i cant code anymore so let ai do it

gameupdate/.gitignore

@@ -0,0 +1,121 @@
+# Ignore all files
+*.*
+
+# File Types
+!*.mps
+!*.json
+!*.txt
+!*.project
+!*.js
+!*.7z
+!*.csv
+!*.ain
+!*.fnl
+!*.ks
+!*.tjs
+!*.yaml
+!*.rb
+!*.rvdata2
+
+# Other Needed Files
+!.gitignore
+!README.md
+!patch-config.txt
+!GameUpdate*
+!patch*
+!Game.dat
+!bsxx.dat*
+!game.ini
+!package.nw
+!SRPG_Unpacker.exe
+!(SRPG_Unpacker Patcher).bat
+
+# Ignore
+previous_patch_sha.txt
+kabe3_save.dat
+kabe3_system.dat
+psbpack.dat
+Save*
+cg.dat
+scene.dat
+BSXScript_*
+
+# Images
+!Layout - コピー (2).png_
+!battleUI_sperm.png_
+!QTE_avoid.png_
+!QTE_avoid2.png_
+!QTE_miss.png_
+!QTE_miss2.png_
+!election.png_
+!name.png_
+!preg_def.png_
+!seiyoku.png_
+!father001_tinpira.png_
+!father002_tinpiraboss.png_
+!father003_goblin.png_
+!father004_ork.png_
+!father005_charao.png_
+!father006_tenta.png_
+!father007_name.png_
+!father008_student.png_
+!father009_wolf.png_
+!father010_tanemaki.png_
+!father011_dog.png_
+!father012_mayor.png_
+!father013_forceman.png_
+!father014_unknown.png_
+!father015_father.png_
+!father016_saimin.png_
+!runa_parents.png_
+!logo2.png_
+!save_window.png_
+!title1.png_
+!title2.png_
+!title3.png_
+!title3_1.png_
+!title4.png_
+!title4_1.png_
+!title4_2.png_
+!title5.png_
+!title5_1.png_
+!trial_end.png_
+!tutorial_01.png_
+!tutorial_02.png_
+!tutorial_02_00.png_
+!tutorial_02_01.png_
+!tutorial_02_02.png_
+!tutorial_02_03.png_
+!tutorial_03.png_
+!tutorial_04.png_
+!tutorial_05.png_
+!tutorial_06.png_
+!tutorial_07.png_
+!tutorial_08.png_
+!tutorial_09.png_
+!tutorial_09_01.png_
+!tutorial_10_00.png_
+!tutorial_10_01.png_
+!tutorial_11_00.png_
+!tutorial_11_01.png_
+!tutorial_12_01.png_
+!tutorial_12_02.png_
+!tutorial_12_03.png_
+!tutorial_13_00.png_
+!tutorial_13_01.png_
+!tutorial_13_02.png_
+!tutorial_14_01.png_
+!tutorial_14_02.png_
+!tutorial_21_01.png_
+!tutorial_21_02.png_
+!tutorial_21_03.png_
+!bar_hentai.png_
+!bar_org.png_
+!bar_preg.png_
+!bar_sex.png_
+!start.png_
+!title_01.png_
+!title_02.png_
+!Command_0.png_
+!Command_1.png_
+!Command_2.png_

gameupdate/gameupdate/patch-config.txt

@@ -0,0 +1,3 @@
+username=dazed-translations
+repo=magical-girls-runa-and-nanami
+branch=main

gameupdate/gameupdate/vocab.txt

@@ -0,0 +1,212 @@
+# MAIN CHARACTERS
+エルミナ (Elmina) - Female; protagonist; elf mage
+フェリス (Feris) - Female; spirit entity; Elmina's guardian & companion
+
+# ALLIES / COMPANIONS
+アンナ (Anna) - Female; witch; lives in 魔女の森; ally
+メープル (Maple) - Female; companion; timid
+ダイヤ (Daiya) - Female; companion; diligent
+トルテ (Torte) - Female; companion; cook/baker
+クレア (Claire) - Female; full name クレアリス・デカート (Clairis Decart); guild-affiliated
+ルイン (Ruin) - Female; Elmina's younger sister (妹); key story motivation
+
+# SIDE / RECURRING CHARACTERS
+バブンスキー (Babunsky) - Male; antagonistic early on, later redeems himself
+ドンフシ (Donfushi) - Male; バブンスキー's younger brother; from プライア村
+ツギカ (Tsugika) - Male; young villager; has feelings for Elmina
+ミゲル (Miguel) - Male; minor NPC; innkeeper or medical role
+
+# ANTAGONISTS
+アナゼマ (Anazema) - Spirit entity; primary antagonist; former co-guardian of 魔界
+ロザリー (Rosalie) - Female; ruler of カロン; arc boss
+キングディブロ (King Diabro) - Boss; Dwarf dungeon; defeated for 水の精霊石
+プーペ (Pupe) - Female; spider-type monster girl (クモ型魔物娘); recurring mini-boss
+魔神 (Demon God) - Entity; final-area antagonist; gave power to アナゼマ
+ビャクダン (Byakudan) - Named boss; Slum dungeon
+# Note: 魔神 = "Demon God"  do NOT render as 'Majin'
+
+# ENEMY / MONSTER TYPES
+スライム (Slime)
+オーク (Orc)
+ゴブリン (Goblin)
+ゴブリン兄弟 (Goblin Brothers)
+こうもり (Bat)
+ミノタウロス (Minotaur)
+魔物娘 (Monster Girl)
+クモ型魔物娘 (Spider-type Monster Girl)  プーペ's species
+
+# LOCATIONS
+## STARTING AREA
+プライア村 (Puraia Village) - Elmina's starting village
+ノストゥスの森 (Nostus Forest) - Forest east of Puraia Village; early dungeon
+
+## KARON CITY
+カロン (Karon) - Major city; mid-game hub; ruled by ロザリー様
+カロン街道 (Karon Road) - Road connecting Puraia Village to Karon
+オプス (Opus) - Karon's high-end inn
+
+## WITCH'S FOREST
+魔女の森 (Witch's Forest) - Anna's domain
+魔女道 (Witch's Road) - Path leading to 魔女の森
+
+## DUNGEONS
+オーク洞窟 (Orc Cave) - Multi-floor orc dungeon
+ロザリー屋敷 (Rosalie's Mansion) - ロザリー's residence; 14+ floors
+
+## DWARF AREA
+ドワーフ街道 (Dwarf Road) - Road from Karon toward Dwarf territory
+ドワーフ隠れ里 (Dwarf Hidden Village) - Dwarf settlement
+ドワーフダンジョン (Dwarf Dungeon) - キングディブロ boss dungeon
+
+## SLUMS
+スラム (Slums) - Poverty district off the Dwarf Road
+娼館 (Brothel / Pleasure House) - Located in the Slums
+
+## ELF VILLAGE
+エルフ里 (Elf Village) - Elmina's homeland; destroyed before the story begins
+
+## SPIRIT STONE DUNGEONS
+精霊石ダンジョン (Spirit Stone Dungeon) - Entrance to three elemental paths
+火 (Fire) - Fire dungeon path
+風 (Wind) - Wind dungeon path
+水 (Water) - Water dungeon path
+
+## DEMON WORLD
+魔界 (Demon World) - The demonic realm; final area
+# Note: render as "Demon World"  NOT "Netherworld" or "Makai"
+
+# WORLD-BUILDING TERMS
+淫紋 (Ero Brand / Lewd Mark) - Contract seal on Elmina's belly; converts desire into magic
+精霊石 (Spirit Stone) - Three elemental key items needed to grant Elmina's wish
+  炎の精霊石 (Fire Spirit Stone)
+  水の精霊石 (Water Spirit Stone)
+  風の精霊石 (Wind Spirit Stone)
+魔力 (Mana / Magical Power)
+ノロイ (Curse) - アナゼマ's seal placed on the Spirit Stones
+冒険者 (Adventurer) - profession/rank system
+ギルド (Guild) - the Adventurers' Guild
+ギルド長 (Guild Master)
+調合師 (Alchemist / Apothecary) - crafts 回復ビン from ingredients
+チャームリンク (Charm Link) - magical bond item between Elmina and a companion
+協会 (Temple / Shrine) - religious institution enshrining the 女神達
+女神 (Goddess) - deities who once saved the continent
+女神達 (Goddesses) - plural; enshrined in the 協会
+魔女 (Witch) - female magic practitioner; distinct from 魔法使い (mage)
+ドワーフ族 (Dwarf Tribe / Dwarf Clan)
+エルフ (Elf) - Elmina's race
+アニキ (Aniki) - "big bro"; ドンフシ's term for バブンスキー
+
+# WEAPON TYPES (Elmina's magic)
+炎の魔法 (Fire Magic) - status ailments; sustained damage
+水の魔法 (Water Magic) - single-hit damage; raises HP and Defense
+風の魔法 (Wind Magic) - multi-hit combos; gains debuffs on growth
+盾の魔法 (Shield Magic) - raises Defense; beginner-recommended
+
+# ARMORS / OUTFITS
+エルミナの普段着 (Elmina's Everyday Clothes)
+絹の肌着 (Silk Undershirt)
+魅惑のバニー (Alluring Bunny)
+メイド服 (Maid Outfit)
+ワンダーウルフ (Wonder Wolf)
+調教服 (Training Outfit / Leash Outfit)
+
+# BATTLE STATES
+戦闘不能 (Incapacitated / KO)
+防御 (Guard / Defending)
+不死身 (Immortal)
+毒 (Poison)
+暗闇 (Blind / Darkness)
+沈黙 (Silence)
+激昂 (Enrage / Fury)
+混乱 (Confusion)
+魅了 (Charm / Enthralled)
+睡眠 (Sleep)
+
+# KEY ITEMS
+怪しげな薬 (Suspicious Medicine)
+冒険者登録証 (Adventurer's Registration Card)
+トルテのお弁当 (Torte's Lunchbox)
+回復ビン (Recovery Vial) - upgrades +1 through +15
+屋敷のカギ (Mansion Key)
+力のグローブ (Power Gloves)
+爆音袋 (Sonic Bag / Bang Bag)
+子猫の入ったカバン (Bag Containing a Kitten)
+チャームリンク・メープル (Charm Link: Maple) - +Defense
+チャームリンク・アンナ (Charm Link: Anna) - +Magic
+チャームリンク・ダイヤ (Charm Link: Daiya) - +Defense
+
+## Crafting Ingredients (素材)
+緑草 (Green Grass)
+黄草 (Yellow Grass)
+赤草 (Red Grass)
+黒草 (Black Grass)
+ロゼの花 (Roze Flower)
+オルラーヤ (Orlaya)
+リンドウ (Gentian)
+ベニダケ (Red Mushroom)
+サボテンの果実 (Cactus Fruit)
+輝くゼンマイ (Glowing Bracken)
+触手の若芽 (Tentacle Sprout)
+水 (Water)
+純水 (Pure Water)
+魔力水 (Mana Water)
+
+## Guild Exchange Drops (ギルド換金)
+スライムの核 (Slime Core)
+スライムの巨大核 (Slime Giant Core)
+スライムの赤核 (Slime Red Core)
+スライムの赤粘膜 (Slime Red Membrane)
+オークの牙 (Orc Fang)
+獣の毛 (Beast Fur)
+焔心核 (Flame Heart Core)
+蜘蛛の糸 (Spider Thread)
+悪魔の角 (Demon Horn)
+海魔の墨 (Kraken Ink)
+悪魔の羽根 (Demon Feather)
+死霊の灰 (Wraith Ash)
+
+## Mana Crystal EXP Items
+小さな魔力結晶 (Small Mana Crystal) - EXP 10
+魔力結晶 (Mana Crystal) - EXP 100
+大きな魔力結晶 (Large Mana Crystal) - EXP 1,000
+巨大な魔力結晶 (Giant Mana Crystal) - EXP 10,000
+
+## Boss Mana Cores (素材)
+プーペの魔力核 (Pupe's Mana Core)
+ビャクダンの魔力核 (Byakudan's Mana Core)
+キングディブロの魔力核 (King Diabro's Mana Core)
+ロザリーの魔力核 (Rosalie's Mana Core)
+タイタンワームの魔力結晶 (Titan Worm Mana Crystal)
+
+# ROMANISATION CONVENTIONS
+エルミナ (Elmina) - not Erumina; matches title
+フェリス (Feris) - matches sprite file "feris1"
+アンナ (Anna)
+メープル (Maple)
+ダイヤ (Daiya) - or Dia; avoid "Diamond" in dialogue
+バブンスキー (Babunsky)
+ドンフシ (Donfushi)
+ツギカ (Tsugika)
+トルテ (Torte) - German: layer-cake
+クレア (Claire)
+クレアリス・デカート (Clairis Decart) - both parts are proper nouns
+アナゼマ (Anazema)
+ルイン (Ruin) - or Luin
+プーペ (Pupe) - or Poupée; French: doll
+ミゲル (Miguel)
+ロザリー (Rosalie)
+キングディブロ (King Diabro)
+ビャクダン (Byakudan)
+カロン (Karon)
+プライア村 (Puraia Village) - keep "Puraia" per map file names
+ノストゥスの森 (Nostus Forest)
+ドワーフ (Dwarf)
+エルフ (Elf)
+魔界 (Demon World) - NOT "Netherworld"; keep distinct from 地獄 "Hell"
+淫紋 (Ero Brand) - or "Lewd Mark"; keep consistent within project
+精霊石 (Spirit Stone)
+ノロイ (Curse)
+魔神 (Demon God) - NOT "Majin"; translate fully
+調合師 (Alchemist) - or Apothecary; be consistent
+ギルド長 (Guild Master)
+アニキ (Aniki) - or "Big Bro"; keep consistent per character

gui/workflow_tab.py

@@ -1,14 +1,15 @@
-"""
+"""
 RPGMaker Workflow Tab - Automation hub for the full translation pipeline.
 
 Provides a guided, step-by-step interface:
 
   Step 0  – Select game project folder and import data files into files/
-  Step 1  – Edit vocab.txt (glossary) inline
+  Step 1  – (Optional) Pre-process game files
-  Step 2  – Actor variable substitution  (\\n[X] ↔ names)
+  Step 2  – Auto-detect speaker format and apply to module settings
-  Step 3  – Auto-detect speaker format and apply to module settings
+  Step 3  – Build glossary: parse speakers, then enrich with AI prompt
-  Step 4  – Phase 1 (safe dialogue codes) and Phase 2 (risky codes) translation
+  Step 4  – Actor variable substitution  (\\n[X] ↔ names)
-  Step 5  – Export translated/ back to the game folder
+  Step 5  – Phase 1 (safe dialogue codes) and Phase 2 (risky codes) translation
+  Step 6  – Export translated/ back to the game folder
 """
 
 from __future__ import annotations
@@ -417,12 +418,12 @@ class WorkflowTab(QWidget):
         vbox.addWidget(_make_hr())
         self._build_preprocess(vbox)
         vbox.addWidget(_make_hr())
+        self._build_step3(vbox)
+        vbox.addWidget(_make_hr())
         self._build_step1(vbox)
         vbox.addWidget(_make_hr())
         self._build_step2(vbox)
         vbox.addWidget(_make_hr())
-        self._build_step3(vbox)
-        vbox.addWidget(_make_hr())
         self._build_step4(vbox)
         vbox.addWidget(_make_hr())
         self._build_step6(vbox)
@@ -543,29 +544,54 @@ class WorkflowTab(QWidget):
     _PROMPT_GLOSSARY = (
         "You are helping me build a complete translation glossary for a Japanese RPGMaker game.\n"
         "\n"
-        "Please scan ALL of these game files (Actors.json, CommonEvents.json, Map*.json, "
+        "⚠️ FILE SIZE WARNING: Map files and CommonEvents.json can be extremely large "
-        "Troops.json, and any other available files) and produce TWO sections:\n"
+        "(hundreds of thousands of lines). Do NOT attempt to read them in full — you will "
+        "hit context limits and miss content. Instead, use this strategy:\n"
+        "\n"
+        "  1. Read the small structured DB files IN FULL first — these are the richest "
+        "source of names and are always small:\n"
+        "     Actors.json, Classes.json, Troops.json, Skills.json, Items.json, "
+        "Armors.json, Weapons.json, States.json, System.json\n"
+        "\n"
+        "  2. For large files (CommonEvents.json, Map*.json), do NOT read sequentially. "
+        "Instead, SEARCH (grep) for:\n"
+        "     - Named character patterns: 【, \\\\n<, \\\\k< at the start of dialogue lines\n"
+        "     - Unique proper nouns: capitalised katakana clusters or kanji compound nouns "
+        "in \"message\" or \"parameters\" fields\n"
+        "     Scan Map001.json through Map010.json at most — early maps have the most "
+        "story-critical dialogue.\n"
+        "\n"
+        "  3. Stop adding entries once you stop finding new names/terms — do not pad.\n"
+        "\n"
+        "Produce TWO sections of output.\n"
+        "\n"
+        "Output the results EXACTLY in the format shown below, including the category headers "
+        "starting with #. Do not add any other text, preamble, or explanation outside the entries.\n"
         "\n"
         "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n"
-        "SECTION 1 — NAMED CHARACTERS\n"
+        "# Game Characters\n"
         "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n"
         "Identify every NAMED CHARACTER that appears in dialogue or descriptions.\n"
         "For each character provide:\n"
         "  - Japanese name (katakana/kanji as it appears in-game)\n"
         "  - English transliteration or translation\n"
         "  - Gender (Male / Female / Unknown) — infer from speech patterns or pronouns\n"
-        "  - Brief role note (e.g. protagonist, antagonist, NPC, merchant)\n"
+        "  - Role (protagonist, antagonist, NPC, etc.)\n"
+        "  - Speech register and personality notes — how they speak, their tone, any nicknames, "
+        "whether their name is player-chosen, etc.\n"
         "\n"
-        "Format — one entry per line:\n"
+        "Format — the header line, then one entry per line:\n"
-        "山田太郎 (Yamada Taro) - Male; protagonist; high-school student\n"
+        "# Game Characters\n"
-        "浅井花子 (Asai Hanako) - Female; childhood friend; lives in 栄町\n"
+        "\u30b7\u30ed (Shiro) - Female; protagonist; player-controlled (Actors.json ID 1); "
+        "speaks in a flustered, cute register with feminine speech markers; nickname \u30d0\u30ab\u732b\u3002\n"
+        "\u30af\u30ed\u30cd (Kurone) - Female; antagonist; cold and terse; speaks in short cutting sentences\n"
         "\n"
         "Rules:\n"
         "  - Named characters only — no generic enemy types or unnamed NPCs.\n"
         "  - If a character has a player-chosen name (e.g. Actors.json id 1), note it explicitly.\n"
         "\n"
         "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n"
-        "SECTION 2 — WORLDBUILDING TERMS\n"
+        "# Worldbuilding Terms\n"
         "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n"
         "Identify lore-specific terms that appear in dialogue, descriptions, or narration "
         "but do NOT have a dedicated database file — i.e. terms the translation tool will NOT "
@@ -573,24 +599,53 @@ class WorkflowTab(QWidget):
         "\n"
         "Target specifically:\n"
         "  - Faction / organisation names (kingdoms, guilds, cults, nations)\n"
-        "  - Location names that are mentioned in dialogue but are not map titles\n"
+        "  - Location names mentioned in dialogue but not map titles\n"
         "  - Unique magic systems, schools of magic, or power classifications\n"
         "  - Lore titles and honorifics unique to this setting\n"
-        "  - Recurring in-universe concepts, events, or proper nouns with no English equivalent\n"
+        "  - Recurring in-universe concepts or proper nouns with no English equivalent\n"
         "\n"
-        "For each term provide:\n"
+        "Format — the header line, then one entry per line:\n"
-        "  - Japanese original (as it appears in-game)\n"
+        "# Worldbuilding Terms\n"
-        "  - Recommended English translation\n"
+        "\u9b54\u4e16\u754c (Demon World) - The demonic realm referenced in NPC dialogue; not a named map\n"
-        "  - Brief note explaining what it is\n"
+        "\u8056\u5263\u6559\u56e3 (Holy Blade Order) - Antagonist faction controlling the eastern territories\n"
-        "\n"
-        "Format — one entry per line:\n"
-        "魔世界 (Demon World) - The demonic realm referenced in NPC dialogue; not a named map\n"
-        "聖剣教団 (Holy Blade Order) - Antagonist faction controlling the eastern territories\n"
         "\n"
         "Rules:\n"
         "  - Do NOT list skill names, item names, weapon names, armour names — the tool handles those.\n"
-        "  - Skip generic RPG words (ポーション, レベル, ステータス, etc.).\n"
+        "  - Skip generic RPG words (\u30dd\u30fc\u30b7\u30e7\u30f3, \u30ec\u30d9\u30eb, \u30b9\u30c6\u30fc\u30bf\u30b9, etc.).\n"
-        "  - Do NOT repeat character names here — those belong in Section 1."
+        "  - Do NOT repeat character names here.\n"
+        "\n"
+        "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n"
+        "COMPLETE EXAMPLE OUTPUT\n"
+        "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n"
+        "Below is what a correct, well-formed response looks like.\n"
+        "Your output should follow this structure exactly:\n"
+        "\n"
+        "```\n"
+        "# Game Characters\n"
+        "アリア (Aria) - Female; protagonist; player-chosen name (Actors.json ID 1); "
+        "speaks cheerfully in casual feminine speech; nicknamed アリアちゃん by her sister\n"
+        "ゼクス (Zex) - Male; antagonist; cold and commanding; addresses others with contempt; "
+        "uses archaic formal register\n"
+        "カナエ (Kanae) - Female; NPC shopkeeper; warm and motherly; ends sentences with わね\n"
+        "\n"
+        "# Worldbuilding Terms\n"
+        "虚無の穴 (Void Rift) - Dimensional tear referenced repeatedly in Act 2 NPC dialogue; "
+        "not a named map location\n"
+        "鋼の誓約 (Iron Vow) - Sacred oath-binding ritual unique to the knightly order; "
+        "appears in story cutscenes\n"
+        "裁定者 (Arbiter) - Title held by the ruling council; lore-specific rank with no "
+        "real-world equivalent\n"
+        "```\n"
+        "\n"
+        "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n"
+        "GLOBAL RULES (apply to both sections)\n"
+        "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n"
+        "  - NEVER give two options for any term (e.g. 'Sylfia / Sylphia' is wrong). "
+        "Always commit to a single best translation. If multiple transliterations exist, "
+        "pick the most etymologically accurate or natural-sounding one and use only that.\n"
+        "  - Use a plain hyphen-minus (-) as the separator between the Japanese entry and "
+        "its description. Never use an em dash (\u2014) or en dash (\u2013) \u2014 the "
+        "translation tool only recognises the plain hyphen."
     )
 
     # ── Step 1 (Optional): Pre-process ────────────────────────────────
@@ -693,8 +748,8 @@ class WorkflowTab(QWidget):
         tc_inner = QVBoxLayout(tc)
         tc_inner.setSpacing(4)
         tc_desc = QLabel(
-            "Copies everything from the <code>gameupdate/</code> folder in the game root "
+            "Copies everything from the <code>gameupdate/</code> folder "
-            "into the game\'s data folder, overwriting existing files."
+            "into the game\'s root folder, overwriting existing files."
         )
         tc_desc.setTextFormat(Qt.RichText)
         tc_desc.setWordWrap(True)
@@ -719,7 +774,7 @@ class WorkflowTab(QWidget):
 
         tc_dst_row = QHBoxLayout()
         tc_dst_row.addWidget(QLabel("Destination:"))
-        self.pp_gameupdate_dst_label = QLabel("(data folder auto-filled from project)")
+        self.pp_gameupdate_dst_label = QLabel("(game root folder auto-filled from project)")
         self.pp_gameupdate_dst_label.setStyleSheet("color:#888;font-size:10px;")
         tc_dst_row.addWidget(self.pp_gameupdate_dst_label, 1)
         tc_inner.addLayout(tc_dst_row)
@@ -752,7 +807,7 @@ class WorkflowTab(QWidget):
         )
 
     def _build_step1(self, layout: QVBoxLayout):
-        layout.addWidget(_make_section_label("Step 2 — Glossary (vocab.txt)"))
+        layout.addWidget(_make_section_label("Step 3 — Glossary (vocab.txt)"))
         hint = QLabel(
             "Edit your glossary below. Character names, genders, and worldbuilding terms "
             "here are injected into every AI prompt for consistent terminology."
@@ -761,8 +816,43 @@ class WorkflowTab(QWidget):
         hint.setStyleSheet("color:#888;font-size:10px;padding-bottom:4px;")
         layout.addWidget(hint)
 
+        # ---- Parse Speakers -------------------------------------------------
+        spk_box = QGroupBox("2a — Parse Speakers (Auto-detect Names)")
+        spk_box.setStyleSheet(
+            "QGroupBox{color:#ccc;border:1px solid #444;border-radius:3px;"
+            "margin-top:8px;font-size:11px;}"
+            "QGroupBox::title{padding:0 6px;}"
+        )
+        spk_inner = QVBoxLayout(spk_box)
+        spk_inner.setSpacing(6)
+
+        spk_hint = QLabel(
+            "Scans every file in <b>files/</b> and extracts all detected speaker names "
+            "into a <b># Speakers</b> section of vocab.txt. "
+            "Run this first — the AI prompt below will then work from the pre-populated "
+            "list and can enrich entries with gender, role, and speech register."
+        )
+        spk_hint.setTextFormat(Qt.RichText)
+        spk_hint.setWordWrap(True)
+        spk_hint.setStyleSheet("color:#888;font-size:10px;")
+        spk_inner.addWidget(spk_hint)
+
+        spk_row = QHBoxLayout()
+        self._parse_speakers_status = QLabel("")
+        self._parse_speakers_status.setStyleSheet("color:#aaa;font-size:10px;")
+        spk_row.addWidget(self._parse_speakers_status, 1)
+        self._parse_speakers_btn = _make_btn("🔍  Parse Speakers", "#007acc")
+        self._parse_speakers_btn.setToolTip(
+            "Scan all game files for speaker names and write them to vocab.txt"
+        )
+        self._parse_speakers_btn.clicked.connect(self._run_parse_speakers)
+        spk_row.addWidget(self._parse_speakers_btn)
+        spk_inner.addLayout(spk_row)
+
+        layout.addWidget(spk_box)
+
         # ---- Copilot / Cursor prompt helpers --------------------------------
-        prompt_box = QGroupBox("AI Prompt Helpers (Copilot / Cursor)")
+        prompt_box = QGroupBox("2b — AI Prompt Helpers (Copilot / Cursor)")
         prompt_box.setStyleSheet(
             "QGroupBox{color:#ccc;border:1px solid #444;border-radius:3px;"
             "margin-top:8px;font-size:11px;}"
@@ -772,8 +862,10 @@ class WorkflowTab(QWidget):
         pb_inner.setSpacing(6)
 
         prompt_hint = QLabel(
-            "Copy a prompt below, paste it into GitHub Copilot Chat or Cursor with your "
+            "After parsing speakers above, copy the prompt below and paste it into "
-            "game files open, then paste the AI’s output into vocab.txt."
+            "GitHub Copilot Chat or Cursor with your game files open. "
+            "The AI will enrich the speaker list with gender, role, and speech register "
+            "and add worldbuilding terms. Paste the AI's output back into vocab.txt."
         )
         prompt_hint.setWordWrap(True)
         prompt_hint.setStyleSheet("color:#888;font-size:10px;")
@@ -800,8 +892,10 @@ class WorkflowTab(QWidget):
         # ---- vocab.txt editor -----------------------------------------------
         layout.addWidget(_make_section_label("vocab.txt editor"))
         format_hint = QLabel(
-            "Expected line format:   Japanese (English) - Gender; role; notes\n"
+            "Put character entries under  # Game Characters  — they are always sent to the AI.\n"
-            "Example:  山田太郎 (Yamada Taro) - Male; protagonist; village blacksmith"
+            "Format:  Japanese (English) - Gender; role; speech register / personality notes\n"
+            "Example:  シロ (Shiro) - Female; protagonist; speaks in a flustered, cute register with feminine speech markers\n"
+            "Universal terms (honorifics, elements, etc.) live in vocab_base.txt and are auto-appended on every save."
         )
         format_hint.setFont(QFont("Consolas", 9))
         format_hint.setStyleSheet(
@@ -834,7 +928,7 @@ class WorkflowTab(QWidget):
     # ── Step 2: Actor Variables ─────────────────────────────────────────────
 
     def _build_step2(self, layout: QVBoxLayout):
-        layout.addWidget(_make_section_label("Step 3 — Actor Variable Substitution"))
+        layout.addWidget(_make_section_label("Step 4 — Actor Variable Substitution"))
         hint = QLabel(
             "Replaces  \\n[X]  RPGMaker name variables with actual actor names "
             "so the AI has proper context. After translation the variables are "
@@ -872,7 +966,7 @@ class WorkflowTab(QWidget):
     # ── Step 3: Speaker Detection ───────────────────────────────────────────
 
     def _build_step3(self, layout: QVBoxLayout):
-        layout.addWidget(_make_section_label("Step 4 — Speaker Format Detection"))
+        layout.addWidget(_make_section_label("Step 2 — Speaker Format Detection"))
         hint = QLabel(
             "Scans map files to determine how speaker names are embedded in dialogue "
             "and automatically sets the correct INLINE401SPEAKERS / FIRSTLINESPEAKERS "
@@ -1099,11 +1193,69 @@ class WorkflowTab(QWidget):
     # Step 1 – Vocab
     # ─────────────────────────────────────────────────────────────────────────
 
+    def _run_parse_speakers(self):
+        """Launch a speaker-parse pass over files/ and write results to vocab.txt."""
+        try:
+            from gui.translation_tab import TranslationWorker
+        except Exception as exc:
+            self._log(f"Could not import TranslationWorker: {exc}")
+            return
+
+        project_root = Path(__file__).parent.parent
+        files_dir = project_root / "files"
+        all_json = sorted(
+            p.name for p in files_dir.glob("*.json") if p.name != ".gitkeep"
+        ) if files_dir.exists() else []
+
+        if not all_json:
+            self._log("No JSON files found in files/. Run Step 1 (import) first.")
+            self._parse_speakers_status.setText("No files found in files/.")
+            return
+
+        self._log(f"Parsing speakers from {len(all_json)} file(s)...")
+        self._parse_speakers_btn.setEnabled(False)
+        self._parse_speakers_status.setText("Running...")
+
+        module_info = ["RPG Maker MV/MZ", [".json"], None]
+        worker = TranslationWorker(
+            project_root, module_info, estimate_only=False,
+            selected_files=all_json,
+            parse_speakers=True,
+        )
+        worker.log_signal.connect(self._log)
+        worker.progress_signal.connect(
+            lambda cur, tot, fn: self._parse_speakers_status.setText(
+                f"[{cur}/{tot}] {Path(fn).name}"
+            )
+        )
+        worker.finished_signal.connect(self._on_parse_speakers_done)
+        self._worker = worker
+        worker.start()
+
+    def _on_parse_speakers_done(self, ok: bool, msg: str):
+        self._parse_speakers_btn.setEnabled(True)
+        if ok:
+            self._parse_speakers_status.setText(
+                "Done — # Speakers section written to vocab.txt"
+            )
+            self._reload_vocab()
+            self._log("Speaker parse complete. vocab.txt updated with # Speakers.")
+        else:
+            self._parse_speakers_status.setText(f"Failed: {msg}")
+            self._log(f"Speaker parse failed: {msg}")
+
+    _BASE_SEPARATOR = "# ── Base Vocabulary (auto-appended from vocab_base.txt — do not edit below) ──\n"
+
     def _reload_vocab(self):
         vocab_path = Path("vocab.txt")
         try:
             if vocab_path.exists():
-                self.vocab_editor.setPlainText(vocab_path.read_text(encoding="utf-8"))
+                text = vocab_path.read_text(encoding="utf-8")
+                # Strip the auto-appended base section so editor shows only game-specific content
+                sep_idx = text.find(self._BASE_SEPARATOR)
+                if sep_idx != -1:
+                    text = text[:sep_idx].rstrip("\n")
+                self.vocab_editor.setPlainText(text)
             else:
                 self.vocab_editor.setPlainText("# Add character glossary entries here\n")
         except Exception as exc:
@@ -1111,10 +1263,12 @@ class WorkflowTab(QWidget):
 
     def _save_vocab(self):
         try:
-            Path("vocab.txt").write_text(
+            game_text = self.vocab_editor.toPlainText().rstrip("\n")
-                self.vocab_editor.toPlainText(), encoding="utf-8"
+            base_path = Path("vocab_base.txt")
-            )
+            base_text = base_path.read_text(encoding="utf-8") if base_path.exists() else ""
-            self._log("✅ vocab.txt saved.")
+            combined = game_text + "\n\n" + self._BASE_SEPARATOR + base_text
+            Path("vocab.txt").write_text(combined, encoding="utf-8")
+            self._log("✅ vocab.txt saved (base terms from vocab_base.txt appended).")
         except Exception as exc:
             self._log(f"❌ Could not save vocab.txt: {exc}")
 
@@ -1442,7 +1596,7 @@ class WorkflowTab(QWidget):
         except Exception:
             pass
         try:
-            self.pp_gameupdate_dst_label.setText(data_path or "(no data folder detected)")
+            self.pp_gameupdate_dst_label.setText(game_root or "(no game folder detected)")
         except Exception:
             pass
 
@@ -1488,12 +1642,12 @@ class WorkflowTab(QWidget):
 
     def _run_gameupdate(self):
         src = self.pp_gameupdate_edit.text().strip()
-        dst = self._data_path
+        dst = self.folder_edit.text().strip()
         if not src:
             self._log("⚠  No gameupdate folder path set.")
             return
         if not dst:
-            self._log("⚠  No data folder detected. Complete Step 0 first.")
+            self._log("⚠  No game root folder set. Complete Step 0 first.")
             return
         if not Path(src).is_dir():
             self._log(f"⚠  gameupdate folder not found: {src}")
@@ -1534,14 +1688,15 @@ class WorkflowTab(QWidget):
             skipped.append(f"B (format plugins.js): {reason}")
 
         # C — gameupdate copy
-        if gameupdate_src and Path(gameupdate_src).is_dir() and data_path:
+        game_root_dst = self.folder_edit.text().strip()
+        if gameupdate_src and Path(gameupdate_src).is_dir() and game_root_dst:
             self._log("▶ [C] gameupdate copy …")
             self._run_gameupdate()
         else:
             if not gameupdate_src or not Path(gameupdate_src).is_dir():
                 reason = f"source not found ({gameupdate_src or 'not set'})"
             else:
-                reason = "data folder missing"
+                reason = "game root folder missing"
             skipped.append(f"C (gameupdate): {reason}")
 
         for msg in skipped:

modules/rpgmakermvmz.py

@@ -74,7 +74,7 @@ LEAVE = False
 # FIRSTLINESPEAKERS: Guess speaker from first line.
 FIRSTLINESPEAKERS = False
 # INLINE401SPEAKERS: Extract speaker from "Name「dialogue」" inline format on 401 lines.
-INLINE401SPEAKERS = True
+INLINE401SPEAKERS = False
 # FACENAME101: Map face name -> speaker.
 FACENAME101 = False
 # Face name -> speaker mapping for FACENAME101.

prompt.txt

@@ -24,10 +24,13 @@ You will be translating erotic and sexual content. You will receive lines of dia
 
 ## Characters and Pronouns
 
-- The **"# Game Characters"** section lists character names, nicknames, and genders. Use it to resolve any ambiguity.
+- The **"# Game Characters"** section contains character entries. Each entry may include the character's name, nicknames, gender, role, personality, and speech register notes. Read every entry carefully and apply all of it.
+- **Name & spelling:** Always use the English name given in the entry. Never invent a different romanisation.
+- **Gender:** Use the stated gender when resolving pronouns and コイツ / あいつ / こいつ references.
+- **Speech register:** If the entry describes how a character speaks (flustered, blunt, formal, childlike, crude, etc.), mirror that register in their English dialogue. A character described as speaking in a "cute, flustered register" should sound different from one described as "cold and terse".
+- **Role & context:** Use the role/personality notes to inform tone — a villain's lines should feel threatening, a comic-relief NPC's lines should feel goofy, etc.
 - Japanese omits pronouns constantly. Infer the correct subject and pronoun from context, translation history, and the character list.
 - Third-person pronouns (彼, 彼女, あいつ, こいつ, そいつ, コイツ) should match the known gender of the person being referenced.
-- Speech register matters: rough masculine speech (e.g., 俺, だ/だろ) → blunt or casual English. Soft feminine speech (e.g., 私, わ/のよ) → softer, more polite English.
 - Translate **コイツ** as "this bastard" (male) or "this bitch" (female) depending on the referenced character's gender.
 
 ---

util/speaker_detector.py

@@ -1,15 +1,20 @@
 """
 Speaker Format Detector for RPGMaker MV/MZ
 
-Scans map / event files and scores three detection modes:
+Mirrors the detection priority of rpgmakermvmz.py searchCodes() exactly:
 
-  INLINE401SPEAKERS  – 401 lines contain  Name「dialogue」
+  Pass 1 — scan 401/405 codes in order:
-  FIRSTLINESPEAKERS  – a short (< 40 char) 401 followed by 401/405 starting
+    1. \\n<Name> / \\k<Name> inline nametag codes   (always active, no flag needed)
-                        with  「  "  (  （  *  [
+    2. 【Name】 alone on a 401 line                  (always active, no flag needed)
-  FACENAME101        – a 101 code (Show Text / face cmd) immediately precedes
+    3. 【Name】dialogue on same 401 line             (always active, no flag needed)
-                       a 401 block
+    4. Name「dialogue」 inline quote                 -> INLINE401SPEAKERS
+    5. Short 401 (<40 chars) followed by 401 whose
+       text starts with 「 " ( （ * [               -> FIRSTLINESPEAKERS
 
-Returns the best mode and confidence scores for each.
+  Pass 2 — only if Pass 1 produced no reliable hits:
+    6. 101 code param[0] is a non-empty name string  -> FACENAME101
+
+Returns the best mode and confidence scores.
 """
 
 from __future__ import annotations
@@ -18,13 +23,28 @@ import json
 import re
 from pathlib import Path
 
-# Regex used to detect inline quote speaker: "Name「..." at start of 401 text
+# ── Regexes matching rpgmakermvmz.py exactly ────────────────────────────────
-_INLINE_RE = re.compile(r"^([^\s「」。、！？…\\\n]{1,20})「")
 
-# Characters that signal the next 401 is dialogue (first-line speaker heuristic)
+# \\n<Name> / \\k<Name> nametag codes (always active in module)
+_NAMETAG_RE = re.compile(
+    r"[\\]+[kKnN][wWcCrRrEe]?[\[<](?:[\\]*\w\[\d+\])?(.*?)(?:[\\]*\w\[\d+\])?[>]"
+)
+
+# 【Name】 alone on the line (with optional trailing control codes)
+_BRACKET_ALONE_RE = re.compile(
+    r"^\s*【[^】]+】(?:\s*|(?:[\\]+[A-Za-z]+(?:\[(?:[^\[\]]|\[[^\]]*\])*\])+\s*)*)$"
+)
+
+# 【Name】dialogue on the same line
+_BRACKET_INLINE_RE = re.compile(r"^\s*【([^】]+)】(.+)", re.DOTALL)
+
+# Inline quote: Name「dialogue…
+_INLINE_QUOTE_RE = re.compile(r"^([^\s「」。、！？…\\\n]{1,20})「")
+
+# Dialogue starters that follow a FIRSTLINESPEAKERS name line
 _DIALOGUE_STARTERS = ("「", '"', "(", "（", "*", "[")
 
-# Minimum number of qualifying events to consider a result reliable
+# Minimum hits to trust a result
 _MIN_HITS = 3
 
 
@@ -32,41 +52,46 @@ def detect_speaker_format(
     files_dir: str | Path = "files",
     sample_size: int = 20,
 ) -> dict:
-    """Scan up to *sample_size* map files and score speaker detection modes.
+    """Scan up to *sample_size* map files and determine the speaker format.
 
     Returns:
     {
-        "best_mode": "INLINE401SPEAKERS" | "FIRSTLINESPEAKERS" | "FACENAME101" | "NONE",
+        "best_mode": "INLINE401SPEAKERS" | "FIRSTLINESPEAKERS" | "FACENAME101"
+                     | "ALWAYS_ON" | "NONE",
         "scores": {
-            "INLINE401SPEAKERS":  int,   # raw hit count
+            "nametag_codes":     int,   # \\n<Name> hits — always handled, no flag
-            "FIRSTLINESPEAKERS":  int,
+            "bracket_401":       int,   # 【Name】 hits  — always handled, no flag
-            "FACENAME101":        int,
+            "INLINE401SPEAKERS": int,
+            "FIRSTLINESPEAKERS": int,
+            "FACENAME101":       int,
         },
-        "total_401_groups": int,   # total dialogue groups examined
+        "total_401_groups": int,
         "files_scanned": int,
-        "recommended_config": {    # ready to apply to rpgmakermvmz.py
+        "recommended_config": {
-            "FIRSTLINESPEAKERS": bool,
             "INLINE401SPEAKERS": bool,
+            "FIRSTLINESPEAKERS": bool,
             "FACENAME101":       bool,
         },
         "confidence": "high" | "medium" | "low",
-        "note": str,               # human-readable explanation
+        "note": str,
     }
     """
     files_dir = Path(files_dir)
     scores = {
+        "nametag_codes":     0,
+        "bracket_401":       0,
         "INLINE401SPEAKERS": 0,
         "FIRSTLINESPEAKERS": 0,
-        "FACENAME101": 0,
+        "FACENAME101":       0,
     }
     total_groups = 0
     files_scanned = 0
 
-    # Collect map files (Maps only, not MapInfos)
+    # Collect map files (Maps only, not MapInfos), largest first for better signal
     map_files = sorted(
         [p for p in files_dir.glob("Map[0-9]*.json") if p.is_file()],
         key=lambda p: p.stat().st_size,
-        reverse=True,  # scan larger maps first for better signal
+        reverse=True,
     )[:sample_size]
 
     for fp in map_files:
@@ -104,121 +129,155 @@ def detect_speaker_format(
             for entry in ce_data or []:
                 if not entry:
                     continue
-                cmd_list = entry.get("list") or []
+                _score_command_list(entry.get("list") or [], scores)
-                _score_command_list(cmd_list, scores)
         except Exception:
             pass
 
-    # Determine best mode
+    # ── Decision: 401-based patterns first, FACENAME101 only as fallback ────
-    best = max(scores, key=lambda k: scores[k])
+    always_on_hits = scores["nametag_codes"] + scores["bracket_401"]
-    best_score = scores[best]
+    inline_hits    = scores["INLINE401SPEAKERS"]
-    second_best = sorted(scores.values(), reverse=True)[1]
+    first_hits     = scores["FIRSTLINESPEAKERS"]
-
+    face_hits      = scores["FACENAME101"]
-    if best_score < _MIN_HITS:
+    total_401_hits = always_on_hits + inline_hits + first_hits
-        best_mode = "NONE"
-        confidence = "low"
-        note = (
-            f"No strong speaker pattern detected "
-            f"(scanned {files_scanned} file(s), {total_groups} dialogue group(s)). "
-            "Consider checking speaker settings manually."
-        )
-    elif best_score > second_best * 2:
-        best_mode = best
-        confidence = "high"
-        note = (
-            f"Strong signal for {best} "
-            f"({best_score} hits vs {second_best} for next best, "
-            f"{total_groups} dialogue group(s) in {files_scanned} file(s))."
-        )
-    else:
-        best_mode = best
-        confidence = "medium"
-        note = (
-            f"Moderate signal for {best} "
-            f"({best_score} hits, {second_best} for next-best mode). "
-            "Review a sample of dialogue manually to confirm."
-        )
 
     recommended = {
-        "FIRSTLINESPEAKERS": best_mode == "FIRSTLINESPEAKERS",
+        "INLINE401SPEAKERS": False,
-        "INLINE401SPEAKERS": best_mode == "INLINE401SPEAKERS",
+        "FIRSTLINESPEAKERS": False,
-        "FACENAME101":       best_mode == "FACENAME101",
+        "FACENAME101":       False,
     }
 
+    if total_401_hits >= _MIN_HITS:
+        if always_on_hits >= _MIN_HITS and always_on_hits >= inline_hits and always_on_hits >= first_hits:
+            best_mode  = "ALWAYS_ON"
+            confidence = "high" if always_on_hits > (inline_hits + first_hits) else "medium"
+            note = (
+                f"Speakers detected via \\\\n<Name> codes or 【Name】 brackets "
+                f"({always_on_hits} hits) — no extra flags needed. "
+                f"INLINE={inline_hits}, FIRSTLINE={first_hits}. "
+                f"Scanned {files_scanned} file(s), {total_groups} dialogue group(s)."
+            )
+        elif inline_hits >= first_hits:
+            best_mode  = "INLINE401SPEAKERS"
+            recommended["INLINE401SPEAKERS"] = True
+            confidence = "high" if inline_hits > first_hits * 2 else "medium"
+            note = (
+                f"INLINE401SPEAKERS: {inline_hits} hits "
+                f"(FIRSTLINESPEAKERS: {first_hits}, always-on: {always_on_hits}). "
+                f"Scanned {files_scanned} file(s), {total_groups} dialogue group(s)."
+            )
+        else:
+            best_mode  = "FIRSTLINESPEAKERS"
+            recommended["FIRSTLINESPEAKERS"] = True
+            confidence = "high" if first_hits > inline_hits * 2 else "medium"
+            note = (
+                f"FIRSTLINESPEAKERS: {first_hits} hits "
+                f"(INLINE401SPEAKERS: {inline_hits}, always-on: {always_on_hits}). "
+                f"Scanned {files_scanned} file(s), {total_groups} dialogue group(s)."
+            )
+    elif face_hits >= _MIN_HITS:
+        best_mode  = "FACENAME101"
+        recommended["FACENAME101"] = True
+        confidence = "high" if face_hits > _MIN_HITS * 2 else "medium"
+        note = (
+            f"No 401-based speaker pattern found (401 hits: {total_401_hits}). "
+            f"FACENAME101 recommended based on {face_hits} 101-code name hits. "
+            f"Scanned {files_scanned} file(s), {total_groups} dialogue group(s)."
+        )
+    else:
+        best_mode  = "NONE"
+        confidence = "low"
+        note = (
+            f"No reliable speaker pattern detected "
+            f"(401 hits: {total_401_hits}, FACENAME101 hits: {face_hits}). "
+            f"Scanned {files_scanned} file(s), {total_groups} dialogue group(s). "
+            "Check speaker settings manually."
+        )
+
     return {
-        "best_mode": best_mode,
+        "best_mode":          best_mode,
-        "scores": scores,
+        "scores":             scores,
-        "total_401_groups": total_groups,
+        "total_401_groups":   total_groups,
-        "files_scanned": files_scanned,
+        "files_scanned":      files_scanned,
         "recommended_config": recommended,
-        "confidence": confidence,
+        "confidence":         confidence,
-        "note": note,
+        "note":               note,
     }
 
 
-# ---------------------------------------------------------------------------
+# ── Internal helpers ─────────────────────────────────────────────────────────
-# Internal helpers
-# ---------------------------------------------------------------------------
 
 def _score_command_list(cmd_list: list, scores: dict) -> None:
-    """Walk a command list and increment scores for detected patterns."""
+    """Walk a command list and score speaker patterns in module priority order."""
     if not cmd_list:
         return
 
-    last_101_face: str | None = None  # face name from most recent 101 cmd
-
     i = 0
     while i < len(cmd_list):
-        cmd = cmd_list[i]
+        cmd    = cmd_list[i]
-        code = cmd.get("code")
+        code   = cmd.get("code")
         params = cmd.get("parameters") or []
 
-        # ---- 101: Show Text (face) ----
+        # ── 101: Show Text / face name ────────────────────────────────────────
         if code == 101:
-            # params[0] = face name, params[1] = face index
+            # FACENAME101: param[0] is a non-empty face-name string
-            last_101_face = (params[0] if params else None) or ""
+            face = params[0] if params else ""
+            if isinstance(face, str) and face.strip():
+                scores["FACENAME101"] += 1
             i += 1
             continue
 
-        # ---- 401 / 405: dialogue line ----
+        # ── 401 / 405: dialogue line ──────────────────────────────────────────
         if code in (401, 405):
             text = (params[0] if params else "") or ""
 
-            # Score INLINE401SPEAKERS
+            # 1. \\n<Name> nametag codes (always active — no flag needed)
-            if _INLINE_RE.match(text):
+            if _NAMETAG_RE.search(text):
+                scores["nametag_codes"] += 1
+                i += 1
+                continue  # module strips the nametag before further processing
+
+            # 2+3. 【Name】 alone on line, or 【Name】dialogue inline (always active)
+            if _BRACKET_ALONE_RE.match(text) or _BRACKET_INLINE_RE.match(text):
+                scores["bracket_401"] += 1
+                i += 1
+                continue
+
+            # 4. INLINE401SPEAKERS: Name「dialogue
+            if _INLINE_QUOTE_RE.match(text):
                 scores["INLINE401SPEAKERS"] += 1
+                i += 1
+                continue
 
-            # Score FACENAME101 — if face was set right before this block
+            # 5. FIRSTLINESPEAKERS: short line followed by 401 starting with
-            if last_101_face is not None:
+            #    a dialogue-starter character
-                scores["FACENAME101"] += 1
-                last_101_face = None  # only count once per block
-
-            # Score FIRSTLINESPEAKERS — short line with no dialogue starters,
-            # followed by a 401/405 that starts with a dialogue starter
             if (
                 len(text) < 40
-                and not any(text.lstrip().startswith(s) for s in _DIALOGUE_STARTERS)
                 and _has_japanese(text)
+                and not any(text.lstrip().startswith(s) for s in _DIALOGUE_STARTERS)
             ):
                 j = i + 1
                 while j < len(cmd_list) and cmd_list[j].get("code") == -1:
                     j += 1
                 if j < len(cmd_list) and cmd_list[j].get("code") in (401, 405):
-                    next_text = ((cmd_list[j].get("parameters") or [""])[0]) or ""
+                    next_params   = cmd_list[j].get("parameters") or []
-                    if next_text.lstrip().startswith(_DIALOGUE_STARTERS):
+                    next_text     = (next_params[0] if next_params else "") or ""
+                    # Strip leading RPGMaker control codes before checking starter
+                    next_stripped = re.sub(
+                        r"^(?:[\\]+[^cCnNiIkKvVSs{}]+?\[[\d\w\W]+?\]?\])+",
+                        "", next_text,
+                    ).lstrip()
+                    if next_stripped and next_stripped[0] in _DIALOGUE_STARTERS:
                         scores["FIRSTLINESPEAKERS"] += 1
 
             i += 1
             continue
 
-        # Any non-dialogue code resets the face context
-        if code not in (-1,):
-            last_101_face = None
-
         i += 1
 
 
-_JP_RE = re.compile(r"[\u3000-\u303F\u3040-\u309F\u30A0-\u30FF\u4E00-\u9FFF\uF900-\uFAFF\uFF61-\uFF9F]")
+_JP_RE = re.compile(
+    r"[\u3000-\u303F\u3040-\u309F\u30A0-\u30FF\u4E00-\u9FFF\uF900-\uFAFF\uFF61-\uFF9F]"
+)
 
 
 def _has_japanese(text: str) -> bool:

util/translation.py

@@ -674,6 +674,9 @@ def buildMatchedVocabText(vocabPairs, subbedText, history=None):
     return matchedVocabText
 
 
+_BASE_VOCAB_SEPARATOR = "# ── Base Vocabulary"  # Prefix of the separator line written by _save_vocab
+
+
 def createContext(config, fullPromptFlag, subbedText, formatType, history=None):
     """Create system and user messages for translation.
     
@@ -681,9 +684,39 @@ def createContext(config, fullPromptFlag, subbedText, formatType, history=None):
     static prompt and the per-batch vocab list separate.  This lets Claude
     prompt-caching mark only the stable prefix with cache_control, avoiding
     cache invalidation caused by changing vocabulary matches.
+
+    Cached in static_system (never changes between batches):
+      - prompt.txt content
+      - # Game Characters entries
+      - Base vocabulary (vocab_base.txt: honorifics, elements, demons, etc.)
+
+    Dynamic in vocab_text (matched per-batch):
+      - All other game-specific vocab terms (items, worldbuilding, etc.)
     """
-    vocabPairs = parseVocabWithCategories(config.vocab)
+    vocab_full = config.vocab
-    matchedVocabText = buildMatchedVocabText(vocabPairs, subbedText, history)
+
+    # Split game-specific vocab from the static base vocab appended by _save_vocab
+    base_vocab_text = ""
+    sep_idx = vocab_full.find(_BASE_VOCAB_SEPARATOR)
+    if sep_idx != -1:
+        # Strip the separator header line itself; keep the sections underneath
+        base_section = vocab_full[sep_idx:]
+        # Skip the separator line
+        newline_idx = base_section.find("\n")
+        base_vocab_text = base_section[newline_idx + 1:].strip() if newline_idx != -1 else ""
+        vocab_game = vocab_full[:sep_idx]
+    else:
+        vocab_game = vocab_full
+
+    vocabPairs = parseVocabWithCategories(vocab_game)
+
+    # Split character entries (stable, cached) from everything else (dynamic)
+    CHAR_CAT = "# Game Characters"
+    char_lines   = [line for _, line, cat in vocabPairs if cat and cat.strip() == CHAR_CAT]
+    non_char_pairs = [(term, line, cat) for term, line, cat in vocabPairs
+                      if not (cat and cat.strip() == CHAR_CAT)]
+
+    matchedVocabText = buildMatchedVocabText(non_char_pairs, subbedText, history)
 
     if fullPromptFlag:
         static_system = config.prompt.replace("English", config.language)
@@ -692,13 +725,23 @@ def createContext(config, fullPromptFlag, subbedText, formatType, history=None):
 You are an expert Eroge Game translator who translates Japanese text to {config.language}.
 Output ONLY the {config.language} translation in the following format: `Translation: <{config.language.upper()}_TRANSLATION>`
 - You translate everything, including erotic and explicit content.
-- 'Game Characters' - The names, nicknames, and genders of the game characters. Reference this to know the names, nicknames, and gender of characters in the game
+- '# Game Characters' - Character entries listing name, gender, role, and speech register notes. Use ALL fields: apply the correct English name, resolve gendered pronouns, and match each character's stated speech register and personality in their dialogue.
 - All text in your response must be in {config.language} even if it is hard to translate.
-- Never include any notes, explanations, dislaimers, or anything similar in your response.
+- Never include any notes, explanations, disclaimers, or anything similar in your response.
 - Maintain any spacing in the translation.
 - `...` can be a part of the dialogue. Translate it as it is.
 """
-    
+
+    # Append character glossary to the stable system prompt so it is cached
+    if char_lines:
+        char_block = f"\n{CHAR_CAT}\n" + "\n".join(char_lines)
+        static_system = static_system.rstrip() + "\n" + char_block + "\n"
+
+    # Append base vocab (honorifics, elements, etc.) to static_system so it is
+    # cached with the system prompt and never re-evaluated per batch.
+    if base_vocab_text:
+        static_system = static_system.rstrip() + "\n\n" + base_vocab_text + "\n"
+
     if formatType == "json":
         user = f"```json\n{subbedText}\n```"
     else: