.CalcSkillDamage | 0x134e8a | formula.il.txt |
| Reflection and shield-before-HP subtraction | BattleFormula.CalcRefDamage / CalcRemainingHpShieldAndHp | 0x134ea4 / 0x134eaa | formula.il.txt |
| EXP/energy multiplicative modifier, return float | BattleFormula.CalcExp / CalcEnergy | 0x134ecb / 0x134ed7 | formula.il.txt |
| Final kill/loss state and int conversion of kill EXP | BattleWorld.OnUnitDead | 0x138c14 | world.il.txt |
| Timer loss and simulated one-second updates | BattleWorld.Step | 0x138de8 | world.il.txt |
| Initial defender count, scheduled count, and time limit | BattleWorld.StartWave | 0x138eec | world.il.txt |
| Deployed regular-unit count includes all placed types | BattleController.GetHeroNum | 0x1374ef | controller.il.txt |
| Ordered ordinary enemies followed by bosses, stage ID join | WaveRuntime.BuildSpawnQueue | 0x13b248 | wave.il.txt |
| Nominal one spawn per second | WaveController.Step / constructor | 0x13b140 / 0x13b125 | wave.il.txt |
| Stage EXP and summoned zero-EXP initialization | BattleFieldDataProvider.GetKillExp / GetMonsterUnitAttr | 0x137fd0 / 0x137dd8 | data.il.txt |
| Summon HP/ATK from owner BaseAttr, not current HP | BattleFieldDataProvider.GetSummonMonsterUnitAttr | 0x137ed8 | data.il.txt |
| Hero attributes do not import raw hero crit | BattleFieldDataProvider.GetHeroUnitAttr | 0x138008 | data.il.txt |
| Active kill EXP multiplier source | BattleUnit.GetKillExp | 0x13c48a | unit.il.txt |
| Paid thresholds and cap | FortressConfig.GetFortressLvByExp | 0x1360d8 | config.il.txt |
| Discard overflow; one level per kill | FortressModel.AddBattlingExp | 0x192614 | model.il.txt |
| Record and reconcile server battle result | FortressModel.OnStagePassed | 0x192a30 | model.il.txt |
| Successful completion after maximum stage | FortressModel nested OnStagePassed callback | 0x2e3658 | callbacks.il.txt |
| Kill-energy gain and capacity clamp | BattleController.OnEnergyChanged | 0x136964 | controller.il.txt |
| Held drain per scaled battle second | BattleController.HandlerUseEnergy | 0x137160 | controller.il.txt |
| Starting energy, full refill | BattleController.SetBeginEnergy / FullEnergy | 0x137286 / 0x1372c0 | controller.il.txt |
| Receive server starting energy | FortressModel.OnBattleStart | 0x192590 | model.il.txt |
| Doubled own attack timer and own buff aging, excluding view 1/99 | BattleUnit.Step | 0x13bba8 | unit.il.txt |
| View 1/99 bypass attack-speed attrs | BattleUnit.GetFinalAtkSpeed | 0x13c42c | unit.il.txt |
| Enemy kill emits energy, including summoned enemies | BattleUnit.TakeDamage | 0x13c008 | unit.il.txt |
| Fixed step defaults to float 0.02 | BattleController constructor | 0x137b3c | controller.il.txt |
| Shield accumulation cap | BattleUnit.UpdateShield | 0x13c378 | unit.il.txt; powerup audit unit-more.il.txt |
| Blitz timed additive ATK stacks | BattleUnit.MonsterAtkAddPerSec_30701 | 0x13be54 | unit.il.txt |
| Fissioner's on-death children | MonsterUnit.Die | 0x13d4b8 | unit.il.txt |
| Supplied priority queue, not verified nearest-unit rule | BattleFieldDataProvider.GetMaxHatredValueHeroTarget | 0x13875c | data.il.txt |
| Crystal/barrier-only formation lacks attacking capacity | FortressUIController.HasFightingCapacityHero | 0x115304 | units-small.il.txt |
| Preparation refresh and >3-unit confirmation | FortressBattlePreBattlePanel.OnRefreshBtnClicked | 0x231ce0 | prebattle-ui.il.txt |
| Start battle and >3-unit confirmation | FortressBattlePreBattlePanel.OnStartBattleBtnClicked | 0x231e3c | prebattle-ui.il.txt |
| Refresh results server-controlled | FortressServiceRmt.PreWarRefresh | 0x19373c | remote.il.txt |
| Talent offers and selection server-controlled | FortressServiceRmt.RandomTalent / SelectTalent | 0x194130 / 0x194270 | remote.il.txt |

`FlightShooting.Data.GameBaseParamConfig.GetFortressRandomBlockCost` is in **Extension_3.dll**, RVA **0x96db**. It reads `fortress_random_hero_cost`, confirming that raw parameter 217's value 15 is the refresh price. `GetFortressInitialEnergy`, `GetFortressEnergyMax`, and `GetFortressEnergyCost` are RVAs 0x96e8, 0x970a, and 0x972c. See `params.il.txt`.

Additional shared evidence from the other independent audits: `BaseBuff.OnAdd` RVA 0x134868 derives shield grants from recipient maximum HP and shield fraction plus the provider's shield bonus; `BattleUnit.UpdateShield` RVA 0x13c378 caps accumulation. Bullet and active executors process exactly one configured skill entry per attack and advance/wrap their index (RVAs 0x1343a8 and 0x134098), which is material for Crimson Wraith's `[20801,20802,20802]` sequence. Those detailed executor and shield claims are also audited by the defender/powerup researchers.

## Raw-table joins and economics

- `fortress_stage` IDs 1–99: all have `monsterNum=50`, `limitedTime=180`. Ordinary ordered blocks precede the boss block. Actual enemy stat key is `100000 + stage × 100 + npcMainId`.
- `fortress_npc_main` IDs 1/2/3 are ordinary visible enemies; 7/8/9 are visible bosses; 4/5 are hidden type-3 summons. No row 6 exists. `fortress_npc_minor` has 329 stage/stat rows.
- Blitz passive `fortress_skill_talent` ID 30701: `atkP +0.02`, timer parameter 0.5, stack cap 50. The runtime tests timer > 0.5 and resets it after applying a stack, so +100% arrives after approximately 25 simulated seconds plus fixed-step effects, not an exact guaranteed wall-clock time. The effect adds percentages rather than compounding 1.02 repeatedly.
- Crimson skill `fortress_skill_bullet` ID 20801: two type-4 summons, 0.5 inherited base ATK and 0.1 inherited base HP; skill 20802 is a coefficient-1 normal shot. Main ID 8 cycles `[20801,20802,20802]`, base cooldown 1.5. Range/target availability controls whether an activation can occur.
- Fission passive `fortress_skill_talent` ID 30901: two type-5 summons, 0.5 inherited base ATK and 0.6 inherited base HP. The children have no split passive.
- `game_base_param` IDs 208–223 cover two daily attempts; item IDs 500–505; initial Cores 15; resurrection bonus 60; refresh price 15; stage-win configured reward 15; initial energy 750; cap 1500; drain 75; modification eligibility maximum level 4; base kill energy 10. Stage rewards and balances are server reconciled; parameter values are described as configured values.
- `_loc_en` keys `fortress_txt_1`, `fortress_txt_8` through `fortress_txt_12`, `fortress_txt_17`, `fortress_txt_24`, `fortress_txt_25`, and the corresponding item descriptions establish carryover, restart/attempt/reset behavior, merge rules, storage persistence, and >3 preparation-unit loss/reset confirmations. Localized rules place the daily reset at 00:00 UTC+8. At event end items 500 and 502–505 are reclaimed and converted to Diamonds; 501 is cleared. No conversion ratio is claimed.
- `fortress_random` authors nested weighted pools, not a locally verified server sampler. Base outer pool 100 totals 1,000: grid expansion pool 100, Wish Crystal pool 18, level-3 ordinary pool 18, level-2 pool 72, level-1 pool 792. Nested pools 2/3/4 each contain seven equal-weight defender entries at levels 3/2/1; pool 5 weights Crystal levels 1/2/3 at 7/2/1. Grid shapes 901–905 are weighted 8/3/3/1/1, with validation flags. Later outer pools move weight from level 1 to level 3. These are authored weights, not unconditional live probabilities; player copy intentionally does not promise those percentages.

## Reproducible calculation checks

Run `/opt/homebrew/bin/python3 output/fortress-rules-audit/derive_and_validate.py` from the workspace root. This validates all table joins, wave totals, boss sequence, late-wave distributions, all 75 XP thresholds, worked arithmetic, section IDs/groups, plain-text content, and table row widths. It writes:

- `waves.json`: complete ordered schedule, joined HP/ATK/move speed, XP, and counts for all 99 stages.
- `bosses.json`: all 27 boss encounter stats.
- `xp-thresholds.json`: every source row, from/to level, threshold, and cumulative retained EXP.
- `xp-simulation.json`: per-stage trace for the explicitly limited, no-bonus, one-clear-per-stage calculation.
- `source-manifest.json`: SHA-256 and sizes for raw source tables, assemblies, IL artifacts, and the derivation script.

Validation result: 99 waves, 27 bosses, 75 paid thresholds, 575,230 EXP to cap. A no-bonus run contains 373,650 potential EXP from 4,950 scheduled kills, retains 372,530 after discarding 1,120 at boundaries, and ends at level 63 with 10,900/15,600 EXP. Summons contribute zero EXP. This is not an assertion about server reconciliation, retries, or additional progression sources.

Worked checks: `750/75=10` boosted battle seconds, `1500/75=20`; stage-99 Fissioner plus children has `5,119,875 + 2 × 3,071,925 = 11,263,725` base HP, and each child has `int(107×0.5)=53` ATK. The generic damage example yields `int(120×2×1×1.25×1.10)=330`.

## Narrow unresolved points

1. The refreshed data and assemblies' exact release/build association is unknown; the repository's older version label must not be used to imply this snapshot's version.
2. Server-controlled refresh legality/weights, talent offer filtering, persistent balances, and battle-result reconciliation have not been live-queried. No unconditional odds, guaranteed cap in one run, or retry farming result is asserted.
3. Enemy selection consumes a supplied defender-priority queue. This audit did not trace the queue's original ranking semantics, so it does not invent a nearest/front-row rule.
4. Timer and attack scheduling use float fixed-step boundaries; pooling and range/target constraints can alter short-lifetime hit counts. No unsupported precise DPS or guaranteed helper-tick total is claimed.
5. The baseline hero crit import is absent from the traced constructor. A different undiscovered source could set a combat critical attribute; no positive baseline is invented.

## Source fingerprints

The complete list is in `output/fortress-rules-audit/source-manifest.json`. Principal inputs:

| Source | SHA-256 |
| --- | --- |
| `data/fortress_attr.json` | `b712d99f106134e7b43ef0ed228fe083a81cf01df7d335f4fb1c2ce4abc053a2` |
| `data/fortress_block.json` | `874d3b3bc259c481481881e86008ed02f5168e977f2e3ba69009d063e0b3b768` |
| `data/fortress_buff.json` | `63f4fa21399631ef80aa6adf7fdde9faebfddb49fd79d9234addf4181fd953a6` |
| `data/fortress_hero.json` | `14af310e5b95e0d0a1b242ea3434658a3db948dd6bc798d680d1d282efacc134` |
| `data/fortress_lv.json` | `21dc424be42f909b03fca4919378e263aed7a0e011cc5754ff047279a1d87d7b` |
| `data/fortress_npc_main.json` | `13df3f19c22881b02513464c5452c3d2f37d2de339ca69665dfdcf21146a579f` |
| `data/fortress_npc_minor.json` | `c530483b0db6c3fb24481deb3aaa6fef317b39d5fd578f597c23a31d5d8ff3ba` |
| `data/fortress_random.json` | `11fc762d72146d800a673ea4c4c2180514cbc15c58d61c29f68b24db5f7347bc` |
| `data/fortress_skill_active.json` | `8f29c72c5ec0b17bdbc6f99a902614964dd4ba7435b820f424d1a5ae15a8b47b` |
| `data/fortress_skill_bullet.json` | `d2d8d4a5a5d4b0b5c5d6774bd4e15554cd27c03afd800db52393934350b6084d` |
| `data/fortress_skill_talent.json` | `6e922324783159e7059631a8c405109f420611d1fdd9e74d6f4d31fd1f84ec47` |
| `data/fortress_stage.json` | `c5d9a4faebf0c1e72bc8a2e59ea19eaf461c19a2fc94424931ecf5b23b97ca99` |
| `data/game_base_param.json` | `dc246858da1a2fd4ae366b3edc11838a9f32d4db89c0947068826b0faa9271f2` |
| `data/_loc_en.json` | `ef622aec6c25190751e7c2ceaf688831840148ffb071307ad05968fe0c1dba0a` |
| `output/crit-cap-audit-2026-09-27/Extension_3.dll` | `169518ab7001173a23b01f689f873678ac8d0602449cc78cec2930f09b528d79` |
| `output/crit-cap-audit-2026-09-27/Extension_4.dll` | `97bcc98eb7cce32de4b13464786b5e39542b9488a66c576d948044cce3c9f973` |
