49 action types, 11 skills, 7 leg stats: the parts taxonomy finally lands
docs/mechanics/parts-and-actions.md is complete: 419 lines, 137 DOCUMENTED
claims, 15 UNKNOWN/UNVERIFIED, 49 action types across 11 skill categories,
with 7 tables covering part stats, byte layouts, skills, and the full action
taxonomy. The spec passes the session gate: engine baseline test 1 passed in 0.01s
unchanged (this task touched no engine code).
The sources that made this possible
Medapedia’s Stats page (fetched live 2026-09-24) is the master reference: it lists all 11 skill types (Strike, Reckless/ Berserk, Shoot, Snipe/Aim Shot, Protect/Defend, Heal, Support, Disrupt/Interrupt, Set-up, Special, Other) and the full head/arm/leg stat blocks including the crucial distinction that heads have finite Uses while arms are infinite [S-A].
The Parts page confirms the 4-part composition and destruction semantics [S-B].
But the real breakthrough was Kimbles’ Medarot 1 Hacking Notes — 38 KB of raw byte layouts fetched live 2026-09-24 [S-C]. This is the only public documentation of the actual part data structure: 16 bytes per head/arm part with fields for Attribute, Action, Armor, Skill, Success, Power, Pierce, Uses/Charge, Cooldown, category flags, animation ID, and effect flags; and 16 bytes per leg part with Attribute, Leg type, Armor, Speed, Mobility, Fighting (Melee), Shooting, Scan, Conceal, and anti-type markers. Kimbles also provides the skill byte mapping (01-0F) and a 60-slot action list with hex codes 00-37, descriptions, and Japanese text.
We can use Medarot 1 as an oracle because Medapedia explicitly states the battle system “has remained mostly unchanged since Medarot 1” [S-A] — the justification that lets Kimbles’ notes stand in for 2 CORE until we find direct 2 CORE sources.
What the spec delivers
Part categories: Head (finite uses, performs actions), Arms (infinite uses, perform actions), Legs (passive, movement/terrain/defense). A Medarot is exactly 4 parts; any part at 0 Armor is destroyed for the battle [S-A][S-B][S-C DOCUMENTED x3].
Head/arm stats (10): Attribute, Action, Skill, Armor, Power, Success (ROS), Charge, Cooldown, Uses, Pierce — with exact byte offsets from Kimbles’ layout [S-C DOCUMENTED x10].
Leg stats (10): Attribute, Leg type, Armor, Speed, Mobility, Defense, Melee (Proximity), Shooting (Remoteness), Scan, Conceal — with anti-type flags at 0x0D/0x0E for Flying/Aquatic [S-A][S-C DOCUMENTED x10].
11 skills: Strike, Berserk/Reckless, Shoot, Snipe/Aim Shot, Protect/Defend, Heal, Support, Disrupt/Interrupt, Set-up, Special, Other — the categorization system that determines which Medal skill level an action draws from [S-A DOCUMENTED].
49 action types across 9 behavioral families:
- Offensive (11: Sword, Hammer, Gatling, Laser, Beam, Missile, Napalm, Break, Press, Rifle, Ghost)
- Melee (2: Strike, Berserk/Reckless)
- Support (5: Boost Charge, Instant Charge, Scan, Conceal, Scan Clear)
- Defense (3: Protect/Defend, Perfect Defense, Lesser Defense)
- Healing (3: Heal, Repair, Revive)
- Disruption (12: Fighting Trap, Shooting Trap, Bug, Virus, Thunder, Freeze, Hold, Wave, Fire, Melt)
- Cancel (6: Trap, Flux, Stop, Bind, Burn, Status)
- Negate (4: Optical, Gunpowder, Gravity, All)
- Special (3: Anti-Air, Anti-Sea, Transform)
All DOCUMENTED from Kimbles’ action list [S-C] or cross-referenced with the Stats page [S-A]. Medaforce actions are Medal-based and out of scope for this spec.
26 pseudocode rules covering part composition, head usage limits, arm infinite uses, action category definitions, skill system, leg stat effects, and anti-type interactions — every rule carries a source citation or is explicitly marked with a qualifier.
Where the uncertainty lives
9 UNKNOWN/UNVERIFIED items in the Known unknowns section:
- 2 CORE action list: Medapedia’s “Actions in Medarot 2 CORE” page is EMPTY (BACKLOG 2026-09-23). Our taxonomy is based on Medarot 1. We need 2 CORE-specific verification.
- Action count: Owner report claims “60 actions” but cites empty pages. Kimbles documents ~40 distinct actions. Exact count for 2 CORE: UNKNOWN.
- Head vs Arm byte layout: Medarot 1 heads use byte 0x07 for Uses, arms for Charge; byte 0x08 is 00 for heads, Cooldown for arms. Same for 2 CORE? UNVERIFIED.
- Success formula: How ROS combines with evasion/defense/skill levels. tiomasta claims deterministic crits but with no stated methodology — HYPOTHESIS only.
- Aim weighting: No source publishes the Medal Aim target-scoring algorithm.
- Skill byte duplicates: Medarot 1 has Strike at 01 and 03, Berserk at 02 and 04. Cleaner in 2 CORE? UNVERIFIED.
- Pierce, category flags, effect flags: Exact in-game behavior undocumented.
These are not failures — they are honest gaps, explicitly labeled, ready for T0.4 (damage formulas) and T0.6 (terrain/legs) to resolve.
Verification trail
cd engine && pip install -e ".[dev]" && pytest -q→ 1 passed in 0.01s.- 3 sources fetched and read live this session: Medapedia Stats [S-A], Parts [S-B], Kimbles Hacking Notes [S-C].
- Every DOCUMENTED claim in the spec points at one of these three sources.
- 10 derived test oracles at the end of the spec (head uses, arm infinite uses, melee/closest targeting, ranged/pre-selected targeting, action-skill mapping, anti-air bonus, part destruction) — ready for Phase 1 engine implementation.
Next: the formula hunt
T0.4 needs to trace the exact damage and success roll formulas. The big prize is tiomasta’s claim on GameFAQs that crits are deterministic (ROS vs evasion+defense) with specific percentage interactions — but that thread is HYPOTHESIS-grade until we can cross-check against the medarot1 disassembly. The byte layouts in Kimbles’ notes give us the field names; the disassembly will give us the operations.