User:HertzDevil/Reverse-engineering notes

• Progress
• Data types

Resonant Battles growth settings

The internal level of all thieves and guardians is the highest level of the player's units in the team, plus a value (lv_offset*) that depends on the player's Interval and chosen battle difficulty:

Interval	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21
Beginner	-8	-8	-7	-7	-6	-6	-5	-5	-4	-3	-2	-1	0	1	2	3	4	5	6	7	8
Intermediate	-4	-4	-3	-3	-2	-2	-1	-1	0	1	2	3	4	5	6	7	8	9	10	11	12
Advanced	0	0	1	1	2	2	3	3	4	5	6	7	8	9	10	11	12	13	14	15	16

For example, entering a Resonant Battle at Advanced difficulty with a Lv. 40 team on Interval 21 results in an internal level of 56. Enemy stats are then calculated from their base stat values and growth rates. After that, thief stats are corrected as follows:

• The HP stat is multiplied by 150%. (multiplier)
• The Def/Res stats are raised to 50% of the unit's internal level if they are lower than that. (min_final_growth)
• A random bonus between 0 and 3 (inclusive) is added to every stat. (random_add)

Guardian stats are corrected as follows:

• The HP and Atk stats are multiplied by 150%.
• The Def/Res stats are raised to 75% of the unit's internal level if they are lower than that.
• A random bonus between 0 and 3 (inclusive) is added to every stat.

Random skills

See also: User:HertzDevil/sandbox/Random skill pool

Training Tower, Tempest Trials, and Allegiance Battles may equip random skills on units.

Game mode	Heroes		Enemy units
	When random skills may be filled	Random skill pool	When random skills may be filled	Random skill pool
Training Tower	Never ~ always (2nd Stratum and below never equip non-weapon skills, 10th Stratum always equip all skills)	Skills owned by the Hero in that category, plus skill on the corresponding unit from the base map (at Normal difficulty), if its tt_inherit_base is true (10th Stratum only), promoted according to derived map rules; all skills with random_mode == 1 if this pool has no legal skills	Never ~ often (2nd Stratum and below never equip non-weapon skills)	All skills with random_mode == 1
Tempest Trials	When the Hero owns any skill from that category (it seems empty slots are also filled at above LV.40)	Skills owned by the Hero in that category; all skills with random_mode == 1 if this pool has no legal skills	N/A	N/A
Allegiance Battles	When the Hero does not own a skill at their rarity, otherwise the Hero equips that skill; Assists are never filled	All skills with random_mode == 1	All non-Assist skills may be filled	All skills with random_mode == 1

Only legal skills from the random pool may be equipped. This means:

• The unit must be able to inherit the skill normally;
• The unit's level is between skill_definition::min_lv and max_lv;
• random_allowed is non-zero;
• random_mode is 1 or 2.

Heroes do not own evolved and refined weapons; additionally, random Heroes cannot equip any weapons they do not already own (note that weapon skills never have tt_inherit_base set).

Growth percentages

The game internally stores growth rates of all Heroes and generic units as percentages. Only multiples of 5% are ever used; however, growth rates are defined for growth percentages that are not multiples of 5 (this has been verified by a debugger). The growth percentage is multiplied by an additional rarity factor:

1	2	3	4	5
0.86	0.93	1.00	1.07	1.14

The truncated result produces the following master growth table:

GP	1	2	3	4	5
-4	0	0	0	0	0
-3	4	4	5	5	5
-2	8	9	10	10	11
-1	12	13	15	16	17
0	17	18	20	21	22
1	21	23	25	26	28
2	25	27	30	32	34
3	30	32	35	37	39
4	34	37	40	42	45
5	38	41	45	48	51
6	43	46	50	53	57
7	47	51	55	58	62
8	51	55	60	64	68
9	55	60	65	69	74
10	60	65	70	74	79
11	64	69	75	80	85
12	68	74	80	85	91
13	73	79	85	90	96

This table is used directly for the randomization of stat growth tables (see below). Stat growth values can be derived from this table by GrowthValue = trunc(0.39 × GrowthPercent), or directly calculated by GrowthValue = trunc(0.39 × trunc((GrowthPoint × 5 + 20) × (0.79 + 0.07 × Rarity))).

Enemy stats for derived maps

Certain game modes (Chain Challenge, Squad Assault, Blessed Gardens) derive their maps from existing story / paralogue maps. Chain Challenge scenarios use base maps at their respective difficulties, whereas the other two game modes always use base maps at Lunatic. Their stats are determined as follows:

• First, if the unit's rarity is increased, then the usual base stat changes for playable Heroes apply, using the base stats defined in either hero_definition or enemy_definition.
• Each stat is increased by trunc((newGP × (newLv - 1) - oldGP × (origLv - 1)) ÷ 100), where origLv is the unit's original level, newLv ≥ origLv is the new level, oldGP is the growth percentage at the unit's old rarity, and newGP ≥ oldGP is the growth percentage at the new rarity. Levels above 40 are defined as follows:

Game mode	Difficulty	Levels
Chain Challenge	Lunatic, single-chapter Story	40/41/42/43/45
	Hard, double-chapter Story	40/40/41/41/43/42/42/43/43/45
	Lunatic, double-chapter Story	45/45/46/46/48/47/47/48/48/50
	Lunatic, single-chapter Paralogue	40/42/45
	Hard, double-chapter Paralogue	40/41/43/42/43/45
	Lunatic, double-chapter Paralogue	45/46/48/47/48/50
Squad Assault	Lunatic	50/50/50/50/50
Blessed Gardens	Lunatic	45
	Infernal	50
Blessed Grounds	Odd-numbered Grounds	45
	Even-numbered Grounds	50
Tempest Trials	Lunatic, Lv. 40	40/41/42/43/44/45/45

A few special cases:

• If only the level is changed, the formula reduces to trunc(GP × (newLv - origLv) ÷ 100).
• If only the rarity is changed, the formula reduces to trunc((newGP - oldGP) × (Lv - 1) ÷ 100).
• If both rarity and level are unchanged, no stat boosts occur at this stage.

• The HP of all enemies is further calculated as trunc(HP × HPFactor), where HPFactor is a global multiplier that depends only on the game mode:

Game mode	Difficulty	HP factor
Chain Challenge	Normal, single-chapter	1.1
	Hard, single-chapter	1.2
	Lunatic, single-chapter	1.3
	Normal, double-chapter Story	1.1 for first half, 1.2 for second half
	Hard, double-chapter Story	1.2 for first half, 1.3 for second half
	Lunatic, double-chapter Story	1.3 for first half, 1.4 for second half
	Normal, double-chapter Paralogue	1.1
	Hard, double-chapter Paralogue	1.2
	Lunatic, double-chapter Paralogue	1.3
Squad Assault	Lunatic	1.3
Blessed Gardens	Normal	0.8
	Hard	0.8
	Lunatic	1.1
	Infernal	1.5
Blessed Grounds	Odd-numbered Grounds	1.1
	Even-numbered Grounds	1.5
Tempest Trials	Normal, Lv. 8	0.90/0.90/0.90
	Normal, Lv. 14	0.90/0.90/0.95
	Normal, Lv. 20	0.95/0.95/1.00
	Hard, Lv. 25	1.00/1.00/1.00/1.05
	Hard, Lv. 30	1.00/1.00/1.05/1.05/1.10
	Lunatic, Lv. 35	1.00/1.05/1.10/1.15/1.20
	Lunatic, Lv. 40	1.00/1.05/1.10/1.15/1.20/1.25/1.30

• Skills are promoted as follows:
  • While skill_definition::next_skill points to a valid skill name, and the unit's rarity is higher than or equal to skill_definition::promotion_rarity, the current skill is replaced by skill_definition::next_skill; this continues until no more promotions can be made. Skills are never demoted, and promotion takes place even if the unit's rarity does not change between the base map and the derived map. Empty skill slots in the base map are never occupied in derived maps.
  • Additionally, only skills with skill_definition::promotion_tier nonzero and also less than or equal to a per-map tier are allowed to promote. Currently this tier number is always 1 on scenarios for derived maps, so Staff weapons, Assists, and Specials are never promoted (see table below).
  • Slight deviations occur for weapon promotion depending on whether the unit is a Hero or an enemy:
    • If a Hero owns the original weapon skill, and the promotion_tier check passes, their default weapon at the new rarity is chosen.
    • If an enemy unit is at 5  and its enemy_definition::top_weapon refers to a valid weapon skill, that skill is chosen, regardless of its promotion_tier value, and regardless of which original weapon skill the unit has (unless it didn't have a weapon).

(Random enemy stats, like those in Tempest Trials and Training Tower, have not been determined. HP factors for Tempest Trials are shown for convenience.)

next_skill exists if and only if the skill is inheritable, not a Sacred Seal, and both promotion_rarity and promotion_tier are nonzero. The table below lists the skill types that have canonical upgrades:

Skills	promotion_rarity	promotion_tier	next_skill
Rank 1 non-staff weapons	3	1	Corresponding rank 2 weapon
Rank 2 non-staff weapons	4	1	Rank 3 weapon with same effect and higher Might
Rank 3 non-staff weapons (excluding generic beast weapons e.g.  Adult (Flier))	5	1	Corresponding rank 4 weapon
Rank 1 staff weapons ( Assault)	3	2	Fear
Rank 2 staff weapons	5	2	Corresponding rank 3 weapon
Rank 1 staff Assists ( Heal)	3	2	Mend
Rank 2 staff Assists ( Mend,  Reconcile)	4	2	Recover and  Martyr respectively
Rank 3 staff Assists	5	2	Corresponding rank 4 Assist
Rank 1 offensive / defensive Specials	4	2	Rank 2 Special with stronger effect but same cooldown count (e.g.  New Moon →  Luna  Moonbow)
Rank 1 Passives (excluding single-rank skills e.g.  Lunge)	4	1	Corresponding rank 2 Passive
Rank 2 Passives	5	1	Corresponding rank 3 Passive

Growth vectors

The file Common/SRPG/Grow.bin.lz contains the seemingly random growth vectors used in the game. The decompressed file contains 2496 XOR-encrypted 64-bit little-endian bit vectors starting from address 0x30, where the n-th bit indicates whether a given stat should increase when the hero reaches level n. An example of reading the vectors directly from the binary file:

#!/usr/bin/env ruby

# this cipher is hardcoded inside libcocos2dcpp.so and appears exactly once
CIPHER = [0x0C, 0xE6, 0xBB, 0x0C, 0x1D, 0x0F, 0x1E, 0xA8]

File.open(File.join('Common', 'SRPG', 'Grow.bin'), 'rb') do |f|
  f.seek 0x30
  (39 * 64).times do |vec_id|
    bytes = f.read(8).bytes
    bytes_xor = bytes.zip(CIPHER).map {|x, y| x ^ y}
    bits = Array.new(64) {|i| bytes_xor[i / 8][i % 8]}
    puts vec_id.to_s + ': ' + bits.join
  end
end

For example, Brave Lyn's neutral growth vectors are: (the used portions are shown in boldface)

Stat	Vector	ID
HP	00100011 01100010 10110110 11100100 01001100 10000000 00000000 00000000	29
Atk	00011111 11001111 01110110 10101101 10101110 00000000 00000000 00000000	26
Spd	00001111 00110111 01011101 11111110 11010100 10000000 00000000 00000000	39
Def	00001010 00000000 01110010 00111111 00010000 00000000 00000000 00000000	00
Res	00011111 00001110 01110010 10011101 01110100 10000000 00000000 00000000	33

Here ID is the lowest 6 bits of the vec_id field, as there are 64 possible variations for each stat growth amount between 1 and 39. This growth vector ID, together with the growth amount determined by the hero's rarity and the stat's GP, uniquely identifies the growth vector used by that stat. It is possible to infer missing entries from the complete stat tables as long as exactly one out of 64 variations correlates with the given entries, or even guess the vector ID if one or two of the given stats are incorrect.

Growth vector IDs change with the hero's IVs: (they warp around 00-63)

Hero	Rarity	Variation	HP	Atk	Spd	Def	Res
Lyn: Brave Lady	5	Bane	23	20	33	58	27
Lyn: Brave Lady	5	Neutral	29	26	39	00	33
Lyn: Brave Lady	5	Boon	35	32	45	05	38

the hero's rarity:

Hero	Rarity	Variation	HP	Atk	Spd	Def	Res
Olivia: Blushing Beauty	1	Neutral	41	20	38	31	35
Olivia: Blushing Beauty	2	Neutral	44	23	42	34	38
Olivia: Blushing Beauty	3	Neutral	48	27	47	38	42
Olivia: Blushing Beauty	4	Neutral	51	30	51	41	45
Olivia: Blushing Beauty	5	Neutral	54	34	55	45	49

and a single byte (hero_definition::base_vector_id) from the structs under Common/SRPG/Person that hold hero data:

Hero	Rarity	Variation	BVID	HP	Atk	Spd	Def	Res
Robin: High Deliverer	5	Neutral	43	15	53	60	03	56
Robin: Mystery Tactician	5	Neutral	BF	11	49	56	63	52
Lucina: Future Witness	5	Neutral	C2	22	02	15	57	46
Marth: Enigmatic Blade	5	Neutral	12	38	18	31	09	62

Deriving growth vector IDs

Growth vector IDs can be calculated as the sum of the following terms, keeping only the 6 lowest bits (i.e. modulo 64):

• 3 × (5-star level 1 neutral base stat);
• A constant offset depending on the stat being calculated for, which increases "randomness":

HP	Atk	Spd	Def	Res
-35	-28	-21	-14	-7

• The growth percentage corresponding to the Hero's rarity and the stat's GP;
• The hero's base vector ID (hero_definition::base_vector_id). It is currently unknown if the highest 2 bits are computationally meaningful since the modulo operation discards them.

Brave Lyn's base vector ID is 0x9D. The growth vector IDs for her neutral stats can be computed as follows:

• HP: (3 × 16 + (−35) + 51 + 157) mod 64 = 221 mod 64 = 29
• Atk: (3 × 7 + (−28) + 68 + 157) mod 64 = 218 mod 64 = 26
• Spd: (3 × 9 + (−21) + 68 + 157) mod 64 = 231 mod 64 = 39
• Def: (3 × 5 + (−14) + 34 + 157) mod 64 = 192 mod 64 = 0
• Res: (3 × 6 + (−7) + 57 + 157) mod 64 = 225 mod 64 = 33

This matches her vector IDs given at the top. This has been verified to match all pre-existing 5-star stat tables on this wiki.

Identical stat growths

Due to the modulo operation it is theoretically possible that two completely different stats on the same hero use the same growth vector: (currently this is limited to HP/Atk base being 7 higher than Def/Res and GPs being the same)

Hero	Stat A	Base	GP	Stat B	Base	GP	GVID	Growth Vector
Chrom: Gifted Leader	Atk Flaw	11	8	Res Asset	6	8	1	01100110 11000111 11100001 11110110 11111101
Tiki: Dragon Scion	HP Flaw	14	7	Def Neutral	8	7	22	00101111 01111011 00111100 01010110 10111100
Tiki: Dragon Scion	HP Neutral	15	8	Def Asset	9	8	28	01001001 11000111 10110011 11111101 11110110

Out-of-bound growth rates

After rarity correction, if the growth rate is 0%, the stat uses a fixed growth vector, regardless of the Hero's base vector ID:

01100000 11001111 10111010 01100001 01110001

This is in fact the XOR cipher used for the growth vectors (see example code above). Because the growth vector files does not store empty vectors, it is possible that the game is accessing memory before the list of growth vectors; if it reads zeroes, the resulting growth vector is simply the XOR value itself. It is also possible that, depending on the Hero's base vector ID, the game may read the file's HSDArc header instead of zeroed memory.

If the growth rate is 100% or above, the integral part is applied to every level up, and the fractional part uses the growth vector mechanism as above. Growth rates can never be below 0%.